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.
6796 Commits
32 Branches
421MB
Tree: e3bbb5910f
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e3bbb5910f'
${ noResults }
FFmpeg/libavcodec/h264.c

8622 lines
323KB
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 library is free software; you can redistribute it and/or

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

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

  8.  * version 2 of the License, or (at your option) any later version.

  9.  *

  10.  * This library is distributed in the hope that it will be useful,

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

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

  13.  * Lesser General Public License for more details.

  14.  *

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

  16.  * License along with this library; if not, write to the Free Software

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

  18.  *

  19.  */

  20. 

  21. /**

  22.  * @file h264.c

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

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

  25.  */

  26. 

  27. #include "common.h"

  28. #include "dsputil.h"

  29. #include "avcodec.h"

  30. #include "mpegvideo.h"

  31. #include "h264data.h"

  32. #include "golomb.h"

  33. 

  34. #include "cabac.h"

  35. 

  36. //#undef NDEBUG

  37. #include <assert.h>

  38. 

  39. #define interlaced_dct interlaced_dct_is_a_bad_name

  40. #define mb_intra mb_intra_isnt_initalized_see_mb_type

  41. 

  42. #define LUMA_DC_BLOCK_INDEX   25

  43. #define CHROMA_DC_BLOCK_INDEX 26

  44. 

  45. #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8

  46. #define COEFF_TOKEN_VLC_BITS           8

  47. #define TOTAL_ZEROS_VLC_BITS           9

  48. #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3

  49. #define RUN_VLC_BITS                   3

  50. #define RUN7_VLC_BITS                  6

  51. 

  52. #define MAX_SPS_COUNT 32

  53. #define MAX_PPS_COUNT 256

  54. 

  55. #define MAX_MMCO_COUNT 66

  56. 

  57. /* Compiling in interlaced support reduces the speed

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

  59. #define ALLOW_INTERLACE

  60. 

  61. #ifdef ALLOW_INTERLACE

  62. #define MB_MBAFF h->mb_mbaff

  63. #define MB_FIELD h->mb_field_decoding_flag

  64. #define FRAME_MBAFF h->mb_aff_frame

  65. #else

  66. #define MB_MBAFF 0

  67. #define MB_FIELD 0

  68. #define FRAME_MBAFF 0

  69. #undef  IS_INTERLACED

  70. #define IS_INTERLACED(mb_type) 0

  71. #endif

  72. 

  73. /**

  74.  * Sequence parameter set

  75.  */

  76. typedef struct SPS{

  77. 

  78.     int profile_idc;

  79.     int level_idc;

  80.     int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag

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

  82.     int poc_type;                      ///< pic_order_cnt_type

  83.     int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4

  84.     int delta_pic_order_always_zero_flag;

  85.     int offset_for_non_ref_pic;

  86.     int offset_for_top_to_bottom_field;

  87.     int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle

  88.     int ref_frame_count;               ///< num_ref_frames

  89.     int gaps_in_frame_num_allowed_flag;

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

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

  92.     int frame_mbs_only_flag;

  93.     int mb_aff;                        ///<mb_adaptive_frame_field_flag

  94.     int direct_8x8_inference_flag;

  95.     int crop;                   ///< frame_cropping_flag

  96.     int crop_left;              ///< frame_cropping_rect_left_offset

  97.     int crop_right;             ///< frame_cropping_rect_right_offset

  98.     int crop_top;               ///< frame_cropping_rect_top_offset

  99.     int crop_bottom;            ///< frame_cropping_rect_bottom_offset

  100.     int vui_parameters_present_flag;

  101.     AVRational sar;

  102.     int timing_info_present_flag;

  103.     uint32_t num_units_in_tick;

  104.     uint32_t time_scale;

  105.     int fixed_frame_rate_flag;

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

  107.     int bitstream_restriction_flag;

  108.     int num_reorder_frames;

  109.     int scaling_matrix_present;

  110.     uint8_t scaling_matrix4[6][16];

  111.     uint8_t scaling_matrix8[2][64];

  112. }SPS;

  113. 

  114. /**

  115.  * Picture parameter set

  116.  */

  117. typedef struct PPS{

  118.     int sps_id;

  119.     int cabac;                  ///< entropy_coding_mode_flag

  120.     int pic_order_present;      ///< pic_order_present_flag

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

  122.     int mb_slice_group_map_type;

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

  124.     int weighted_pred;          ///< weighted_pred_flag

  125.     int weighted_bipred_idc;

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

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

  128.     int chroma_qp_index_offset;

  129.     int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag

  130.     int constrained_intra_pred; ///< constrained_intra_pred_flag

  131.     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag

  132.     int transform_8x8_mode;     ///< transform_8x8_mode_flag

  133.     uint8_t scaling_matrix4[6][16];

  134.     uint8_t scaling_matrix8[2][64];

  135. }PPS;

  136. 

  137. /**

  138.  * Memory management control operation opcode.

  139.  */

  140. typedef enum MMCOOpcode{

  141.     MMCO_END=0,

  142.     MMCO_SHORT2UNUSED,

  143.     MMCO_LONG2UNUSED,

  144.     MMCO_SHORT2LONG,

  145.     MMCO_SET_MAX_LONG,

  146.     MMCO_RESET,

  147.     MMCO_LONG,

  148. } MMCOOpcode;

  149. 

  150. /**

  151.  * Memory management control operation.

  152.  */

  153. typedef struct MMCO{

  154.     MMCOOpcode opcode;

  155.     int short_frame_num;

  156.     int long_index;

  157. } MMCO;

  158. 

  159. /**

  160.  * H264Context

  161.  */

  162. typedef struct H264Context{

  163.     MpegEncContext s;

  164.     int nal_ref_idc;

  165.     int nal_unit_type;

  166. #define NAL_SLICE                1

  167. #define NAL_DPA                  2

  168. #define NAL_DPB                  3

  169. #define NAL_DPC                  4

  170. #define NAL_IDR_SLICE            5

  171. #define NAL_SEI                  6

  172. #define NAL_SPS                  7

  173. #define NAL_PPS                  8

  174. #define NAL_AUD                  9

  175. #define NAL_END_SEQUENCE        10

  176. #define NAL_END_STREAM          11

  177. #define NAL_FILLER_DATA         12

  178. #define NAL_SPS_EXT             13

  179. #define NAL_AUXILIARY_SLICE     19

  180.     uint8_t *rbsp_buffer;

  181.     unsigned int rbsp_buffer_size;

  182. 

  183.     /**

  184.       * Used to parse AVC variant of h264

  185.       */

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

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

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

  189. 

  190.     int chroma_qp; //QPc

  191. 

  192.     int prev_mb_skipped;

  193.     int next_mb_skipped;

  194. 

  195.     //prediction stuff

  196.     int chroma_pred_mode;

  197.     int intra16x16_pred_mode;

  198. 

  199.     int top_mb_xy;

  200.     int left_mb_xy[2];

  201. 

  202.     int8_t intra4x4_pred_mode_cache[5*8];

  203.     int8_t (*intra4x4_pred_mode)[8];

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

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

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

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

  208.     unsigned int topleft_samples_available;

  209.     unsigned int top_samples_available;

  210.     unsigned int topright_samples_available;

  211.     unsigned int left_samples_available;

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

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

  214. 

  215.     /**

  216.      * non zero coeff count cache.

  217.      * is 64 if not available.

  218.      */

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

  220.     uint8_t (*non_zero_count)[16];

  221. 

  222.     /**

  223.      * Motion vector cache.

  224.      */

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

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

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

  228. #define PART_NOT_AVAILABLE -2

  229. 

  230.     /**

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

  232.      */

  233.     int mv_cache_clean[2];

  234. 

  235.     /**

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

  237.      */

  238.     int neighbor_transform_size;

  239. 

  240.     /**

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

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

  243.      */

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

  245. 

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

  247.     uint32_t *mb2b8_xy;

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

  249.     int b8_stride;

  250. 

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

  252.     int mb_uvlinesize;

  253. 

  254.     int emu_edge_width;

  255.     int emu_edge_height;

  256. 

  257.     int halfpel_flag;

  258.     int thirdpel_flag;

  259. 

  260.     int unknown_svq3_flag;

  261.     int next_slice_index;

  262. 

  263.     SPS sps_buffer[MAX_SPS_COUNT];

  264.     SPS sps; ///< current sps

  265. 

  266.     PPS pps_buffer[MAX_PPS_COUNT];

  267.     /**

  268.      * current pps

  269.      */

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

  271. 

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

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

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

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

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

  277. 

  278.     int slice_num;

  279.     uint8_t *slice_table_base;

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

  281.     int slice_type;

  282.     int slice_type_fixed;

  283. 

  284.     //interlacing specific flags

  285.     int mb_aff_frame;

  286.     int mb_field_decoding_flag;

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

  288. 

  289.     int sub_mb_type[4];

  290. 

  291.     //POC stuff

  292.     int poc_lsb;

  293.     int poc_msb;

  294.     int delta_poc_bottom;

  295.     int delta_poc[2];

  296.     int frame_num;

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

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

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

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

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

  302. 

  303.     /**

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

  305.      */

  306.     int curr_pic_num;

  307. 

  308.     /**

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

  310.      */

  311.     int max_pic_num;

  312. 

  313.     //Weighted pred stuff

  314.     int use_weight;

  315.     int use_weight_chroma;

  316.     int luma_log2_weight_denom;

  317.     int chroma_log2_weight_denom;

  318.     int luma_weight[2][48];

  319.     int luma_offset[2][48];

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

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

  322.     int implicit_weight[48][48];

  323. 

  324.     //deblock

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

  326.     int slice_alpha_c0_offset;

  327.     int slice_beta_offset;

  328. 

  329.     int redundant_pic_count;

  330. 

  331.     int direct_spatial_mv_pred;

  332.     int dist_scale_factor[16];

  333.     int dist_scale_factor_field[32];

  334.     int map_col_to_list0[2][16];

  335.     int map_col_to_list0_field[2][32];

  336. 

  337.     /**

  338.      * num_ref_idx_l0/1_active_minus1 + 1

  339.      */

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

  341.     Picture *short_ref[32];

  342.     Picture *long_ref[32];

  343.     Picture default_ref_list[2][32];

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

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

  346.     Picture *delayed_output_pic;

  347. 

  348.     /**

  349.      * memory management control operations buffer.

  350.      */

  351.     MMCO mmco[MAX_MMCO_COUNT];

  352.     int mmco_index;

  353. 

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

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

  356. 

  357.     //data partitioning

  358.     GetBitContext intra_gb;

  359.     GetBitContext inter_gb;

  360.     GetBitContext *intra_gb_ptr;

  361.     GetBitContext *inter_gb_ptr;

  362. 

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

  364. 

  365.     /**

  366.      * Cabac

  367.      */

  368.     CABACContext cabac;

  369.     uint8_t      cabac_state[460];

  370.     int          cabac_init_idc;

  371. 

  372.     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */

  373.     uint16_t     *cbp_table;

  374.     int cbp;

  375.     int top_cbp;

  376.     int left_cbp;

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

  378.     uint8_t     *chroma_pred_mode_table;

  379.     int         last_qscale_diff;

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

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

  382.     uint8_t     *direct_table;

  383.     uint8_t     direct_cache[5*8];

  384. 

  385.     uint8_t zigzag_scan[16];

  386.     uint8_t zigzag_scan8x8[64];

  387.     uint8_t zigzag_scan8x8_cavlc[64];

  388.     uint8_t field_scan[16];

  389.     uint8_t field_scan8x8[64];

  390.     uint8_t field_scan8x8_cavlc[64];

  391.     const uint8_t *zigzag_scan_q0;

  392.     const uint8_t *zigzag_scan8x8_q0;

  393.     const uint8_t *zigzag_scan8x8_cavlc_q0;

  394.     const uint8_t *field_scan_q0;

  395.     const uint8_t *field_scan8x8_q0;

  396.     const uint8_t *field_scan8x8_cavlc_q0;

  397. 

  398.     int x264_build;

  399. }H264Context;

  400. 

  401. static VLC coeff_token_vlc[4];

  402. static VLC chroma_dc_coeff_token_vlc;

  403. 

  404. static VLC total_zeros_vlc[15];

  405. static VLC chroma_dc_total_zeros_vlc[3];

  406. 

  407. static VLC run_vlc[6];

  408. static VLC run7_vlc;

  409. 

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

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

  412. 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);

  413. 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);

  414. 

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

  416. #ifdef WORDS_BIGENDIAN

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

  418. #else

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

  420. #endif

  421. }

  422. 

  423. /**

  424.  * fill a rectangle.

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

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

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

  428.  */

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

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

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

  432.     assert(w<=4);

  433. 

  434.     w      *= size;

  435.     stride *= size;

  436. 

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

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

  439.     if(w==2){

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

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

  442.         if(h==1) return;

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

  444.         if(h==2) return;

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

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

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

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

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

  450.         if(h==1) return;

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

  452.         if(h==2) return;

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

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

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

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

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

  458.         const uint64_t v= val*0x0100000001ULL;

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

  460.         if(h==1) return;

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

  462.         if(h==2) return;

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

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

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

  466.         const uint64_t v= val*0x0100000001ULL;

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

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

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

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

  471.         if(h==2) return;

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

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

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

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

  476. #else

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

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

  479.         if(h==1) return;

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

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

  482.         if(h==2) return;

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

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

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

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

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

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

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

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

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

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

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

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

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

  496.         if(h==2) return;

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

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

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

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

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

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

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

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

  505. #endif

  506.     }else

  507.         assert(0);

  508.     assert(h==4);

  509. }

  510. 

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

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

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

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

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

  516.     int left_block[8];

  517.     int i;

  518. 

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

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

  521.         return;

  522. 

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

  524. 

  525.     top_xy     = mb_xy  - s->mb_stride;

  526.     topleft_xy = top_xy - 1;

  527.     topright_xy= top_xy + 1;

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

  529.     left_block[0]= 0;

  530.     left_block[1]= 1;

  531.     left_block[2]= 2;

  532.     left_block[3]= 3;

  533.     left_block[4]= 7;

  534.     left_block[5]= 10;

  535.     left_block[6]= 8;

  536.     left_block[7]= 11;

  537.     if(FRAME_MBAFF){

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

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

  540.         const int topleft_pair_xy  = top_pair_xy - 1;

  541.         const int topright_pair_xy = top_pair_xy + 1;

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

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

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

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

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

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

  548.         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);

  549.         if (bottom

  550.                 ? !curr_mb_frame_flag // bottom macroblock

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

  552.                 ) {

  553.             top_xy -= s->mb_stride;

  554.         }

  555.         if (bottom

  556.                 ? !curr_mb_frame_flag // bottom macroblock

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

  558.                 ) {

  559.             topleft_xy -= s->mb_stride;

  560.         }

  561.         if (bottom

  562.                 ? !curr_mb_frame_flag // bottom macroblock

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

  564.                 ) {

  565.             topright_xy -= s->mb_stride;

  566.         }

  567.         if (left_mb_frame_flag != curr_mb_frame_flag) {

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

  569.             if (curr_mb_frame_flag) {

  570.                 if (bottom) {

  571.                     left_block[0]= 2;

  572.                     left_block[1]= 2;

  573.                     left_block[2]= 3;

  574.                     left_block[3]= 3;

  575.                     left_block[4]= 8;

  576.                     left_block[5]= 11;

  577.                     left_block[6]= 8;

  578.                     left_block[7]= 11;

  579.                 } else {

  580.                     left_block[0]= 0;

  581.                     left_block[1]= 0;

  582.                     left_block[2]= 1;

  583.                     left_block[3]= 1;

  584.                     left_block[4]= 7;

  585.                     left_block[5]= 10;

  586.                     left_block[6]= 7;

  587.                     left_block[7]= 10;

  588.                 }

  589.             } else {

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

  591.                 //left_block[0]= 0;

  592.                 left_block[1]= 2;

  593.                 left_block[2]= 0;

  594.                 left_block[3]= 2;

  595.                 //left_block[4]= 7;

  596.                 left_block[5]= 10;

  597.                 left_block[6]= 7;

  598.                 left_block[7]= 10;

  599.             }

  600.         }

  601.     }

  602. 

  603.     h->top_mb_xy = top_xy;

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

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

  606.     if(for_deblock){

  607.         topleft_type = 0;

  608.         topright_type = 0;

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

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

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

  612. 

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

  614.             int list;

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

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

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

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

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

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

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

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

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

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

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

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

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

  628.                     }

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

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

  631.                     ref += h->b8_stride;

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

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

  634.                 }else{

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

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

  637.                 }

  638.             }

  639.         }

  640.     }else{

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

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

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

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

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

  646.     }

  647. 

  648.     if(IS_INTRA(mb_type)){

  649.         h->topleft_samples_available=

  650.         h->top_samples_available=

  651.         h->left_samples_available= 0xFFFF;

  652.         h->topright_samples_available= 0xEEEA;

  653. 

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

  655.             h->topleft_samples_available= 0xB3FF;

  656.             h->top_samples_available= 0x33FF;

  657.             h->topright_samples_available= 0x26EA;

  658.         }

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

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

  661.                 h->topleft_samples_available&= 0xDF5F;

  662.                 h->left_samples_available&= 0x5F5F;

  663.             }

  664.         }

  665. 

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

  667.             h->topleft_samples_available&= 0x7FFF;

  668. 

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

  670.             h->topright_samples_available&= 0xFBFF;

  671. 

  672.         if(IS_INTRA4x4(mb_type)){

  673.             if(IS_INTRA4x4(top_type)){

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

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

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

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

  678.             }else{

  679.                 int pred;

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

  681.                     pred= -1;

  682.                 else{

  683.                     pred= 2;

  684.                 }

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

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

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

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

  689.             }

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

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

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

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

  694.                 }else{

  695.                     int pred;

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

  697.                         pred= -1;

  698.                     else{

  699.                         pred= 2;

  700.                     }

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

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

  703.                 }

  704.             }

  705.         }

  706.     }

  707. 

  708. 

  709. /*

  710. 0 . T T. T T T T

  711. 1 L . .L . . . .

  712. 2 L . .L . . . .

  713. 3 . T TL . . . .

  714. 4 L . .L . . . .

  715. 5 L . .. . . . .

  716. */

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

  718.     if(top_type){

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

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

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

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

  723. 

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

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

  726. 

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

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

  729. 

  730.     }else{

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

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

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

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

  735. 

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

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

  738. 

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

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

  741. 

  742.     }

  743. 

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

  745.         if(left_type[i]){

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

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

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

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

  750.         }else{

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

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

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

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

  755.         }

  756.     }

  757. 

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

  759.         // top_cbp

  760.         if(top_type) {

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

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

  763.             h->top_cbp = 0x1C0;

  764.         } else {

  765.             h->top_cbp = 0;

  766.         }

  767.         // left_cbp

  768.         if (left_type[0]) {

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

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

  771.             h->left_cbp = 0x1C0;

  772.         } else {

  773.             h->left_cbp = 0;

  774.         }

  775.         if (left_type[0]) {

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

  777.         }

  778.         if (left_type[1]) {

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

  780.         }

  781.     }

  782. 

  783. #if 1

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

  785.         int list;

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

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

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

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

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

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

  792.                 }*/

  793.                 continue;

  794.             }

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

  796. 

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

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

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

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

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

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

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

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

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

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

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

  808.             }else{

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

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

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

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

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

  814.             }

  815. 

  816.             //FIXME unify cleanup or sth

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

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

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

  820.                 *(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]];

  821.                 *(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]];

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

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

  824.             }else{

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

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

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

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

  829.             }

  830. 

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

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

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

  834.                 *(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]];

  835.                 *(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]];

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

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

  838.             }else{

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

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

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

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

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

  844.             }

  845. 

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

  847.                 continue;

  848. 

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

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

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

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

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

  854.             }else{

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

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

  857.             }

  858. 

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

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

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

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

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

  864.             }else{

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

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

  867.             }

  868. 

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

  870.                 continue;

  871. 

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

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

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

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

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

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

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

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

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

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

  882. 

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

  884.                 /* XXX beurk, Load mvd */

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

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

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

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

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

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

  891.                 }else{

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

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

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

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

  896.                 }

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

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

  899.                     *(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]];

  900.                     *(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]];

  901.                 }else{

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

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

  904.                 }

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

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

  907.                     *(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]];

  908.                     *(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]];

  909.                 }else{

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

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

  912.                 }

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

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

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

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

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

  918. 

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

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

  921. 

  922.                     if(IS_DIRECT(top_type)){

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

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

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

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

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

  928.                     }else{

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

  930.                     }

  931. 

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

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

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

  935.                         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)];

  936.                     else

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

  938. 

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

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

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

  942.                         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)];

  943.                     else

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

  945.                 }

  946.             }

  947. 

  948.             if(FRAME_MBAFF){

  949. #define MAP_MVS\

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

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

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

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

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

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

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

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

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

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

  960.                 if(MB_FIELD){

  961. #define MAP_F2F(idx, mb_type)\

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

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

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

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

  966.                     }

  967.                     MAP_MVS

  968. #undef MAP_F2F

  969.                 }else{

  970. #define MAP_F2F(idx, mb_type)\

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

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

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

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

  975.                     }

  976.                     MAP_MVS

  977. #undef MAP_F2F

  978.                 }

  979.             }

  980.         }

  981.     }

  982. #endif

  983. 

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

  985. }

  986. 

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

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

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

  990. 

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

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

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

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

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

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

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

  998. }

  999. 

  1000. /**

  1001.  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.

  1002.  */

  1003. static inline int check_intra4x4_pred_mode(H264Context *h){

  1004.     MpegEncContext * const s = &h->s;

  1005.     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};

  1006.     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};

  1007.     int i;

  1008. 

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

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

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

  1012.             if(status<0){

  1013.                 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);

  1014.                 return -1;

  1015.             } else if(status){

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

  1017.             }

  1018.         }

  1019.     }

  1020. 

  1021.     if(!(h->left_samples_available&0x8000)){

  1022.         for(i=0; i<4; i++){

  1023.             int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];

  1024.             if(status<0){

  1025.                 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);

  1026.                 return -1;

  1027.             } else if(status){

  1028.                 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;

  1029.             }

  1030.         }

  1031.     }

  1032. 

  1033.     return 0;

  1034. } //FIXME cleanup like next

  1035. 

  1036. /**

  1037.  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.

  1038.  */

  1039. static inline int check_intra_pred_mode(H264Context *h, int mode){

  1040.     MpegEncContext * const s = &h->s;

  1041.     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};

  1042.     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};

  1043. 

  1044.     if(mode < 0 || mode > 6) {

  1045.         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);

  1046.         return -1;

  1047.     }

  1048. 

  1049.     if(!(h->top_samples_available&0x8000)){

  1050.         mode= top[ mode ];

  1051.         if(mode<0){

  1052.             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);

  1053.             return -1;

  1054.         }

  1055.     }

  1056. 

  1057.     if(!(h->left_samples_available&0x8000)){

  1058.         mode= left[ mode ];

  1059.         if(mode<0){

  1060.             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);

  1061.             return -1;

  1062.         }

  1063.     }

  1064. 

  1065.     return mode;

  1066. }

  1067. 

  1068. /**

  1069.  * gets the predicted intra4x4 prediction mode.

  1070.  */

  1071. static inline int pred_intra_mode(H264Context *h, int n){

  1072.     const int index8= scan8[n];

  1073.     const int left= h->intra4x4_pred_mode_cache[index8 - 1];

  1074.     const int top = h->intra4x4_pred_mode_cache[index8 - 8];

  1075.     const int min= FFMIN(left, top);

  1076. 

  1077.     tprintf("mode:%d %d min:%d\n", left ,top, min);

  1078. 

  1079.     if(min<0) return DC_PRED;

  1080.     else      return min;

  1081. }

  1082. 

  1083. static inline void write_back_non_zero_count(H264Context *h){

  1084.     MpegEncContext * const s = &h->s;

  1085.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  1086. 

  1087.     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];

  1088.     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];

  1089.     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];

  1090.     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];

  1091.     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];

  1092.     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];

  1093.     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];

  1094. 

  1095.     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];

  1096.     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];

  1097.     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];

  1098. 

  1099.     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];

  1100.     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];

  1101.     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];

  1102. 

  1103.     if(FRAME_MBAFF){

  1104.         // store all luma nnzs, for deblocking

  1105.         int v = 0, i;

  1106.         for(i=0; i<16; i++)

  1107.             v += (!!h->non_zero_count_cache[scan8[i]]) << i;

  1108.         *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;

  1109.     }

  1110. }

  1111. 

  1112. /**

  1113.  * gets the predicted number of non zero coefficients.

  1114.  * @param n block index

  1115.  */

  1116. static inline int pred_non_zero_count(H264Context *h, int n){

  1117.     const int index8= scan8[n];

  1118.     const int left= h->non_zero_count_cache[index8 - 1];

  1119.     const int top = h->non_zero_count_cache[index8 - 8];

  1120.     int i= left + top;

  1121. 

  1122.     if(i<64) i= (i+1)>>1;

  1123. 

  1124.     tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);

  1125. 

  1126.     return i&31;

  1127. }

  1128. 

  1129. static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){

  1130.     const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];

  1131. 

  1132.     /* there is no consistent mapping of mvs to neighboring locations that will

  1133.      * make mbaff happy, so we can't move all this logic to fill_caches */

  1134.     if(FRAME_MBAFF){

  1135.         MpegEncContext *s = &h->s;

  1136.         const uint32_t *mb_types = s->current_picture_ptr->mb_type;

  1137.         const int16_t *mv;

  1138.         *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;

  1139.         *C = h->mv_cache[list][scan8[0]-2];

  1140. 

  1141.         if(!MB_FIELD

  1142.            && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){

  1143.             int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);

  1144.             if(IS_INTERLACED(mb_types[topright_xy])){

  1145. #define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\

  1146.                 const int x4 = X4, y4 = Y4;\

  1147.                 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\

  1148.                 if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\

  1149.                     return LIST_NOT_USED;\

  1150.                 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\

  1151.                 h->mv_cache[list][scan8[0]-2][0] = mv[0];\

  1152.                 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\

  1153.                 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;

  1154. 

  1155.                 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);

  1156.             }

  1157.         }

  1158.         if(topright_ref == PART_NOT_AVAILABLE

  1159.            && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4

  1160.            && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){

  1161.             if(!MB_FIELD

  1162.                && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){

  1163.                 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);

  1164.             }

  1165.             if(MB_FIELD

  1166.                && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])

  1167.                && i >= scan8[0]+8){

  1168.                 // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.

  1169.                 SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);

  1170.             }

  1171.         }

  1172. #undef SET_DIAG_MV

  1173.     }

  1174. 

  1175.     if(topright_ref != PART_NOT_AVAILABLE){

  1176.         *C= h->mv_cache[list][ i - 8 + part_width ];

  1177.         return topright_ref;

  1178.     }else{

  1179.         tprintf("topright MV not available\n");

  1180. 

  1181.         *C= h->mv_cache[list][ i - 8 - 1 ];

  1182.         return h->ref_cache[list][ i - 8 - 1 ];

  1183.     }

  1184. }

  1185. 

  1186. /**

  1187.  * gets the predicted MV.

  1188.  * @param n the block index

  1189.  * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)

  1190.  * @param mx the x component of the predicted motion vector

  1191.  * @param my the y component of the predicted motion vector

  1192.  */

  1193. static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){

  1194.     const int index8= scan8[n];

  1195.     const int top_ref=      h->ref_cache[list][ index8 - 8 ];

  1196.     const int left_ref=     h->ref_cache[list][ index8 - 1 ];

  1197.     const int16_t * const A= h->mv_cache[list][ index8 - 1 ];

  1198.     const int16_t * const B= h->mv_cache[list][ index8 - 8 ];

  1199.     const int16_t * C;

  1200.     int diagonal_ref, match_count;

  1201. 

  1202.     assert(part_width==1 || part_width==2 || part_width==4);

  1203. 

  1204. /* mv_cache

  1205.   B . . A T T T T

  1206.   U . . L . . , .

  1207.   U . . L . . . .

  1208.   U . . L . . , .

  1209.   . . . L . . . .

  1210. */

  1211. 

  1212.     diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);

  1213.     match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);

  1214.     tprintf("pred_motion match_count=%d\n", match_count);

  1215.     if(match_count > 1){ //most common

  1216.         *mx= mid_pred(A[0], B[0], C[0]);

  1217.         *my= mid_pred(A[1], B[1], C[1]);

  1218.     }else if(match_count==1){

  1219.         if(left_ref==ref){

  1220.             *mx= A[0];

  1221.             *my= A[1];

  1222.         }else if(top_ref==ref){

  1223.             *mx= B[0];

  1224.             *my= B[1];

  1225.         }else{

  1226.             *mx= C[0];

  1227.             *my= C[1];

  1228.         }

  1229.     }else{

  1230.         if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){

  1231.             *mx= A[0];

  1232.             *my= A[1];

  1233.         }else{

  1234.             *mx= mid_pred(A[0], B[0], C[0]);

  1235.             *my= mid_pred(A[1], B[1], C[1]);

  1236.         }

  1237.     }

  1238. 

  1239.     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);

  1240. }

  1241. 

  1242. /**

  1243.  * gets the directionally predicted 16x8 MV.

  1244.  * @param n the block index

  1245.  * @param mx the x component of the predicted motion vector

  1246.  * @param my the y component of the predicted motion vector

  1247.  */

  1248. static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){

  1249.     if(n==0){

  1250.         const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];

  1251.         const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];

  1252. 

  1253.         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);

  1254. 

  1255.         if(top_ref == ref){

  1256.             *mx= B[0];

  1257.             *my= B[1];

  1258.             return;

  1259.         }

  1260.     }else{

  1261.         const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];

  1262.         const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];

  1263. 

  1264.         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);

  1265. 

  1266.         if(left_ref == ref){

  1267.             *mx= A[0];

  1268.             *my= A[1];

  1269.             return;

  1270.         }

  1271.     }

  1272. 

  1273.     //RARE

  1274.     pred_motion(h, n, 4, list, ref, mx, my);

  1275. }

  1276. 

  1277. /**

  1278.  * gets the directionally predicted 8x16 MV.

  1279.  * @param n the block index

  1280.  * @param mx the x component of the predicted motion vector

  1281.  * @param my the y component of the predicted motion vector

  1282.  */

  1283. static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){

  1284.     if(n==0){

  1285.         const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];

  1286.         const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];

  1287. 

  1288.         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);

  1289. 

  1290.         if(left_ref == ref){

  1291.             *mx= A[0];

  1292.             *my= A[1];

  1293.             return;

  1294.         }

  1295.     }else{

  1296.         const int16_t * C;

  1297.         int diagonal_ref;

  1298. 

  1299.         diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);

  1300. 

  1301.         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);

  1302. 

  1303.         if(diagonal_ref == ref){

  1304.             *mx= C[0];

  1305.             *my= C[1];

  1306.             return;

  1307.         }

  1308.     }

  1309. 

  1310.     //RARE

  1311.     pred_motion(h, n, 2, list, ref, mx, my);

  1312. }

  1313. 

  1314. static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){

  1315.     const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];

  1316.     const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];

  1317. 

  1318.     tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);

  1319. 

  1320.     if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE

  1321.        || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)

  1322.        || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){

  1323. 

  1324.         *mx = *my = 0;

  1325.         return;

  1326.     }

  1327. 

  1328.     pred_motion(h, 0, 4, 0, 0, mx, my);

  1329. 

  1330.     return;

  1331. }

  1332. 

  1333. static inline void direct_dist_scale_factor(H264Context * const h){

  1334.     const int poc = h->s.current_picture_ptr->poc;

  1335.     const int poc1 = h->ref_list[1][0].poc;

  1336.     int i;

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

  1338.         int poc0 = h->ref_list[0][i].poc;

  1339.         int td = clip(poc1 - poc0, -128, 127);

  1340.         if(td == 0 /* FIXME || pic0 is a long-term ref */){

  1341.             h->dist_scale_factor[i] = 256;

  1342.         }else{

  1343.             int tb = clip(poc - poc0, -128, 127);

  1344.             int tx = (16384 + (ABS(td) >> 1)) / td;

  1345.             h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);

  1346.         }

  1347.     }

  1348.     if(FRAME_MBAFF){

  1349.         for(i=0; i<h->ref_count[0]; i++){

  1350.             h->dist_scale_factor_field[2*i] =

  1351.             h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];

  1352.         }

  1353.     }

  1354. }

  1355. static inline void direct_ref_list_init(H264Context * const h){

  1356.     MpegEncContext * const s = &h->s;

  1357.     Picture * const ref1 = &h->ref_list[1][0];

  1358.     Picture * const cur = s->current_picture_ptr;

  1359.     int list, i, j;

  1360.     if(cur->pict_type == I_TYPE)

  1361.         cur->ref_count[0] = 0;

  1362.     if(cur->pict_type != B_TYPE)

  1363.         cur->ref_count[1] = 0;

  1364.     for(list=0; list<2; list++){

  1365.         cur->ref_count[list] = h->ref_count[list];

  1366.         for(j=0; j<h->ref_count[list]; j++)

  1367.             cur->ref_poc[list][j] = h->ref_list[list][j].poc;

  1368.     }

  1369.     if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)

  1370.         return;

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

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

  1373.             const int poc = ref1->ref_poc[list][i];

  1374.             h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */

  1375.             for(j=0; j<h->ref_count[list]; j++)

  1376.                 if(h->ref_list[list][j].poc == poc){

  1377.                     h->map_col_to_list0[list][i] = j;

  1378.                     break;

  1379.                 }

  1380.         }

  1381.     }

  1382.     if(FRAME_MBAFF){

  1383.         for(list=0; list<2; list++){

  1384.             for(i=0; i<ref1->ref_count[list]; i++){

  1385.                 j = h->map_col_to_list0[list][i];

  1386.                 h->map_col_to_list0_field[list][2*i] = 2*j;

  1387.                 h->map_col_to_list0_field[list][2*i+1] = 2*j+1;

  1388.             }

  1389.         }

  1390.     }

  1391. }

  1392. 

  1393. static inline void pred_direct_motion(H264Context * const h, int *mb_type){

  1394.     MpegEncContext * const s = &h->s;

  1395.     const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;

  1396.     const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;

  1397.     const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;

  1398.     const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];

  1399.     const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];

  1400.     const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];

  1401.     const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];

  1402.     const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];

  1403.     const int is_b8x8 = IS_8X8(*mb_type);

  1404.     int sub_mb_type;

  1405.     int i8, i4;

  1406. 

  1407. #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)

  1408.     if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){

  1409.         /* FIXME save sub mb types from previous frames (or derive from MVs)

  1410.          * so we know exactly what block size to use */

  1411.         sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */

  1412.         *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;

  1413.     }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){

  1414.         sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  1415.         *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */

  1416.     }else{

  1417.         sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  1418.         *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;

  1419.     }

  1420.     if(!is_b8x8)

  1421.         *mb_type |= MB_TYPE_DIRECT2;

  1422.     if(MB_FIELD)

  1423.         *mb_type |= MB_TYPE_INTERLACED;

  1424. 

  1425.     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);

  1426. 

  1427.     if(h->direct_spatial_mv_pred){

  1428.         int ref[2];

  1429.         int mv[2][2];

  1430.         int list;

  1431. 

  1432.         /* FIXME interlacing + spatial direct uses wrong colocated block positions */

  1433. 

  1434.         /* ref = min(neighbors) */

  1435.         for(list=0; list<2; list++){

  1436.             int refa = h->ref_cache[list][scan8[0] - 1];

  1437.             int refb = h->ref_cache[list][scan8[0] - 8];

  1438.             int refc = h->ref_cache[list][scan8[0] - 8 + 4];

  1439.             if(refc == -2)

  1440.                 refc = h->ref_cache[list][scan8[0] - 8 - 1];

  1441.             ref[list] = refa;

  1442.             if(ref[list] < 0 || (refb < ref[list] && refb >= 0))

  1443.                 ref[list] = refb;

  1444.             if(ref[list] < 0 || (refc < ref[list] && refc >= 0))

  1445.                 ref[list] = refc;

  1446.             if(ref[list] < 0)

  1447.                 ref[list] = -1;

  1448.         }

  1449. 

  1450.         if(ref[0] < 0 && ref[1] < 0){

  1451.             ref[0] = ref[1] = 0;

  1452.             mv[0][0] = mv[0][1] =

  1453.             mv[1][0] = mv[1][1] = 0;

  1454.         }else{

  1455.             for(list=0; list<2; list++){

  1456.                 if(ref[list] >= 0)

  1457.                     pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);

  1458.                 else

  1459.                     mv[list][0] = mv[list][1] = 0;

  1460.             }

  1461.         }

  1462. 

  1463.         if(ref[1] < 0){

  1464.             *mb_type &= ~MB_TYPE_P0L1;

  1465.             sub_mb_type &= ~MB_TYPE_P0L1;

  1466.         }else if(ref[0] < 0){

  1467.             *mb_type &= ~MB_TYPE_P0L0;

  1468.             sub_mb_type &= ~MB_TYPE_P0L0;

  1469.         }

  1470. 

  1471.         if(IS_16X16(*mb_type)){

  1472.             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);

  1473.             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);

  1474.             if(!IS_INTRA(mb_type_col)

  1475.                && (   (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)

  1476.                    || (l1ref0[0]  < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1

  1477.                        && (h->x264_build>33 || !h->x264_build)))){

  1478.                 if(ref[0] > 0)

  1479.                     fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);

  1480.                 else

  1481.                     fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);

  1482.                 if(ref[1] > 0)

  1483.                     fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);

  1484.                 else

  1485.                     fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);

  1486.             }else{

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

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

  1489.             }

  1490.         }else{

  1491.             for(i8=0; i8<4; i8++){

  1492.                 const int x8 = i8&1;

  1493.                 const int y8 = i8>>1;

  1494. 

  1495.                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  1496.                     continue;

  1497.                 h->sub_mb_type[i8] = sub_mb_type;

  1498. 

  1499.                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);

  1500.                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);

  1501.                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);

  1502.                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);

  1503. 

  1504.                 /* col_zero_flag */

  1505.                 if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0

  1506.                                               || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0

  1507.                                                   && (h->x264_build>33 || !h->x264_build)))){

  1508.                     const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;

  1509.                     if(IS_SUB_8X8(sub_mb_type)){

  1510.                         const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];

  1511.                         if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){

  1512.                             if(ref[0] == 0)

  1513.                                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

  1514.                             if(ref[1] == 0)

  1515.                                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);

  1516.                         }

  1517.                     }else

  1518.                     for(i4=0; i4<4; i4++){

  1519.                         const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];

  1520.                         if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){

  1521.                             if(ref[0] == 0)

  1522.                                 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;

  1523.                             if(ref[1] == 0)

  1524.                                 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;

  1525.                         }

  1526.                     }

  1527.                 }

  1528.             }

  1529.         }

  1530.     }else{ /* direct temporal mv pred */

  1531.         const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};

  1532.         const int *dist_scale_factor = h->dist_scale_factor;

  1533. 

  1534.         if(FRAME_MBAFF){

  1535.             if(IS_INTERLACED(*mb_type)){

  1536.                 map_col_to_list0[0] = h->map_col_to_list0_field[0];

  1537.                 map_col_to_list0[1] = h->map_col_to_list0_field[1];

  1538.                 dist_scale_factor = h->dist_scale_factor_field;

  1539.             }

  1540.             if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){

  1541.                 /* FIXME assumes direct_8x8_inference == 1 */

  1542.                 const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;

  1543.                 int mb_types_col[2];

  1544.                 int y_shift;

  1545. 

  1546.                 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1

  1547.                          | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)

  1548.                          | (*mb_type & MB_TYPE_INTERLACED);

  1549.                 sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;

  1550. 

  1551.                 if(IS_INTERLACED(*mb_type)){

  1552.                     /* frame to field scaling */

  1553.                     mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];

  1554.                     mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];

  1555.                     if(s->mb_y&1){

  1556.                         l1ref0 -= 2*h->b8_stride;

  1557.                         l1ref1 -= 2*h->b8_stride;

  1558.                         l1mv0 -= 4*h->b_stride;

  1559.                         l1mv1 -= 4*h->b_stride;

  1560.                     }

  1561.                     y_shift = 0;

  1562. 

  1563.                     if(   (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)

  1564.                        && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)

  1565.                        && !is_b8x8)

  1566.                         *mb_type |= MB_TYPE_16x8;

  1567.                     else

  1568.                         *mb_type |= MB_TYPE_8x8;

  1569.                 }else{

  1570.                     /* field to frame scaling */

  1571.                     /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)

  1572.                      * but in MBAFF, top and bottom POC are equal */

  1573.                     int dy = (s->mb_y&1) ? 1 : 2;

  1574.                     mb_types_col[0] =

  1575.                     mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];

  1576.                     l1ref0 += dy*h->b8_stride;

  1577.                     l1ref1 += dy*h->b8_stride;

  1578.                     l1mv0 += 2*dy*h->b_stride;

  1579.                     l1mv1 += 2*dy*h->b_stride;

  1580.                     y_shift = 2;

  1581. 

  1582.                     if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))

  1583.                        && !is_b8x8)

  1584.                         *mb_type |= MB_TYPE_16x16;

  1585.                     else

  1586.                         *mb_type |= MB_TYPE_8x8;

  1587.                 }

  1588. 

  1589.                 for(i8=0; i8<4; i8++){

  1590.                     const int x8 = i8&1;

  1591.                     const int y8 = i8>>1;

  1592.                     int ref0, scale;

  1593.                     const int16_t (*l1mv)[2]= l1mv0;

  1594. 

  1595.                     if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  1596.                         continue;

  1597.                     h->sub_mb_type[i8] = sub_mb_type;

  1598. 

  1599.                     fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);

  1600.                     if(IS_INTRA(mb_types_col[y8])){

  1601.                         fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);

  1602.                         fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

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

  1604.                         continue;

  1605.                     }

  1606. 

  1607.                     ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];

  1608.                     if(ref0 >= 0)

  1609.                         ref0 = map_col_to_list0[0][ref0*2>>y_shift];

  1610.                     else{

  1611.                         ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];

  1612.                         l1mv= l1mv1;

  1613.                     }

  1614.                     scale = dist_scale_factor[ref0];

  1615.                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);

  1616. 

  1617.                     {

  1618.                         const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];

  1619.                         int my_col = (mv_col[1]<<y_shift)/2;

  1620.                         int mx = (scale * mv_col[0] + 128) >> 8;

  1621.                         int my = (scale * my_col + 128) >> 8;

  1622.                         fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);

  1623.                         fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);

  1624.                     }

  1625.                 }

  1626.                 return;

  1627.             }

  1628.         }

  1629. 

  1630.         /* one-to-one mv scaling */

  1631. 

  1632.         if(IS_16X16(*mb_type)){

  1633.             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);

  1634.             if(IS_INTRA(mb_type_col)){

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

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

  1637.                 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);

  1638.             }else{

  1639.                 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]

  1640.                                                 : map_col_to_list0[1][l1ref1[0]];

  1641.                 const int scale = dist_scale_factor[ref0];

  1642.                 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];

  1643.                 int mv_l0[2];

  1644.                 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;

  1645.                 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;

  1646.                 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);

  1647.                 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);

  1648.                 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);

  1649.             }

  1650.         }else{

  1651.             for(i8=0; i8<4; i8++){

  1652.                 const int x8 = i8&1;

  1653.                 const int y8 = i8>>1;

  1654.                 int ref0, scale;

  1655.                 const int16_t (*l1mv)[2]= l1mv0;

  1656. 

  1657.                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  1658.                     continue;

  1659.                 h->sub_mb_type[i8] = sub_mb_type;

  1660.                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);

  1661.                 if(IS_INTRA(mb_type_col)){

  1662.                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);

  1663.                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

  1664.                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);

  1665.                     continue;

  1666.                 }

  1667. 

  1668.                 ref0 = l1ref0[x8 + y8*h->b8_stride];

  1669.                 if(ref0 >= 0)

  1670.                     ref0 = map_col_to_list0[0][ref0];

  1671.                 else{

  1672.                     ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];

  1673.                     l1mv= l1mv1;

  1674.                 }

  1675.                 scale = dist_scale_factor[ref0];

  1676. 

  1677.                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);

  1678.                 if(IS_SUB_8X8(sub_mb_type)){

  1679.                     const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];

  1680.                     int mx = (scale * mv_col[0] + 128) >> 8;

  1681.                     int my = (scale * mv_col[1] + 128) >> 8;

  1682.                     fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);

  1683.                     fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);

  1684.                 }else

  1685.                 for(i4=0; i4<4; i4++){

  1686.                     const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];

  1687.                     int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];

  1688.                     mv_l0[0] = (scale * mv_col[0] + 128) >> 8;

  1689.                     mv_l0[1] = (scale * mv_col[1] + 128) >> 8;

  1690.                     *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =

  1691.                         pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);

  1692.                 }

  1693.             }

  1694.         }

  1695.     }

  1696. }

  1697. 

  1698. static inline void write_back_motion(H264Context *h, int mb_type){

  1699.     MpegEncContext * const s = &h->s;

  1700.     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;

  1701.     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;

  1702.     int list;

  1703. 

  1704.     if(!USES_LIST(mb_type, 0))

  1705.         fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);

  1706. 

  1707.     for(list=0; list<2; list++){

  1708.         int y;

  1709.         if(!USES_LIST(mb_type, list))

  1710.             continue;

  1711. 

  1712.         for(y=0; y<4; y++){

  1713.             *(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];

  1714.             *(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];

  1715.         }

  1716.         if( h->pps.cabac ) {

  1717.             for(y=0; y<4; y++){

  1718.                 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];

  1719.                 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];

  1720.             }

  1721.         }

  1722. 

  1723.         {

  1724.             int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];

  1725.             ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];

  1726.             ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];

  1727.             ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];

  1728.             ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];

  1729.         }

  1730.     }

  1731. 

  1732.     if(h->slice_type == B_TYPE && h->pps.cabac){

  1733.         if(IS_8X8(mb_type)){

  1734.             uint8_t *direct_table = &h->direct_table[b8_xy];

  1735.             direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;

  1736.             direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;

  1737.             direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;

  1738.         }

  1739.     }

  1740. }

  1741. 

  1742. /**

  1743.  * Decodes a network abstraction layer unit.

  1744.  * @param consumed is the number of bytes used as input

  1745.  * @param length is the length of the array

  1746.  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?

  1747.  * @returns decoded bytes, might be src+1 if no escapes

  1748.  */

  1749. static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){

  1750.     int i, si, di;

  1751.     uint8_t *dst;

  1752. 

  1753. //    src[0]&0x80;                //forbidden bit

  1754.     h->nal_ref_idc= src[0]>>5;

  1755.     h->nal_unit_type= src[0]&0x1F;

  1756. 

  1757.     src++; length--;

  1758. #if 0

  1759.     for(i=0; i<length; i++)

  1760.         printf("%2X ", src[i]);

  1761. #endif

  1762.     for(i=0; i+1<length; i+=2){

  1763.         if(src[i]) continue;

  1764.         if(i>0 && src[i-1]==0) i--;

  1765.         if(i+2<length && src[i+1]==0 && src[i+2]<=3){

  1766.             if(src[i+2]!=3){

  1767.                 /* startcode, so we must be past the end */

  1768.                 length=i;

  1769.             }

  1770.             break;

  1771.         }

  1772.     }

  1773. 

  1774.     if(i>=length-1){ //no escaped 0

  1775.         *dst_length= length;

  1776.         *consumed= length+1; //+1 for the header

  1777.         return src;

  1778.     }

  1779. 

  1780.     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);

  1781.     dst= h->rbsp_buffer;

  1782. 

  1783. //printf("decoding esc\n");

  1784.     si=di=0;

  1785.     while(si<length){

  1786.         //remove escapes (very rare 1:2^22)

  1787.         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){

  1788.             if(src[si+2]==3){ //escape

  1789.                 dst[di++]= 0;

  1790.                 dst[di++]= 0;

  1791.                 si+=3;

  1792.                 continue;

  1793.             }else //next start code

  1794.                 break;

  1795.         }

  1796. 

  1797.         dst[di++]= src[si++];

  1798.     }

  1799. 

  1800.     *dst_length= di;

  1801.     *consumed= si + 1;//+1 for the header

  1802. //FIXME store exact number of bits in the getbitcontext (its needed for decoding)

  1803.     return dst;

  1804. }

  1805. 

  1806. #if 0

  1807. /**

  1808.  * @param src the data which should be escaped

  1809.  * @param dst the target buffer, dst+1 == src is allowed as a special case

  1810.  * @param length the length of the src data

  1811.  * @param dst_length the length of the dst array

  1812.  * @returns length of escaped data in bytes or -1 if an error occured

  1813.  */

  1814. static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){

  1815.     int i, escape_count, si, di;

  1816.     uint8_t *temp;

  1817. 

  1818.     assert(length>=0);

  1819.     assert(dst_length>0);

  1820. 

  1821.     dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;

  1822. 

  1823.     if(length==0) return 1;

  1824. 

  1825.     escape_count= 0;

  1826.     for(i=0; i<length; i+=2){

  1827.         if(src[i]) continue;

  1828.         if(i>0 && src[i-1]==0)

  1829.             i--;

  1830.         if(i+2<length && src[i+1]==0 && src[i+2]<=3){

  1831.             escape_count++;

  1832.             i+=2;

  1833.         }

  1834.     }

  1835. 

  1836.     if(escape_count==0){

  1837.         if(dst+1 != src)

  1838.             memcpy(dst+1, src, length);

  1839.         return length + 1;

  1840.     }

  1841. 

  1842.     if(length + escape_count + 1> dst_length)

  1843.         return -1;

  1844. 

  1845.     //this should be damn rare (hopefully)

  1846. 

  1847.     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);

  1848.     temp= h->rbsp_buffer;

  1849. //printf("encoding esc\n");

  1850. 

  1851.     si= 0;

  1852.     di= 0;

  1853.     while(si < length){

  1854.         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){

  1855.             temp[di++]= 0; si++;

  1856.             temp[di++]= 0; si++;

  1857.             temp[di++]= 3;

  1858.             temp[di++]= src[si++];

  1859.         }

  1860.         else

  1861.             temp[di++]= src[si++];

  1862.     }

  1863.     memcpy(dst+1, temp, length+escape_count);

  1864. 

  1865.     assert(di == length+escape_count);

  1866. 

  1867.     return di + 1;

  1868. }

  1869. 

  1870. /**

  1871.  * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4

  1872.  */

  1873. static void encode_rbsp_trailing(PutBitContext *pb){

  1874.     int length;

  1875.     put_bits(pb, 1, 1);

  1876.     length= (-put_bits_count(pb))&7;

  1877.     if(length) put_bits(pb, length, 0);

  1878. }

  1879. #endif

  1880. 

  1881. /**

  1882.  * identifies the exact end of the bitstream

  1883.  * @return the length of the trailing, or 0 if damaged

  1884.  */

  1885. static int decode_rbsp_trailing(uint8_t *src){

  1886.     int v= *src;

  1887.     int r;

  1888. 

  1889.     tprintf("rbsp trailing %X\n", v);

  1890. 

  1891.     for(r=1; r<9; r++){

  1892.         if(v&1) return r;

  1893.         v>>=1;

  1894.     }

  1895.     return 0;

  1896. }

  1897. 

  1898. /**

  1899.  * idct tranforms the 16 dc values and dequantize them.

  1900.  * @param qp quantization parameter

  1901.  */

  1902. static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){

  1903. #define stride 16

  1904.     int i;

  1905.     int temp[16]; //FIXME check if this is a good idea

  1906.     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};

  1907.     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

  1908. 

  1909. //memset(block, 64, 2*256);

  1910. //return;

  1911.     for(i=0; i<4; i++){

  1912.         const int offset= y_offset[i];

  1913.         const int z0= block[offset+stride*0] + block[offset+stride*4];

  1914.         const int z1= block[offset+stride*0] - block[offset+stride*4];

  1915.         const int z2= block[offset+stride*1] - block[offset+stride*5];

  1916.         const int z3= block[offset+stride*1] + block[offset+stride*5];

  1917. 

  1918.         temp[4*i+0]= z0+z3;

  1919.         temp[4*i+1]= z1+z2;

  1920.         temp[4*i+2]= z1-z2;

  1921.         temp[4*i+3]= z0-z3;

  1922.     }

  1923. 

  1924.     for(i=0; i<4; i++){

  1925.         const int offset= x_offset[i];

  1926.         const int z0= temp[4*0+i] + temp[4*2+i];

  1927.         const int z1= temp[4*0+i] - temp[4*2+i];

  1928.         const int z2= temp[4*1+i] - temp[4*3+i];

  1929.         const int z3= temp[4*1+i] + temp[4*3+i];

  1930. 

  1931.         block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual

  1932.         block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));

  1933.         block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));

  1934.         block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));

  1935.     }

  1936. }

  1937. 

  1938. #if 0

  1939. /**

  1940.  * dct tranforms the 16 dc values.

  1941.  * @param qp quantization parameter ??? FIXME

  1942.  */

  1943. static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){

  1944. //    const int qmul= dequant_coeff[qp][0];

  1945.     int i;

  1946.     int temp[16]; //FIXME check if this is a good idea

  1947.     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};

  1948.     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

  1949. 

  1950.     for(i=0; i<4; i++){

  1951.         const int offset= y_offset[i];

  1952.         const int z0= block[offset+stride*0] + block[offset+stride*4];

  1953.         const int z1= block[offset+stride*0] - block[offset+stride*4];

  1954.         const int z2= block[offset+stride*1] - block[offset+stride*5];

  1955.         const int z3= block[offset+stride*1] + block[offset+stride*5];

  1956. 

  1957.         temp[4*i+0]= z0+z3;

  1958.         temp[4*i+1]= z1+z2;

  1959.         temp[4*i+2]= z1-z2;

  1960.         temp[4*i+3]= z0-z3;

  1961.     }

  1962. 

  1963.     for(i=0; i<4; i++){

  1964.         const int offset= x_offset[i];

  1965.         const int z0= temp[4*0+i] + temp[4*2+i];

  1966.         const int z1= temp[4*0+i] - temp[4*2+i];

  1967.         const int z2= temp[4*1+i] - temp[4*3+i];

  1968.         const int z3= temp[4*1+i] + temp[4*3+i];

  1969. 

  1970.         block[stride*0 +offset]= (z0 + z3)>>1;

  1971.         block[stride*2 +offset]= (z1 + z2)>>1;

  1972.         block[stride*8 +offset]= (z1 - z2)>>1;

  1973.         block[stride*10+offset]= (z0 - z3)>>1;

  1974.     }

  1975. }

  1976. #endif

  1977. 

  1978. #undef xStride

  1979. #undef stride

  1980. 

  1981. static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){

  1982.     const int stride= 16*2;

  1983.     const int xStride= 16;

  1984.     int a,b,c,d,e;

  1985. 

  1986.     a= block[stride*0 + xStride*0];

  1987.     b= block[stride*0 + xStride*1];

  1988.     c= block[stride*1 + xStride*0];

  1989.     d= block[stride*1 + xStride*1];

  1990. 

  1991.     e= a-b;

  1992.     a= a+b;

  1993.     b= c-d;

  1994.     c= c+d;

  1995. 

  1996.     block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;

  1997.     block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;

  1998.     block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;

  1999.     block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;

  2000. }

  2001. 

  2002. #if 0

  2003. static void chroma_dc_dct_c(DCTELEM *block){

  2004.     const int stride= 16*2;

  2005.     const int xStride= 16;

  2006.     int a,b,c,d,e;

  2007. 

  2008.     a= block[stride*0 + xStride*0];

  2009.     b= block[stride*0 + xStride*1];

  2010.     c= block[stride*1 + xStride*0];

  2011.     d= block[stride*1 + xStride*1];

  2012. 

  2013.     e= a-b;

  2014.     a= a+b;

  2015.     b= c-d;

  2016.     c= c+d;

  2017. 

  2018.     block[stride*0 + xStride*0]= (a+c);

  2019.     block[stride*0 + xStride*1]= (e+b);

  2020.     block[stride*1 + xStride*0]= (a-c);

  2021.     block[stride*1 + xStride*1]= (e-b);

  2022. }

  2023. #endif

  2024. 

  2025. /**

  2026.  * gets the chroma qp.

  2027.  */

  2028. static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){

  2029. 

  2030.     return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];

  2031. }

  2032. 

  2033. 

  2034. #if 0

  2035. static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){

  2036.     int i;

  2037.     //FIXME try int temp instead of block

  2038. 

  2039.     for(i=0; i<4; i++){

  2040.         const int d0= src1[0 + i*stride] - src2[0 + i*stride];

  2041.         const int d1= src1[1 + i*stride] - src2[1 + i*stride];

  2042.         const int d2= src1[2 + i*stride] - src2[2 + i*stride];

  2043.         const int d3= src1[3 + i*stride] - src2[3 + i*stride];

  2044.         const int z0= d0 + d3;

  2045.         const int z3= d0 - d3;

  2046.         const int z1= d1 + d2;

  2047.         const int z2= d1 - d2;

  2048. 

  2049.         block[0 + 4*i]=   z0 +   z1;

  2050.         block[1 + 4*i]= 2*z3 +   z2;

  2051.         block[2 + 4*i]=   z0 -   z1;

  2052.         block[3 + 4*i]=   z3 - 2*z2;

  2053.     }

  2054. 

  2055.     for(i=0; i<4; i++){

  2056.         const int z0= block[0*4 + i] + block[3*4 + i];

  2057.         const int z3= block[0*4 + i] - block[3*4 + i];

  2058.         const int z1= block[1*4 + i] + block[2*4 + i];

  2059.         const int z2= block[1*4 + i] - block[2*4 + i];

  2060. 

  2061.         block[0*4 + i]=   z0 +   z1;

  2062.         block[1*4 + i]= 2*z3 +   z2;

  2063.         block[2*4 + i]=   z0 -   z1;

  2064.         block[3*4 + i]=   z3 - 2*z2;

  2065.     }

  2066. }

  2067. #endif

  2068. 

  2069. //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close

  2070. //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)

  2071. static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){

  2072.     int i;

  2073.     const int * const quant_table= quant_coeff[qscale];

  2074.     const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;

  2075.     const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;

  2076.     const unsigned int threshold2= (threshold1<<1);

  2077.     int last_non_zero;

  2078. 

  2079.     if(seperate_dc){

  2080.         if(qscale<=18){

  2081.             //avoid overflows

  2082.             const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;

  2083.             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;

  2084.             const unsigned int dc_threshold2= (dc_threshold1<<1);

  2085. 

  2086.             int level= block[0]*quant_coeff[qscale+18][0];

  2087.             if(((unsigned)(level+dc_threshold1))>dc_threshold2){

  2088.                 if(level>0){

  2089.                     level= (dc_bias + level)>>(QUANT_SHIFT-2);

  2090.                     block[0]= level;

  2091.                 }else{

  2092.                     level= (dc_bias - level)>>(QUANT_SHIFT-2);

  2093.                     block[0]= -level;

  2094.                 }

  2095. //                last_non_zero = i;

  2096.             }else{

  2097.                 block[0]=0;

  2098.             }

  2099.         }else{

  2100.             const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;

  2101.             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;

  2102.             const unsigned int dc_threshold2= (dc_threshold1<<1);

  2103. 

  2104.             int level= block[0]*quant_table[0];

  2105.             if(((unsigned)(level+dc_threshold1))>dc_threshold2){

  2106.                 if(level>0){

  2107.                     level= (dc_bias + level)>>(QUANT_SHIFT+1);

  2108.                     block[0]= level;

  2109.                 }else{

  2110.                     level= (dc_bias - level)>>(QUANT_SHIFT+1);

  2111.                     block[0]= -level;

  2112.                 }

  2113. //                last_non_zero = i;

  2114.             }else{

  2115.                 block[0]=0;

  2116.             }

  2117.         }

  2118.         last_non_zero= 0;

  2119.         i=1;

  2120.     }else{

  2121.         last_non_zero= -1;

  2122.         i=0;

  2123.     }

  2124. 

  2125.     for(; i<16; i++){

  2126.         const int j= scantable[i];

  2127.         int level= block[j]*quant_table[j];

  2128. 

  2129. //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))

  2130. //           || bias-level >= (1<<(QMAT_SHIFT - 3))){

  2131.         if(((unsigned)(level+threshold1))>threshold2){

  2132.             if(level>0){

  2133.                 level= (bias + level)>>QUANT_SHIFT;

  2134.                 block[j]= level;

  2135.             }else{

  2136.                 level= (bias - level)>>QUANT_SHIFT;

  2137.                 block[j]= -level;

  2138.             }

  2139.             last_non_zero = i;

  2140.         }else{

  2141.             block[j]=0;

  2142.         }

  2143.     }

  2144. 

  2145.     return last_non_zero;

  2146. }

  2147. 

  2148. static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){

  2149.     const uint32_t a= ((uint32_t*)(src-stride))[0];

  2150.     ((uint32_t*)(src+0*stride))[0]= a;

  2151.     ((uint32_t*)(src+1*stride))[0]= a;

  2152.     ((uint32_t*)(src+2*stride))[0]= a;

  2153.     ((uint32_t*)(src+3*stride))[0]= a;

  2154. }

  2155. 

  2156. static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){

  2157.     ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;

  2158.     ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;

  2159.     ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;

  2160.     ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;

  2161. }

  2162. 

  2163. static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){

  2164.     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]

  2165.                    + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;

  2166. 

  2167.     ((uint32_t*)(src+0*stride))[0]=

  2168.     ((uint32_t*)(src+1*stride))[0]=

  2169.     ((uint32_t*)(src+2*stride))[0]=

  2170.     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;

  2171. }

  2172. 

  2173. static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){

  2174.     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;

  2175. 

  2176.     ((uint32_t*)(src+0*stride))[0]=

  2177.     ((uint32_t*)(src+1*stride))[0]=

  2178.     ((uint32_t*)(src+2*stride))[0]=

  2179.     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;

  2180. }

  2181. 

  2182. static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){

  2183.     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;

  2184. 

  2185.     ((uint32_t*)(src+0*stride))[0]=

  2186.     ((uint32_t*)(src+1*stride))[0]=

  2187.     ((uint32_t*)(src+2*stride))[0]=

  2188.     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;

  2189. }

  2190. 

  2191. static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){

  2192.     ((uint32_t*)(src+0*stride))[0]=

  2193.     ((uint32_t*)(src+1*stride))[0]=

  2194.     ((uint32_t*)(src+2*stride))[0]=

  2195.     ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;

  2196. }

  2197. 

  2198. 

  2199. #define LOAD_TOP_RIGHT_EDGE\

  2200.     const int t4= topright[0];\

  2201.     const int t5= topright[1];\

  2202.     const int t6= topright[2];\

  2203.     const int t7= topright[3];\

  2204. 

  2205. #define LOAD_LEFT_EDGE\

  2206.     const int l0= src[-1+0*stride];\

  2207.     const int l1= src[-1+1*stride];\

  2208.     const int l2= src[-1+2*stride];\

  2209.     const int l3= src[-1+3*stride];\

  2210. 

  2211. #define LOAD_TOP_EDGE\

  2212.     const int t0= src[ 0-1*stride];\

  2213.     const int t1= src[ 1-1*stride];\

  2214.     const int t2= src[ 2-1*stride];\

  2215.     const int t3= src[ 3-1*stride];\

  2216. 

  2217. static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){

  2218.     const int lt= src[-1-1*stride];

  2219.     LOAD_TOP_EDGE

  2220.     LOAD_LEFT_EDGE

  2221. 

  2222.     src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;

  2223.     src[0+2*stride]=

  2224.     src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;

  2225.     src[0+1*stride]=

  2226.     src[1+2*stride]=

  2227.     src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;

  2228.     src[0+0*stride]=

  2229.     src[1+1*stride]=

  2230.     src[2+2*stride]=

  2231.     src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;

  2232.     src[1+0*stride]=

  2233.     src[2+1*stride]=

  2234.     src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;

  2235.     src[2+0*stride]=

  2236.     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;

  2237.     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;

  2238. }

  2239. 

  2240. static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){

  2241.     LOAD_TOP_EDGE

  2242.     LOAD_TOP_RIGHT_EDGE

  2243. //    LOAD_LEFT_EDGE

  2244. 

  2245.     src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;

  2246.     src[1+0*stride]=

  2247.     src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;

  2248.     src[2+0*stride]=

  2249.     src[1+1*stride]=

  2250.     src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;

  2251.     src[3+0*stride]=

  2252.     src[2+1*stride]=

  2253.     src[1+2*stride]=

  2254.     src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;

  2255.     src[3+1*stride]=

  2256.     src[2+2*stride]=

  2257.     src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;

  2258.     src[3+2*stride]=

  2259.     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;

  2260.     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;

  2261. }

  2262. 

  2263. static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){

  2264.     const int lt= src[-1-1*stride];

  2265.     LOAD_TOP_EDGE

  2266.     LOAD_LEFT_EDGE

  2267.     const __attribute__((unused)) int unu= l3;

  2268. 

  2269.     src[0+0*stride]=

  2270.     src[1+2*stride]=(lt + t0 + 1)>>1;

  2271.     src[1+0*stride]=

  2272.     src[2+2*stride]=(t0 + t1 + 1)>>1;

  2273.     src[2+0*stride]=

  2274.     src[3+2*stride]=(t1 + t2 + 1)>>1;

  2275.     src[3+0*stride]=(t2 + t3 + 1)>>1;

  2276.     src[0+1*stride]=

  2277.     src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;

  2278.     src[1+1*stride]=

  2279.     src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;

  2280.     src[2+1*stride]=

  2281.     src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;

  2282.     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;

  2283.     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;

  2284.     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;

  2285. }

  2286. 

  2287. static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){

  2288.     LOAD_TOP_EDGE

  2289.     LOAD_TOP_RIGHT_EDGE

  2290.     const __attribute__((unused)) int unu= t7;

  2291. 

  2292.     src[0+0*stride]=(t0 + t1 + 1)>>1;

  2293.     src[1+0*stride]=

  2294.     src[0+2*stride]=(t1 + t2 + 1)>>1;

  2295.     src[2+0*stride]=

  2296.     src[1+2*stride]=(t2 + t3 + 1)>>1;

  2297.     src[3+0*stride]=

  2298.     src[2+2*stride]=(t3 + t4+ 1)>>1;

  2299.     src[3+2*stride]=(t4 + t5+ 1)>>1;

  2300.     src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;

  2301.     src[1+1*stride]=

  2302.     src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;

  2303.     src[2+1*stride]=

  2304.     src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;

  2305.     src[3+1*stride]=

  2306.     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;

  2307.     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;

  2308. }

  2309. 

  2310. static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){

  2311.     LOAD_LEFT_EDGE

  2312. 

  2313.     src[0+0*stride]=(l0 + l1 + 1)>>1;

  2314.     src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;

  2315.     src[2+0*stride]=

  2316.     src[0+1*stride]=(l1 + l2 + 1)>>1;

  2317.     src[3+0*stride]=

  2318.     src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;

  2319.     src[2+1*stride]=

  2320.     src[0+2*stride]=(l2 + l3 + 1)>>1;

  2321.     src[3+1*stride]=

  2322.     src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;

  2323.     src[3+2*stride]=

  2324.     src[1+3*stride]=

  2325.     src[0+3*stride]=

  2326.     src[2+2*stride]=

  2327.     src[2+3*stride]=

  2328.     src[3+3*stride]=l3;

  2329. }

  2330. 

  2331. static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){

  2332.     const int lt= src[-1-1*stride];

  2333.     LOAD_TOP_EDGE

  2334.     LOAD_LEFT_EDGE

  2335.     const __attribute__((unused)) int unu= t3;

  2336. 

  2337.     src[0+0*stride]=

  2338.     src[2+1*stride]=(lt + l0 + 1)>>1;

  2339.     src[1+0*stride]=

  2340.     src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;

  2341.     src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;

  2342.     src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;

  2343.     src[0+1*stride]=

  2344.     src[2+2*stride]=(l0 + l1 + 1)>>1;

  2345.     src[1+1*stride]=

  2346.     src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;

  2347.     src[0+2*stride]=

  2348.     src[2+3*stride]=(l1 + l2+ 1)>>1;

  2349.     src[1+2*stride]=

  2350.     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;

  2351.     src[0+3*stride]=(l2 + l3 + 1)>>1;

  2352.     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;

  2353. }

  2354. 

  2355. static void pred16x16_vertical_c(uint8_t *src, int stride){

  2356.     int i;

  2357.     const uint32_t a= ((uint32_t*)(src-stride))[0];

  2358.     const uint32_t b= ((uint32_t*)(src-stride))[1];

  2359.     const uint32_t c= ((uint32_t*)(src-stride))[2];

  2360.     const uint32_t d= ((uint32_t*)(src-stride))[3];

  2361. 

  2362.     for(i=0; i<16; i++){

  2363.         ((uint32_t*)(src+i*stride))[0]= a;

  2364.         ((uint32_t*)(src+i*stride))[1]= b;

  2365.         ((uint32_t*)(src+i*stride))[2]= c;

  2366.         ((uint32_t*)(src+i*stride))[3]= d;

  2367.     }

  2368. }

  2369. 

  2370. static void pred16x16_horizontal_c(uint8_t *src, int stride){

  2371.     int i;

  2372. 

  2373.     for(i=0; i<16; i++){

  2374.         ((uint32_t*)(src+i*stride))[0]=

  2375.         ((uint32_t*)(src+i*stride))[1]=

  2376.         ((uint32_t*)(src+i*stride))[2]=

  2377.         ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;

  2378.     }

  2379. }

  2380. 

  2381. static void pred16x16_dc_c(uint8_t *src, int stride){

  2382.     int i, dc=0;

  2383. 

  2384.     for(i=0;i<16; i++){

  2385.         dc+= src[-1+i*stride];

  2386.     }

  2387. 

  2388.     for(i=0;i<16; i++){

  2389.         dc+= src[i-stride];

  2390.     }

  2391. 

  2392.     dc= 0x01010101*((dc + 16)>>5);

  2393. 

  2394.     for(i=0; i<16; i++){

  2395.         ((uint32_t*)(src+i*stride))[0]=

  2396.         ((uint32_t*)(src+i*stride))[1]=

  2397.         ((uint32_t*)(src+i*stride))[2]=

  2398.         ((uint32_t*)(src+i*stride))[3]= dc;

  2399.     }

  2400. }

  2401. 

  2402. static void pred16x16_left_dc_c(uint8_t *src, int stride){

  2403.     int i, dc=0;

  2404. 

  2405.     for(i=0;i<16; i++){

  2406.         dc+= src[-1+i*stride];

  2407.     }

  2408. 

  2409.     dc= 0x01010101*((dc + 8)>>4);

  2410. 

  2411.     for(i=0; i<16; i++){

  2412.         ((uint32_t*)(src+i*stride))[0]=

  2413.         ((uint32_t*)(src+i*stride))[1]=

  2414.         ((uint32_t*)(src+i*stride))[2]=

  2415.         ((uint32_t*)(src+i*stride))[3]= dc;

  2416.     }

  2417. }

  2418. 

  2419. static void pred16x16_top_dc_c(uint8_t *src, int stride){

  2420.     int i, dc=0;

  2421. 

  2422.     for(i=0;i<16; i++){

  2423.         dc+= src[i-stride];

  2424.     }

  2425.     dc= 0x01010101*((dc + 8)>>4);

  2426. 

  2427.     for(i=0; i<16; i++){

  2428.         ((uint32_t*)(src+i*stride))[0]=

  2429.         ((uint32_t*)(src+i*stride))[1]=

  2430.         ((uint32_t*)(src+i*stride))[2]=

  2431.         ((uint32_t*)(src+i*stride))[3]= dc;

  2432.     }

  2433. }

  2434. 

  2435. static void pred16x16_128_dc_c(uint8_t *src, int stride){

  2436.     int i;

  2437. 

  2438.     for(i=0; i<16; i++){

  2439.         ((uint32_t*)(src+i*stride))[0]=

  2440.         ((uint32_t*)(src+i*stride))[1]=

  2441.         ((uint32_t*)(src+i*stride))[2]=

  2442.         ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;

  2443.     }

  2444. }

  2445. 

  2446. static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){

  2447.   int i, j, k;

  2448.   int a;

  2449.   uint8_t *cm = cropTbl + MAX_NEG_CROP;

  2450.   const uint8_t * const src0 = src+7-stride;

  2451.   const uint8_t *src1 = src+8*stride-1;

  2452.   const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;

  2453.   int H = src0[1] - src0[-1];

  2454.   int V = src1[0] - src2[ 0];

  2455.   for(k=2; k<=8; ++k) {

  2456.     src1 += stride; src2 -= stride;

  2457.     H += k*(src0[k] - src0[-k]);

  2458.     V += k*(src1[0] - src2[ 0]);

  2459.   }

  2460.   if(svq3){

  2461.     H = ( 5*(H/4) ) / 16;

  2462.     V = ( 5*(V/4) ) / 16;

  2463. 

  2464.     /* required for 100% accuracy */

  2465.     i = H; H = V; V = i;

  2466.   }else{

  2467.     H = ( 5*H+32 ) >> 6;

  2468.     V = ( 5*V+32 ) >> 6;

  2469.   }

  2470. 

  2471.   a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);

  2472.   for(j=16; j>0; --j) {

  2473.     int b = a;

  2474.     a += V;

  2475.     for(i=-16; i<0; i+=4) {

  2476.       src[16+i] = cm[ (b    ) >> 5 ];

  2477.       src[17+i] = cm[ (b+  H) >> 5 ];

  2478.       src[18+i] = cm[ (b+2*H) >> 5 ];

  2479.       src[19+i] = cm[ (b+3*H) >> 5 ];

  2480.       b += 4*H;

  2481.     }

  2482.     src += stride;

  2483.   }

  2484. }

  2485. 

  2486. static void pred16x16_plane_c(uint8_t *src, int stride){

  2487.     pred16x16_plane_compat_c(src, stride, 0);

  2488. }

  2489. 

  2490. static void pred8x8_vertical_c(uint8_t *src, int stride){

  2491.     int i;

  2492.     const uint32_t a= ((uint32_t*)(src-stride))[0];

  2493.     const uint32_t b= ((uint32_t*)(src-stride))[1];

  2494. 

  2495.     for(i=0; i<8; i++){

  2496.         ((uint32_t*)(src+i*stride))[0]= a;

  2497.         ((uint32_t*)(src+i*stride))[1]= b;

  2498.     }

  2499. }

  2500. 

  2501. static void pred8x8_horizontal_c(uint8_t *src, int stride){

  2502.     int i;

  2503. 

  2504.     for(i=0; i<8; i++){

  2505.         ((uint32_t*)(src+i*stride))[0]=

  2506.         ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;

  2507.     }

  2508. }

  2509. 

  2510. static void pred8x8_128_dc_c(uint8_t *src, int stride){

  2511.     int i;

  2512. 

  2513.     for(i=0; i<8; i++){

  2514.         ((uint32_t*)(src+i*stride))[0]=

  2515.         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;

  2516.     }

  2517. }

  2518. 

  2519. static void pred8x8_left_dc_c(uint8_t *src, int stride){

  2520.     int i;

  2521.     int dc0, dc2;

  2522. 

  2523.     dc0=dc2=0;

  2524.     for(i=0;i<4; i++){

  2525.         dc0+= src[-1+i*stride];

  2526.         dc2+= src[-1+(i+4)*stride];

  2527.     }

  2528.     dc0= 0x01010101*((dc0 + 2)>>2);

  2529.     dc2= 0x01010101*((dc2 + 2)>>2);

  2530. 

  2531.     for(i=0; i<4; i++){

  2532.         ((uint32_t*)(src+i*stride))[0]=

  2533.         ((uint32_t*)(src+i*stride))[1]= dc0;

  2534.     }

  2535.     for(i=4; i<8; i++){

  2536.         ((uint32_t*)(src+i*stride))[0]=

  2537.         ((uint32_t*)(src+i*stride))[1]= dc2;

  2538.     }

  2539. }

  2540. 

  2541. static void pred8x8_top_dc_c(uint8_t *src, int stride){

  2542.     int i;

  2543.     int dc0, dc1;

  2544. 

  2545.     dc0=dc1=0;

  2546.     for(i=0;i<4; i++){

  2547.         dc0+= src[i-stride];

  2548.         dc1+= src[4+i-stride];

  2549.     }

  2550.     dc0= 0x01010101*((dc0 + 2)>>2);

  2551.     dc1= 0x01010101*((dc1 + 2)>>2);

  2552. 

  2553.     for(i=0; i<4; i++){

  2554.         ((uint32_t*)(src+i*stride))[0]= dc0;

  2555.         ((uint32_t*)(src+i*stride))[1]= dc1;

  2556.     }

  2557.     for(i=4; i<8; i++){

  2558.         ((uint32_t*)(src+i*stride))[0]= dc0;

  2559.         ((uint32_t*)(src+i*stride))[1]= dc1;

  2560.     }

  2561. }

  2562. 

  2563. 

  2564. static void pred8x8_dc_c(uint8_t *src, int stride){

  2565.     int i;

  2566.     int dc0, dc1, dc2, dc3;

  2567. 

  2568.     dc0=dc1=dc2=0;

  2569.     for(i=0;i<4; i++){

  2570.         dc0+= src[-1+i*stride] + src[i-stride];

  2571.         dc1+= src[4+i-stride];

  2572.         dc2+= src[-1+(i+4)*stride];

  2573.     }

  2574.     dc3= 0x01010101*((dc1 + dc2 + 4)>>3);

  2575.     dc0= 0x01010101*((dc0 + 4)>>3);

  2576.     dc1= 0x01010101*((dc1 + 2)>>2);

  2577.     dc2= 0x01010101*((dc2 + 2)>>2);

  2578. 

  2579.     for(i=0; i<4; i++){

  2580.         ((uint32_t*)(src+i*stride))[0]= dc0;

  2581.         ((uint32_t*)(src+i*stride))[1]= dc1;

  2582.     }

  2583.     for(i=4; i<8; i++){

  2584.         ((uint32_t*)(src+i*stride))[0]= dc2;

  2585.         ((uint32_t*)(src+i*stride))[1]= dc3;

  2586.     }

  2587. }

  2588. 

  2589. static void pred8x8_plane_c(uint8_t *src, int stride){

  2590.   int j, k;

  2591.   int a;

  2592.   uint8_t *cm = cropTbl + MAX_NEG_CROP;

  2593.   const uint8_t * const src0 = src+3-stride;

  2594.   const uint8_t *src1 = src+4*stride-1;

  2595.   const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;

  2596.   int H = src0[1] - src0[-1];

  2597.   int V = src1[0] - src2[ 0];

  2598.   for(k=2; k<=4; ++k) {

  2599.     src1 += stride; src2 -= stride;

  2600.     H += k*(src0[k] - src0[-k]);

  2601.     V += k*(src1[0] - src2[ 0]);

  2602.   }

  2603.   H = ( 17*H+16 ) >> 5;

  2604.   V = ( 17*V+16 ) >> 5;

  2605. 

  2606.   a = 16*(src1[0] + src2[8]+1) - 3*(V+H);

  2607.   for(j=8; j>0; --j) {

  2608.     int b = a;

  2609.     a += V;

  2610.     src[0] = cm[ (b    ) >> 5 ];

  2611.     src[1] = cm[ (b+  H) >> 5 ];

  2612.     src[2] = cm[ (b+2*H) >> 5 ];

  2613.     src[3] = cm[ (b+3*H) >> 5 ];

  2614.     src[4] = cm[ (b+4*H) >> 5 ];

  2615.     src[5] = cm[ (b+5*H) >> 5 ];

  2616.     src[6] = cm[ (b+6*H) >> 5 ];

  2617.     src[7] = cm[ (b+7*H) >> 5 ];

  2618.     src += stride;

  2619.   }

  2620. }

  2621. 

  2622. #define SRC(x,y) src[(x)+(y)*stride]

  2623. #define PL(y) \

  2624.     const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;

  2625. #define PREDICT_8x8_LOAD_LEFT \

  2626.     const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \

  2627.                      + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \

  2628.     PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \

  2629.     const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2

  2630. 

  2631. #define PT(x) \

  2632.     const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;

  2633. #define PREDICT_8x8_LOAD_TOP \

  2634.     const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \

  2635.                      + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \

  2636.     PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \

  2637.     const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \

  2638.                      + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2

  2639. 

  2640. #define PTR(x) \

  2641.     t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;

  2642. #define PREDICT_8x8_LOAD_TOPRIGHT \

  2643.     int t8, t9, t10, t11, t12, t13, t14, t15; \

  2644.     if(has_topright) { \

  2645.         PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \

  2646.         t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \

  2647.     } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);

  2648. 

  2649. #define PREDICT_8x8_LOAD_TOPLEFT \

  2650.     const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2

  2651. 

  2652. #define PREDICT_8x8_DC(v) \

  2653.     int y; \

  2654.     for( y = 0; y < 8; y++ ) { \

  2655.         ((uint32_t*)src)[0] = \

  2656.         ((uint32_t*)src)[1] = v; \

  2657.         src += stride; \

  2658.     }

  2659. 

  2660. static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2661. {

  2662.     PREDICT_8x8_DC(0x80808080);

  2663. }

  2664. static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2665. {

  2666.     PREDICT_8x8_LOAD_LEFT;

  2667.     const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;

  2668.     PREDICT_8x8_DC(dc);

  2669. }

  2670. static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2671. {

  2672.     PREDICT_8x8_LOAD_TOP;

  2673.     const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;

  2674.     PREDICT_8x8_DC(dc);

  2675. }

  2676. static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2677. {

  2678.     PREDICT_8x8_LOAD_LEFT;

  2679.     PREDICT_8x8_LOAD_TOP;

  2680.     const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7

  2681.                          +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;

  2682.     PREDICT_8x8_DC(dc);

  2683. }

  2684. static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2685. {

  2686.     PREDICT_8x8_LOAD_LEFT;

  2687. #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\

  2688.                ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y

  2689.     ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);

  2690. #undef ROW

  2691. }

  2692. static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2693. {

  2694.     int y;

  2695.     PREDICT_8x8_LOAD_TOP;

  2696.     src[0] = t0;

  2697.     src[1] = t1;

  2698.     src[2] = t2;

  2699.     src[3] = t3;

  2700.     src[4] = t4;

  2701.     src[5] = t5;

  2702.     src[6] = t6;

  2703.     src[7] = t7;

  2704.     for( y = 1; y < 8; y++ )

  2705.         *(uint64_t*)(src+y*stride) = *(uint64_t*)src;

  2706. }

  2707. static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2708. {

  2709.     PREDICT_8x8_LOAD_TOP;

  2710.     PREDICT_8x8_LOAD_TOPRIGHT;

  2711.     SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;

  2712.     SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;

  2713.     SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;

  2714.     SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;

  2715.     SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;

  2716.     SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;

  2717.     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;

  2718.     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;

  2719.     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;

  2720.     SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;

  2721.     SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;

  2722.     SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;

  2723.     SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;

  2724.     SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;

  2725.     SRC(7,7)= (t14 + 3*t15 + 2) >> 2;

  2726. }

  2727. static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2728. {

  2729.     PREDICT_8x8_LOAD_TOP;

  2730.     PREDICT_8x8_LOAD_LEFT;

  2731.     PREDICT_8x8_LOAD_TOPLEFT;

  2732.     SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;

  2733.     SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;

  2734.     SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;

  2735.     SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;

  2736.     SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;

  2737.     SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;

  2738.     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;

  2739.     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;

  2740.     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;

  2741.     SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;

  2742.     SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;

  2743.     SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;

  2744.     SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;

  2745.     SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;

  2746.     SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;

  2747. 

  2748. }

  2749. static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2750. {

  2751.     PREDICT_8x8_LOAD_TOP;

  2752.     PREDICT_8x8_LOAD_LEFT;

  2753.     PREDICT_8x8_LOAD_TOPLEFT;

  2754.     SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;

  2755.     SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;

  2756.     SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;

  2757.     SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;

  2758.     SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;

  2759.     SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;

  2760.     SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;

  2761.     SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;

  2762.     SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;

  2763.     SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;

  2764.     SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;

  2765.     SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;

  2766.     SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;

  2767.     SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;

  2768.     SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;

  2769.     SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;

  2770.     SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;

  2771.     SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;

  2772.     SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;

  2773.     SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;

  2774.     SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;

  2775.     SRC(7,0)= (t6 + t7 + 1) >> 1;

  2776. }

  2777. static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2778. {

  2779.     PREDICT_8x8_LOAD_TOP;

  2780.     PREDICT_8x8_LOAD_LEFT;

  2781.     PREDICT_8x8_LOAD_TOPLEFT;

  2782.     SRC(0,7)= (l6 + l7 + 1) >> 1;

  2783.     SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;

  2784.     SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;

  2785.     SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;

  2786.     SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;

  2787.     SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;

  2788.     SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;

  2789.     SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;

  2790.     SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;

  2791.     SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;

  2792.     SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;

  2793.     SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;

  2794.     SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;

  2795.     SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;

  2796.     SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;

  2797.     SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;

  2798.     SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;

  2799.     SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;

  2800.     SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;

  2801.     SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;

  2802.     SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;

  2803.     SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;

  2804. }

  2805. static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2806. {

  2807.     PREDICT_8x8_LOAD_TOP;

  2808.     PREDICT_8x8_LOAD_TOPRIGHT;

  2809.     SRC(0,0)= (t0 + t1 + 1) >> 1;

  2810.     SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;

  2811.     SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;

  2812.     SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;

  2813.     SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;

  2814.     SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;

  2815.     SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;

  2816.     SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;

  2817.     SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;

  2818.     SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;

  2819.     SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;

  2820.     SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;

  2821.     SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;

  2822.     SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;

  2823.     SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;

  2824.     SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;

  2825.     SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;

  2826.     SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;

  2827.     SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;

  2828.     SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;

  2829.     SRC(7,6)= (t10 + t11 + 1) >> 1;

  2830.     SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;

  2831. }

  2832. static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)

  2833. {

  2834.     PREDICT_8x8_LOAD_LEFT;

  2835.     SRC(0,0)= (l0 + l1 + 1) >> 1;

  2836.     SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;

  2837.     SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;

  2838.     SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;

  2839.     SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;

  2840.     SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;

  2841.     SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;

  2842.     SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;

  2843.     SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;

  2844.     SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;

  2845.     SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;

  2846.     SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;

  2847.     SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;

  2848.     SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;

  2849.     SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=

  2850.     SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=

  2851.     SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=

  2852.     SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;

  2853. }

  2854. #undef PREDICT_8x8_LOAD_LEFT

  2855. #undef PREDICT_8x8_LOAD_TOP

  2856. #undef PREDICT_8x8_LOAD_TOPLEFT

  2857. #undef PREDICT_8x8_LOAD_TOPRIGHT

  2858. #undef PREDICT_8x8_DC

  2859. #undef PTR

  2860. #undef PT

  2861. #undef PL

  2862. #undef SRC

  2863. 

  2864. static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,

  2865.                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  2866.                            int src_x_offset, int src_y_offset,

  2867.                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){

  2868.     MpegEncContext * const s = &h->s;

  2869.     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;

  2870.     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;

  2871.     const int luma_xy= (mx&3) + ((my&3)<<2);

  2872.     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;

  2873.     uint8_t * src_cb, * src_cr;

  2874.     int extra_width= h->emu_edge_width;

  2875.     int extra_height= h->emu_edge_height;

  2876.     int emu=0;

  2877.     const int full_mx= mx>>2;

  2878.     const int full_my= my>>2;

  2879.     const int pic_width  = 16*s->mb_width;

  2880.     const int pic_height = 16*s->mb_height >> MB_MBAFF;

  2881. 

  2882.     if(!pic->data[0])

  2883.         return;

  2884. 

  2885.     if(mx&7) extra_width -= 3;

  2886.     if(my&7) extra_height -= 3;

  2887. 

  2888.     if(   full_mx < 0-extra_width

  2889.        || full_my < 0-extra_height

  2890.        || full_mx + 16/*FIXME*/ > pic_width + extra_width

  2891.        || full_my + 16/*FIXME*/ > pic_height + extra_height){

  2892.         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);

  2893.             src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;

  2894.         emu=1;

  2895.     }

  2896. 

  2897.     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?

  2898.     if(!square){

  2899.         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);

  2900.     }

  2901. 

  2902.     if(s->flags&CODEC_FLAG_GRAY) return;

  2903. 

  2904.     if(MB_MBAFF){

  2905.         // chroma offset when predicting from a field of opposite parity

  2906.         my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));

  2907.         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);

  2908.     }

  2909.     src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;

  2910.     src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;

  2911. 

  2912.     if(emu){

  2913.         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);

  2914.             src_cb= s->edge_emu_buffer;

  2915.     }

  2916.     chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);

  2917. 

  2918.     if(emu){

  2919.         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);

  2920.             src_cr= s->edge_emu_buffer;

  2921.     }

  2922.     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);

  2923. }

  2924. 

  2925. static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,

  2926.                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  2927.                            int x_offset, int y_offset,

  2928.                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,

  2929.                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,

  2930.                            int list0, int list1){

  2931.     MpegEncContext * const s = &h->s;

  2932.     qpel_mc_func *qpix_op=  qpix_put;

  2933.     h264_chroma_mc_func chroma_op= chroma_put;

  2934. 

  2935.     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;

  2936.     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;

  2937.     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;

  2938.     x_offset += 8*s->mb_x;

  2939.     y_offset += 8*(s->mb_y >> MB_MBAFF);

  2940. 

  2941.     if(list0){

  2942.         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];

  2943.         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,

  2944.                            dest_y, dest_cb, dest_cr, x_offset, y_offset,

  2945.                            qpix_op, chroma_op);

  2946. 

  2947.         qpix_op=  qpix_avg;

  2948.         chroma_op= chroma_avg;

  2949.     }

  2950. 

  2951.     if(list1){

  2952.         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];

  2953.         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,

  2954.                            dest_y, dest_cb, dest_cr, x_offset, y_offset,

  2955.                            qpix_op, chroma_op);

  2956.     }

  2957. }

  2958. 

  2959. static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,

  2960.                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  2961.                            int x_offset, int y_offset,

  2962.                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,

  2963.                            h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,

  2964.                            h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,

  2965.                            int list0, int list1){

  2966.     MpegEncContext * const s = &h->s;

  2967. 

  2968.     dest_y  += 2*x_offset + 2*y_offset*h->  mb_linesize;

  2969.     dest_cb +=   x_offset +   y_offset*h->mb_uvlinesize;

  2970.     dest_cr +=   x_offset +   y_offset*h->mb_uvlinesize;

  2971.     x_offset += 8*s->mb_x;

  2972.     y_offset += 8*(s->mb_y >> MB_MBAFF);

  2973. 

  2974.     if(list0 && list1){

  2975.         /* don't optimize for luma-only case, since B-frames usually

  2976.          * use implicit weights => chroma too. */

  2977.         uint8_t *tmp_cb = s->obmc_scratchpad;

  2978.         uint8_t *tmp_cr = s->obmc_scratchpad + 8;

  2979.         uint8_t *tmp_y  = s->obmc_scratchpad + 8*h->mb_uvlinesize;

  2980.         int refn0 = h->ref_cache[0][ scan8[n] ];

  2981.         int refn1 = h->ref_cache[1][ scan8[n] ];

  2982. 

  2983.         mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,

  2984.                     dest_y, dest_cb, dest_cr,

  2985.                     x_offset, y_offset, qpix_put, chroma_put);

  2986.         mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,

  2987.                     tmp_y, tmp_cb, tmp_cr,

  2988.                     x_offset, y_offset, qpix_put, chroma_put);

  2989. 

  2990.         if(h->use_weight == 2){

  2991.             int weight0 = h->implicit_weight[refn0][refn1];

  2992.             int weight1 = 64 - weight0;

  2993.             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);

  2994.             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);

  2995.             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);

  2996.         }else{

  2997.             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,

  2998.                             h->luma_weight[0][refn0], h->luma_weight[1][refn1],

  2999.                             h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);

  3000.             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,

  3001.                             h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],

  3002.                             h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);

  3003.             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,

  3004.                             h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],

  3005.                             h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);

  3006.         }

  3007.     }else{

  3008.         int list = list1 ? 1 : 0;

  3009.         int refn = h->ref_cache[list][ scan8[n] ];

  3010.         Picture *ref= &h->ref_list[list][refn];

  3011.         mc_dir_part(h, ref, n, square, chroma_height, delta, list,

  3012.                     dest_y, dest_cb, dest_cr, x_offset, y_offset,

  3013.                     qpix_put, chroma_put);

  3014. 

  3015.         luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,

  3016.                        h->luma_weight[list][refn], h->luma_offset[list][refn]);

  3017.         if(h->use_weight_chroma){

  3018.             chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,

  3019.                              h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);

  3020.             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,

  3021.                              h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);

  3022.         }

  3023.     }

  3024. }

  3025. 

  3026. static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,

  3027.                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  3028.                            int x_offset, int y_offset,

  3029.                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,

  3030.                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,

  3031.                            h264_weight_func *weight_op, h264_biweight_func *weight_avg,

  3032.                            int list0, int list1){

  3033.     if((h->use_weight==2 && list0 && list1

  3034.         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))

  3035.        || h->use_weight==1)

  3036.         mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,

  3037.                          x_offset, y_offset, qpix_put, chroma_put,

  3038.                          weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);

  3039.     else

  3040.         mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,

  3041.                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);

  3042. }

  3043. 

  3044. static inline void prefetch_motion(H264Context *h, int list){

  3045.     /* fetch pixels for estimated mv 4 macroblocks ahead

  3046.      * optimized for 64byte cache lines */

  3047.     MpegEncContext * const s = &h->s;

  3048.     const int refn = h->ref_cache[list][scan8[0]];

  3049.     if(refn >= 0){

  3050.         const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;

  3051.         const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;

  3052.         uint8_t **src= h->ref_list[list][refn].data;

  3053.         int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;

  3054.         s->dsp.prefetch(src[0]+off, s->linesize, 4);

  3055.         off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;

  3056.         s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);

  3057.     }

  3058. }

  3059. 

  3060. static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,

  3061.                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),

  3062.                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),

  3063.                       h264_weight_func *weight_op, h264_biweight_func *weight_avg){

  3064.     MpegEncContext * const s = &h->s;

  3065.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  3066.     const int mb_type= s->current_picture.mb_type[mb_xy];

  3067. 

  3068.     assert(IS_INTER(mb_type));

  3069. 

  3070.     prefetch_motion(h, 0);

  3071. 

  3072.     if(IS_16X16(mb_type)){

  3073.         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,

  3074.                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],

  3075.                 &weight_op[0], &weight_avg[0],

  3076.                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));

  3077.     }else if(IS_16X8(mb_type)){

  3078.         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,

  3079.                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],

  3080.                 &weight_op[1], &weight_avg[1],

  3081.                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));

  3082.         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,

  3083.                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],

  3084.                 &weight_op[1], &weight_avg[1],

  3085.                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));

  3086.     }else if(IS_8X16(mb_type)){

  3087.         mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,

  3088.                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],

  3089.                 &weight_op[2], &weight_avg[2],

  3090.                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));

  3091.         mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,

  3092.                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],

  3093.                 &weight_op[2], &weight_avg[2],

  3094.                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));

  3095.     }else{

  3096.         int i;

  3097. 

  3098.         assert(IS_8X8(mb_type));

  3099. 

  3100.         for(i=0; i<4; i++){

  3101.             const int sub_mb_type= h->sub_mb_type[i];

  3102.             const int n= 4*i;

  3103.             int x_offset= (i&1)<<2;

  3104.             int y_offset= (i&2)<<1;

  3105. 

  3106.             if(IS_SUB_8X8(sub_mb_type)){

  3107.                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,

  3108.                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],

  3109.                     &weight_op[3], &weight_avg[3],

  3110.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3111.             }else if(IS_SUB_8X4(sub_mb_type)){

  3112.                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,

  3113.                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],

  3114.                     &weight_op[4], &weight_avg[4],

  3115.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3116.                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,

  3117.                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],

  3118.                     &weight_op[4], &weight_avg[4],

  3119.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3120.             }else if(IS_SUB_4X8(sub_mb_type)){

  3121.                 mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,

  3122.                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],

  3123.                     &weight_op[5], &weight_avg[5],

  3124.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3125.                 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,

  3126.                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],

  3127.                     &weight_op[5], &weight_avg[5],

  3128.                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3129.             }else{

  3130.                 int j;

  3131.                 assert(IS_SUB_4X4(sub_mb_type));

  3132.                 for(j=0; j<4; j++){

  3133.                     int sub_x_offset= x_offset + 2*(j&1);

  3134.                     int sub_y_offset= y_offset +   (j&2);

  3135.                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,

  3136.                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],

  3137.                         &weight_op[6], &weight_avg[6],

  3138.                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));

  3139.                 }

  3140.             }

  3141.         }

  3142.     }

  3143. 

  3144.     prefetch_motion(h, 1);

  3145. }

  3146. 

  3147. static void decode_init_vlc(H264Context *h){

  3148.     static int done = 0;

  3149. 

  3150.     if (!done) {

  3151.         int i;

  3152.         done = 1;

  3153. 

  3154.         init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,

  3155.                  &chroma_dc_coeff_token_len [0], 1, 1,

  3156.                  &chroma_dc_coeff_token_bits[0], 1, 1, 1);

  3157. 

  3158.         for(i=0; i<4; i++){

  3159.             init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,

  3160.                      &coeff_token_len [i][0], 1, 1,

  3161.                      &coeff_token_bits[i][0], 1, 1, 1);

  3162.         }

  3163. 

  3164.         for(i=0; i<3; i++){

  3165.             init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,

  3166.                      &chroma_dc_total_zeros_len [i][0], 1, 1,

  3167.                      &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);

  3168.         }

  3169.         for(i=0; i<15; i++){

  3170.             init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,

  3171.                      &total_zeros_len [i][0], 1, 1,

  3172.                      &total_zeros_bits[i][0], 1, 1, 1);

  3173.         }

  3174. 

  3175.         for(i=0; i<6; i++){

  3176.             init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,

  3177.                      &run_len [i][0], 1, 1,

  3178.                      &run_bits[i][0], 1, 1, 1);

  3179.         }

  3180.         init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,

  3181.                  &run_len [6][0], 1, 1,

  3182.                  &run_bits[6][0], 1, 1, 1);

  3183.     }

  3184. }

  3185. 

  3186. /**

  3187.  * Sets the intra prediction function pointers.

  3188.  */

  3189. static void init_pred_ptrs(H264Context *h){

  3190. //    MpegEncContext * const s = &h->s;

  3191. 

  3192.     h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;

  3193.     h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;

  3194.     h->pred4x4[DC_PRED             ]= pred4x4_dc_c;

  3195.     h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;

  3196.     h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;

  3197.     h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;

  3198.     h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;

  3199.     h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;

  3200.     h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;

  3201.     h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;

  3202.     h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;

  3203.     h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;

  3204. 

  3205.     h->pred8x8l[VERT_PRED           ]= pred8x8l_vertical_c;

  3206.     h->pred8x8l[HOR_PRED            ]= pred8x8l_horizontal_c;

  3207.     h->pred8x8l[DC_PRED             ]= pred8x8l_dc_c;

  3208.     h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;

  3209.     h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;

  3210.     h->pred8x8l[VERT_RIGHT_PRED     ]= pred8x8l_vertical_right_c;

  3211.     h->pred8x8l[HOR_DOWN_PRED       ]= pred8x8l_horizontal_down_c;

  3212.     h->pred8x8l[VERT_LEFT_PRED      ]= pred8x8l_vertical_left_c;

  3213.     h->pred8x8l[HOR_UP_PRED         ]= pred8x8l_horizontal_up_c;

  3214.     h->pred8x8l[LEFT_DC_PRED        ]= pred8x8l_left_dc_c;

  3215.     h->pred8x8l[TOP_DC_PRED         ]= pred8x8l_top_dc_c;

  3216.     h->pred8x8l[DC_128_PRED         ]= pred8x8l_128_dc_c;

  3217. 

  3218.     h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;

  3219.     h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;

  3220.     h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;

  3221.     h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;

  3222.     h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;

  3223.     h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;

  3224.     h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;

  3225. 

  3226.     h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;

  3227.     h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;

  3228.     h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;

  3229.     h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;

  3230.     h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;

  3231.     h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;

  3232.     h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;

  3233. }

  3234. 

  3235. static void free_tables(H264Context *h){

  3236.     av_freep(&h->intra4x4_pred_mode);

  3237.     av_freep(&h->chroma_pred_mode_table);

  3238.     av_freep(&h->cbp_table);

  3239.     av_freep(&h->mvd_table[0]);

  3240.     av_freep(&h->mvd_table[1]);

  3241.     av_freep(&h->direct_table);

  3242.     av_freep(&h->non_zero_count);

  3243.     av_freep(&h->slice_table_base);

  3244.     av_freep(&h->top_borders[1]);

  3245.     av_freep(&h->top_borders[0]);

  3246.     h->slice_table= NULL;

  3247. 

  3248.     av_freep(&h->mb2b_xy);

  3249.     av_freep(&h->mb2b8_xy);

  3250. 

  3251.     av_freep(&h->s.obmc_scratchpad);

  3252. }

  3253. 

  3254. static void init_dequant8_coeff_table(H264Context *h){

  3255.     int i,q,x;

  3256.     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly

  3257.     h->dequant8_coeff[0] = h->dequant8_buffer[0];

  3258.     h->dequant8_coeff[1] = h->dequant8_buffer[1];

  3259. 

  3260.     for(i=0; i<2; i++ ){

  3261.         if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){

  3262.             h->dequant8_coeff[1] = h->dequant8_buffer[0];

  3263.             break;

  3264.         }

  3265. 

  3266.         for(q=0; q<52; q++){

  3267.             int shift = div6[q];

  3268.             int idx = rem6[q];

  3269.             for(x=0; x<64; x++)

  3270.                 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =

  3271.                     ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *

  3272.                     h->pps.scaling_matrix8[i][x]) << shift;

  3273.         }

  3274.     }

  3275. }

  3276. 

  3277. static void init_dequant4_coeff_table(H264Context *h){

  3278.     int i,j,q,x;

  3279.     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly

  3280.     for(i=0; i<6; i++ ){

  3281.         h->dequant4_coeff[i] = h->dequant4_buffer[i];

  3282.         for(j=0; j<i; j++){

  3283.             if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){

  3284.                 h->dequant4_coeff[i] = h->dequant4_buffer[j];

  3285.                 break;

  3286.             }

  3287.         }

  3288.         if(j<i)

  3289.             continue;

  3290. 

  3291.         for(q=0; q<52; q++){

  3292.             int shift = div6[q] + 2;

  3293.             int idx = rem6[q];

  3294.             for(x=0; x<16; x++)

  3295.                 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =

  3296.                     ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *

  3297.                     h->pps.scaling_matrix4[i][x]) << shift;

  3298.         }

  3299.     }

  3300. }

  3301. 

  3302. static void init_dequant_tables(H264Context *h){

  3303.     int i,x;

  3304.     init_dequant4_coeff_table(h);

  3305.     if(h->pps.transform_8x8_mode)

  3306.         init_dequant8_coeff_table(h);

  3307.     if(h->sps.transform_bypass){

  3308.         for(i=0; i<6; i++)

  3309.             for(x=0; x<16; x++)

  3310.                 h->dequant4_coeff[i][0][x] = 1<<6;

  3311.         if(h->pps.transform_8x8_mode)

  3312.             for(i=0; i<2; i++)

  3313.                 for(x=0; x<64; x++)

  3314.                     h->dequant8_coeff[i][0][x] = 1<<6;

  3315.     }

  3316. }

  3317. 

  3318. 

  3319. /**

  3320.  * allocates tables.

  3321.  * needs width/height

  3322.  */

  3323. static int alloc_tables(H264Context *h){

  3324.     MpegEncContext * const s = &h->s;

  3325.     const int big_mb_num= s->mb_stride * (s->mb_height+1);

  3326.     int x,y;

  3327. 

  3328.     CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))

  3329. 

  3330.     CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))

  3331.     CHECKED_ALLOCZ(h->slice_table_base  , (big_mb_num+s->mb_stride) * sizeof(uint8_t))

  3332.     CHECKED_ALLOCZ(h->top_borders[0]    , s->mb_width * (16+8+8) * sizeof(uint8_t))

  3333.     CHECKED_ALLOCZ(h->top_borders[1]    , s->mb_width * (16+8+8) * sizeof(uint8_t))

  3334.     CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))

  3335. 

  3336.     if( h->pps.cabac ) {

  3337.         CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))

  3338.         CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));

  3339.         CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));

  3340.         CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));

  3341.     }

  3342. 

  3343.     memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride)  * sizeof(uint8_t));

  3344.     h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;

  3345. 

  3346.     CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint32_t));

  3347.     CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));

  3348.     for(y=0; y<s->mb_height; y++){

  3349.         for(x=0; x<s->mb_width; x++){

  3350.             const int mb_xy= x + y*s->mb_stride;

  3351.             const int b_xy = 4*x + 4*y*h->b_stride;

  3352.             const int b8_xy= 2*x + 2*y*h->b8_stride;

  3353. 

  3354.             h->mb2b_xy [mb_xy]= b_xy;

  3355.             h->mb2b8_xy[mb_xy]= b8_xy;

  3356.         }

  3357.     }

  3358. 

  3359.     s->obmc_scratchpad = NULL;

  3360. 

  3361.     if(!h->dequant4_coeff[0])

  3362.         init_dequant_tables(h);

  3363. 

  3364.     return 0;

  3365. fail:

  3366.     free_tables(h);

  3367.     return -1;

  3368. }

  3369. 

  3370. static void common_init(H264Context *h){

  3371.     MpegEncContext * const s = &h->s;

  3372. 

  3373.     s->width = s->avctx->width;

  3374.     s->height = s->avctx->height;

  3375.     s->codec_id= s->avctx->codec->id;

  3376. 

  3377.     init_pred_ptrs(h);

  3378. 

  3379.     h->dequant_coeff_pps= -1;

  3380.     s->unrestricted_mv=1;

  3381.     s->decode=1; //FIXME

  3382. 

  3383.     memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));

  3384.     memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));

  3385. }

  3386. 

  3387. static int decode_init(AVCodecContext *avctx){

  3388.     H264Context *h= avctx->priv_data;

  3389.     MpegEncContext * const s = &h->s;

  3390. 

  3391.     MPV_decode_defaults(s);

  3392. 

  3393.     s->avctx = avctx;

  3394.     common_init(h);

  3395. 

  3396.     s->out_format = FMT_H264;

  3397.     s->workaround_bugs= avctx->workaround_bugs;

  3398. 

  3399.     // set defaults

  3400. //    s->decode_mb= ff_h263_decode_mb;

  3401.     s->low_delay= 1;

  3402.     avctx->pix_fmt= PIX_FMT_YUV420P;

  3403. 

  3404.     decode_init_vlc(h);

  3405. 

  3406.     if(avctx->extradata_size > 0 && avctx->extradata &&

  3407.        *(char *)avctx->extradata == 1){

  3408.         h->is_avc = 1;

  3409.         h->got_avcC = 0;

  3410.     } else {

  3411.         h->is_avc = 0;

  3412.     }

  3413. 

  3414.     return 0;

  3415. }

  3416. 

  3417. static int frame_start(H264Context *h){

  3418.     MpegEncContext * const s = &h->s;

  3419.     int i;

  3420. 

  3421.     if(MPV_frame_start(s, s->avctx) < 0)

  3422.         return -1;

  3423.     ff_er_frame_start(s);

  3424. 

  3425.     assert(s->linesize && s->uvlinesize);

  3426. 

  3427.     for(i=0; i<16; i++){

  3428.         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);

  3429.         h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);

  3430.     }

  3431.     for(i=0; i<4; i++){

  3432.         h->block_offset[16+i]=

  3433.         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);

  3434.         h->block_offset[24+16+i]=

  3435.         h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);

  3436.     }

  3437. 

  3438.     /* can't be in alloc_tables because linesize isn't known there.

  3439.      * FIXME: redo bipred weight to not require extra buffer? */

  3440.     if(!s->obmc_scratchpad)

  3441.         s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);

  3442. 

  3443.     /* some macroblocks will be accessed before they're available */

  3444.     if(FRAME_MBAFF)

  3445.         memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));

  3446. 

  3447. //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;

  3448.     return 0;

  3449. }

  3450. 

  3451. static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){

  3452.     MpegEncContext * const s = &h->s;

  3453.     int i;

  3454. 

  3455.     src_y  -=   linesize;

  3456.     src_cb -= uvlinesize;

  3457.     src_cr -= uvlinesize;

  3458. 

  3459.     // There are two lines saved, the line above the the top macroblock of a pair,

  3460.     // and the line above the bottom macroblock

  3461.     h->left_border[0]= h->top_borders[0][s->mb_x][15];

  3462.     for(i=1; i<17; i++){

  3463.         h->left_border[i]= src_y[15+i*  linesize];

  3464.     }

  3465. 

  3466.     *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);

  3467.     *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);

  3468. 

  3469.     if(!(s->flags&CODEC_FLAG_GRAY)){

  3470.         h->left_border[17  ]= h->top_borders[0][s->mb_x][16+7];

  3471.         h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];

  3472.         for(i=1; i<9; i++){

  3473.             h->left_border[i+17  ]= src_cb[7+i*uvlinesize];

  3474.             h->left_border[i+17+9]= src_cr[7+i*uvlinesize];

  3475.         }

  3476.         *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);

  3477.         *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);

  3478.     }

  3479. }

  3480. 

  3481. 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){

  3482.     MpegEncContext * const s = &h->s;

  3483.     int temp8, i;

  3484.     uint64_t temp64;

  3485.     int deblock_left = (s->mb_x > 0);

  3486.     int deblock_top  = (s->mb_y > 0);

  3487. 

  3488.     src_y  -=   linesize + 1;

  3489.     src_cb -= uvlinesize + 1;

  3490.     src_cr -= 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; i<17; 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.         if(s->mb_x+1 < s->mb_width){

  3508.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);

  3509.         }

  3510.     }

  3511. 

  3512.     if(!(s->flags&CODEC_FLAG_GRAY)){

  3513.         if(deblock_left){

  3514.             for(i = !deblock_top; i<9; i++){

  3515.                 XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);

  3516.                 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);

  3517.             }

  3518.         }

  3519.         if(deblock_top){

  3520.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);

  3521.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);

  3522.         }

  3523.     }

  3524. }

  3525. 

  3526. static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){

  3527.     MpegEncContext * const s = &h->s;

  3528.     int i;

  3529. 

  3530.     src_y  -= 2 *   linesize;

  3531.     src_cb -= 2 * uvlinesize;

  3532.     src_cr -= 2 * uvlinesize;

  3533. 

  3534.     // There are two lines saved, the line above the the top macroblock of a pair,

  3535.     // and the line above the bottom macroblock

  3536.     h->left_border[0]= h->top_borders[0][s->mb_x][15];

  3537.     h->left_border[1]= h->top_borders[1][s->mb_x][15];

  3538.     for(i=2; i<34; i++){

  3539.         h->left_border[i]= src_y[15+i*  linesize];

  3540.     }

  3541. 

  3542.     *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y +  32*linesize);

  3543.     *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);

  3544.     *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y +  33*linesize);

  3545.     *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);

  3546. 

  3547.     if(!(s->flags&CODEC_FLAG_GRAY)){

  3548.         h->left_border[34     ]= h->top_borders[0][s->mb_x][16+7];

  3549.         h->left_border[34+   1]= h->top_borders[1][s->mb_x][16+7];

  3550.         h->left_border[34+18  ]= h->top_borders[0][s->mb_x][24+7];

  3551.         h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];

  3552.         for(i=2; i<18; i++){

  3553.             h->left_border[i+34   ]= src_cb[7+i*uvlinesize];

  3554.             h->left_border[i+34+18]= src_cr[7+i*uvlinesize];

  3555.         }

  3556.         *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);

  3557.         *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);

  3558.         *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);

  3559.         *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);

  3560.     }

  3561. }

  3562. 

  3563. 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){

  3564.     MpegEncContext * const s = &h->s;

  3565.     int temp8, i;

  3566.     uint64_t temp64;

  3567.     int deblock_left = (s->mb_x > 0);

  3568.     int deblock_top  = (s->mb_y > 1);

  3569. 

  3570.     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);

  3571. 

  3572.     src_y  -= 2 *   linesize + 1;

  3573.     src_cb -= 2 * uvlinesize + 1;

  3574.     src_cr -= 2 * uvlinesize + 1;

  3575. 

  3576. #define XCHG(a,b,t,xchg)\

  3577. t= a;\

  3578. if(xchg)\

  3579.     a= b;\

  3580. b= t;

  3581. 

  3582.     if(deblock_left){

  3583.         for(i = (!deblock_top)<<1; i<34; i++){

  3584.             XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);

  3585.         }

  3586.     }

  3587. 

  3588.     if(deblock_top){

  3589.         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);

  3590.         XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);

  3591.         XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);

  3592.         XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);

  3593.         if(s->mb_x+1 < s->mb_width){

  3594.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);

  3595.             XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x+1]), *(uint64_t*)(src_y +17 +linesize), temp64, 1);

  3596.         }

  3597.     }

  3598. 

  3599.     if(!(s->flags&CODEC_FLAG_GRAY)){

  3600.         if(deblock_left){

  3601.             for(i = (!deblock_top) << 1; i<18; i++){

  3602.                 XCHG(h->left_border[i+34   ], src_cb[i*uvlinesize], temp8, xchg);

  3603.                 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);

  3604.             }

  3605.         }

  3606.         if(deblock_top){

  3607.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);

  3608.             XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);

  3609.             XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);

  3610.             XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);

  3611.         }

  3612.     }

  3613. }

  3614. 

  3615. static void hl_decode_mb(H264Context *h){

  3616.     MpegEncContext * const s = &h->s;

  3617.     const int mb_x= s->mb_x;

  3618.     const int mb_y= s->mb_y;

  3619.     const int mb_xy= mb_x + mb_y*s->mb_stride;

  3620.     const int mb_type= s->current_picture.mb_type[mb_xy];

  3621.     uint8_t  *dest_y, *dest_cb, *dest_cr;

  3622.     int linesize, uvlinesize /*dct_offset*/;

  3623.     int i;

  3624.     int *block_offset = &h->block_offset[0];

  3625.     const unsigned int bottom = mb_y & 1;

  3626.     const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);

  3627.     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);

  3628.     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);

  3629. 

  3630.     if(!s->decode)

  3631.         return;

  3632. 

  3633.     dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;

  3634.     dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;

  3635.     dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;

  3636. 

  3637.     if (MB_FIELD) {

  3638.         linesize   = h->mb_linesize   = s->linesize * 2;

  3639.         uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;

  3640.         block_offset = &h->block_offset[24];

  3641.         if(mb_y&1){ //FIXME move out of this func?

  3642.             dest_y -= s->linesize*15;

  3643.             dest_cb-= s->uvlinesize*7;

  3644.             dest_cr-= s->uvlinesize*7;

  3645.         }

  3646.         if(FRAME_MBAFF) {

  3647.             int list;

  3648.             for(list=0; list<2; list++){

  3649.                 if(!USES_LIST(mb_type, list))

  3650.                     continue;

  3651.                 if(IS_16X16(mb_type)){

  3652.                     int8_t *ref = &h->ref_cache[list][scan8[0]];

  3653.                     fill_rectangle(ref, 4, 4, 8, 16+*ref^(s->mb_y&1), 1);

  3654.                 }else{

  3655.                     for(i=0; i<16; i+=4){

  3656.                         //FIXME can refs be smaller than 8x8 when !direct_8x8_inference ?

  3657.                         int ref = h->ref_cache[list][scan8[i]];

  3658.                         if(ref >= 0)

  3659.                             fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, 16+ref^(s->mb_y&1), 1);

  3660.                     }

  3661.                 }

  3662.             }

  3663.         }

  3664.     } else {

  3665.         linesize   = h->mb_linesize   = s->linesize;

  3666.         uvlinesize = h->mb_uvlinesize = s->uvlinesize;

  3667. //        dct_offset = s->linesize * 16;

  3668.     }

  3669. 

  3670.     if(transform_bypass){

  3671.         idct_dc_add =

  3672.         idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;

  3673.     }else if(IS_8x8DCT(mb_type)){

  3674.         idct_dc_add = s->dsp.h264_idct8_dc_add;

  3675.         idct_add = s->dsp.h264_idct8_add;

  3676.     }else{

  3677.         idct_dc_add = s->dsp.h264_idct_dc_add;

  3678.         idct_add = s->dsp.h264_idct_add;

  3679.     }

  3680. 

  3681.     if(FRAME_MBAFF && h->deblocking_filter && IS_INTRA(mb_type)

  3682.        && (!bottom || !IS_INTRA(s->current_picture.mb_type[mb_xy-s->mb_stride]))){

  3683.         int mbt_y = mb_y&~1;

  3684.         uint8_t *top_y  = s->current_picture.data[0] + (mbt_y * 16* s->linesize  ) + mb_x * 16;

  3685.         uint8_t *top_cb = s->current_picture.data[1] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;

  3686.         uint8_t *top_cr = s->current_picture.data[2] + (mbt_y * 8 * s->uvlinesize) + mb_x * 8;

  3687.         xchg_pair_border(h, top_y, top_cb, top_cr, s->linesize, s->uvlinesize, 1);

  3688.     }

  3689. 

  3690.     if (IS_INTRA_PCM(mb_type)) {

  3691.         unsigned int x, y;

  3692. 

  3693.         // The pixels are stored in h->mb array in the same order as levels,

  3694.         // copy them in output in the correct order.

  3695.         for(i=0; i<16; i++) {

  3696.             for (y=0; y<4; y++) {

  3697.                 for (x=0; x<4; x++) {

  3698.                     *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];

  3699.                 }

  3700.             }

  3701.         }

  3702.         for(i=16; i<16+4; i++) {

  3703.             for (y=0; y<4; y++) {

  3704.                 for (x=0; x<4; x++) {

  3705.                     *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];

  3706.                 }

  3707.             }

  3708.         }

  3709.         for(i=20; i<20+4; i++) {

  3710.             for (y=0; y<4; y++) {

  3711.                 for (x=0; x<4; x++) {

  3712.                     *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];

  3713.                 }

  3714.             }

  3715.         }

  3716.     } else {

  3717.         if(IS_INTRA(mb_type)){

  3718.             if(h->deblocking_filter && !FRAME_MBAFF)

  3719.                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);

  3720. 

  3721.             if(!(s->flags&CODEC_FLAG_GRAY)){

  3722.                 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);

  3723.                 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);

  3724.             }

  3725. 

  3726.             if(IS_INTRA4x4(mb_type)){

  3727.                 if(!s->encoding){

  3728.                     if(IS_8x8DCT(mb_type)){

  3729.                         for(i=0; i<16; i+=4){

  3730.                             uint8_t * const ptr= dest_y + block_offset[i];

  3731.                             const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];

  3732.                             const int nnz = h->non_zero_count_cache[ scan8[i] ];

  3733.                             h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,

  3734.                                                    (h->topright_samples_available<<(i+1))&0x8000, linesize);

  3735.                             if(nnz){

  3736.                                 if(nnz == 1 && h->mb[i*16])

  3737.                                     idct_dc_add(ptr, h->mb + i*16, linesize);

  3738.                                 else

  3739.                                     idct_add(ptr, h->mb + i*16, linesize);

  3740.                             }

  3741.                         }

  3742.                     }else

  3743.                     for(i=0; i<16; i++){

  3744.                         uint8_t * const ptr= dest_y + block_offset[i];

  3745.                         uint8_t *topright;

  3746.                         const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];

  3747.                         int nnz, tr;

  3748. 

  3749.                         if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){

  3750.                             const int topright_avail= (h->topright_samples_available<<i)&0x8000;

  3751.                             assert(mb_y || linesize <= block_offset[i]);

  3752.                             if(!topright_avail){

  3753.                                 tr= ptr[3 - linesize]*0x01010101;

  3754.                                 topright= (uint8_t*) &tr;

  3755.                             }else

  3756.                                 topright= ptr + 4 - linesize;

  3757.                         }else

  3758.                             topright= NULL;

  3759. 

  3760.                         h->pred4x4[ dir ](ptr, topright, linesize);

  3761.                         nnz = h->non_zero_count_cache[ scan8[i] ];

  3762.                         if(nnz){

  3763.                             if(s->codec_id == CODEC_ID_H264){

  3764.                                 if(nnz == 1 && h->mb[i*16])

  3765.                                     idct_dc_add(ptr, h->mb + i*16, linesize);

  3766.                                 else

  3767.                                     idct_add(ptr, h->mb + i*16, linesize);

  3768.                             }else

  3769.                                 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);

  3770.                         }

  3771.                     }

  3772.                 }

  3773.             }else{

  3774.                 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);

  3775.                 if(s->codec_id == CODEC_ID_H264){

  3776.                     if(!transform_bypass)

  3777.                         h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);

  3778.                 }else

  3779.                     svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);

  3780.             }

  3781.             if(h->deblocking_filter && !FRAME_MBAFF)

  3782.                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);

  3783.         }else if(s->codec_id == CODEC_ID_H264){

  3784.             hl_motion(h, dest_y, dest_cb, dest_cr,

  3785.                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,

  3786.                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,

  3787.                       s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);

  3788.         }

  3789. 

  3790. 

  3791.         if(!IS_INTRA4x4(mb_type)){

  3792.             if(s->codec_id == CODEC_ID_H264){

  3793.                 if(IS_INTRA16x16(mb_type)){

  3794.                     for(i=0; i<16; i++){

  3795.                         if(h->non_zero_count_cache[ scan8[i] ])

  3796.                             idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);

  3797.                         else if(h->mb[i*16])

  3798.                             idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);

  3799.                     }

  3800.                 }else{

  3801.                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;

  3802.                     for(i=0; i<16; i+=di){

  3803.                         int nnz = h->non_zero_count_cache[ scan8[i] ];

  3804.                         if(nnz){

  3805.                             if(nnz==1 && h->mb[i*16])

  3806.                                 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);

  3807.                             else

  3808.                                 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);

  3809.                         }

  3810.                     }

  3811.                 }

  3812.             }else{

  3813.                 for(i=0; i<16; i++){

  3814.                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below

  3815.                         uint8_t * const ptr= dest_y + block_offset[i];

  3816.                         svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);

  3817.                     }

  3818.                 }

  3819.             }

  3820.         }

  3821. 

  3822.         if(!(s->flags&CODEC_FLAG_GRAY)){

  3823.             uint8_t *dest[2] = {dest_cb, dest_cr};

  3824.             if(transform_bypass){

  3825.                 idct_add = idct_dc_add = s->dsp.add_pixels4;

  3826.             }else{

  3827.                 idct_add = s->dsp.h264_idct_add;

  3828.                 idct_dc_add = s->dsp.h264_idct_dc_add;

  3829.                 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]);

  3830.                 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]);

  3831.             }

  3832.             if(s->codec_id == CODEC_ID_H264){

  3833.                 for(i=16; i<16+8; i++){

  3834.                     if(h->non_zero_count_cache[ scan8[i] ])

  3835.                         idct_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);

  3836.                     else if(h->mb[i*16])

  3837.                         idct_dc_add(dest[(i&4)>>2] + block_offset[i], h->mb + i*16, uvlinesize);

  3838.                 }

  3839.             }else{

  3840.                 for(i=16; i<16+8; i++){

  3841.                     if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){

  3842.                         uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];

  3843.                         svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);

  3844.                     }

  3845.                 }

  3846.             }

  3847.         }

  3848.     }

  3849.     if(h->deblocking_filter) {

  3850.         if (FRAME_MBAFF) {

  3851.             //FIXME try deblocking one mb at a time?

  3852.             // the reduction in load/storing mvs and such might outweigh the extra backup/xchg_border

  3853.             const int mb_y = s->mb_y - 1;

  3854.             uint8_t  *pair_dest_y, *pair_dest_cb, *pair_dest_cr;

  3855.             const int mb_xy= mb_x + mb_y*s->mb_stride;

  3856.             const int mb_type_top   = s->current_picture.mb_type[mb_xy];

  3857.             const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];

  3858.             if (!bottom) return;

  3859.             pair_dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;

  3860.             pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;

  3861.             pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;

  3862. 

  3863.             if(IS_INTRA(mb_type_top | mb_type_bottom))

  3864.                 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);

  3865. 

  3866.             backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);

  3867.             // deblock a pair

  3868.             // top

  3869.             s->mb_y--;

  3870.             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);

  3871.             fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb

  3872.             h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy]);

  3873.             filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);

  3874.             // bottom

  3875.             s->mb_y++;

  3876.             tprintf("call mbaff filter_mb\n");

  3877.             fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb

  3878.             h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy+s->mb_stride]);

  3879.             filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);

  3880.         } else {

  3881.             tprintf("call filter_mb\n");

  3882.             backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);

  3883.             fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb

  3884.             filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);

  3885.         }

  3886.     }

  3887. }

  3888. 

  3889. /**

  3890.  * fills the default_ref_list.

  3891.  */

  3892. static int fill_default_ref_list(H264Context *h){

  3893.     MpegEncContext * const s = &h->s;

  3894.     int i;

  3895.     int smallest_poc_greater_than_current = -1;

  3896.     Picture sorted_short_ref[32];

  3897. 

  3898.     if(h->slice_type==B_TYPE){

  3899.         int out_i;

  3900.         int limit= INT_MIN;

  3901. 

  3902.         /* sort frame according to poc in B slice */

  3903.         for(out_i=0; out_i<h->short_ref_count; out_i++){

  3904.             int best_i=INT_MIN;

  3905.             int best_poc=INT_MAX;

  3906. 

  3907.             for(i=0; i<h->short_ref_count; i++){

  3908.                 const int poc= h->short_ref[i]->poc;

  3909.                 if(poc > limit && poc < best_poc){

  3910.                     best_poc= poc;

  3911.                     best_i= i;

  3912.                 }

  3913.             }

  3914. 

  3915.             assert(best_i != INT_MIN);

  3916. 

  3917.             limit= best_poc;

  3918.             sorted_short_ref[out_i]= *h->short_ref[best_i];

  3919.             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);

  3920.             if (-1 == smallest_poc_greater_than_current) {

  3921.                 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {

  3922.                     smallest_poc_greater_than_current = out_i;

  3923.                 }

  3924.             }

  3925.         }

  3926.     }

  3927. 

  3928.     if(s->picture_structure == PICT_FRAME){

  3929.         if(h->slice_type==B_TYPE){

  3930.             int list;

  3931.             tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);

  3932. 

  3933.             // find the largest poc

  3934.             for(list=0; list<2; list++){

  3935.                 int index = 0;

  3936.                 int j= -99;

  3937.                 int step= list ? -1 : 1;

  3938. 

  3939.                 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {

  3940.                     while(j<0 || j>= h->short_ref_count){

  3941.                         if(j != -99 && step == (list ? -1 : 1))

  3942.                             return -1;

  3943.                         step = -step;

  3944.                         j= smallest_poc_greater_than_current + (step>>1);

  3945.                     }

  3946.                     if(sorted_short_ref[j].reference != 3) continue;

  3947.                     h->default_ref_list[list][index  ]= sorted_short_ref[j];

  3948.                     h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;

  3949.                 }

  3950. 

  3951.                 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){

  3952.                     if(h->long_ref[i] == NULL) continue;

  3953.                     if(h->long_ref[i]->reference != 3) continue;

  3954. 

  3955.                     h->default_ref_list[ list ][index  ]= *h->long_ref[i];

  3956.                     h->default_ref_list[ list ][index++].pic_id= i;;

  3957.                 }

  3958. 

  3959.                 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){

  3960.                     // swap the two first elements of L1 when

  3961.                     // L0 and L1 are identical

  3962.                     Picture temp= h->default_ref_list[1][0];

  3963.                     h->default_ref_list[1][0] = h->default_ref_list[1][1];

  3964.                     h->default_ref_list[1][1] = temp;

  3965.                 }

  3966. 

  3967.                 if(index < h->ref_count[ list ])

  3968.                     memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));

  3969.             }

  3970.         }else{

  3971.             int index=0;

  3972.             for(i=0; i<h->short_ref_count; i++){

  3973.                 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit

  3974.                 h->default_ref_list[0][index  ]= *h->short_ref[i];

  3975.                 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;

  3976.             }

  3977.             for(i = 0; i < 16; i++){

  3978.                 if(h->long_ref[i] == NULL) continue;

  3979.                 if(h->long_ref[i]->reference != 3) continue;

  3980.                 h->default_ref_list[0][index  ]= *h->long_ref[i];

  3981.                 h->default_ref_list[0][index++].pic_id= i;;

  3982.             }

  3983.             if(index < h->ref_count[0])

  3984.                 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));

  3985.         }

  3986.     }else{ //FIELD

  3987.         if(h->slice_type==B_TYPE){

  3988.         }else{

  3989.             //FIXME second field balh

  3990.         }

  3991.     }

  3992. #ifdef TRACE

  3993.     for (i=0; i<h->ref_count[0]; i++) {

  3994.         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]);

  3995.     }

  3996.     if(h->slice_type==B_TYPE){

  3997.         for (i=0; i<h->ref_count[1]; i++) {

  3998.             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]);

  3999.         }

  4000.     }

  4001. #endif

  4002.     return 0;

  4003. }

  4004. 

  4005. static void print_short_term(H264Context *h);

  4006. static void print_long_term(H264Context *h);

  4007. 

  4008. static int decode_ref_pic_list_reordering(H264Context *h){

  4009.     MpegEncContext * const s = &h->s;

  4010.     int list, index;

  4011. 

  4012.     print_short_term(h);

  4013.     print_long_term(h);

  4014.     if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func

  4015. 

  4016.     for(list=0; list<2; list++){

  4017.         memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);

  4018. 

  4019.         if(get_bits1(&s->gb)){

  4020.             int pred= h->curr_pic_num;

  4021. 

  4022.             for(index=0; ; index++){

  4023.                 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);

  4024.                 int pic_id;

  4025.                 int i;

  4026.                 Picture *ref = NULL;

  4027. 

  4028.                 if(reordering_of_pic_nums_idc==3)

  4029.                     break;

  4030. 

  4031.                 if(index >= h->ref_count[list]){

  4032.                     av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");

  4033.                     return -1;

  4034.                 }

  4035. 

  4036.                 if(reordering_of_pic_nums_idc<3){

  4037.                     if(reordering_of_pic_nums_idc<2){

  4038.                         const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;

  4039. 

  4040.                         if(abs_diff_pic_num >= h->max_pic_num){

  4041.                             av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");

  4042.                             return -1;

  4043.                         }

  4044. 

  4045.                         if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;

  4046.                         else                                pred+= abs_diff_pic_num;

  4047.                         pred &= h->max_pic_num - 1;

  4048. 

  4049.                         for(i= h->short_ref_count-1; i>=0; i--){

  4050.                             ref = h->short_ref[i];

  4051.                             assert(ref->reference == 3);

  4052.                             assert(!ref->long_ref);

  4053.                             if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer

  4054.                                 break;

  4055.                         }

  4056.                         if(i>=0)

  4057.                             ref->pic_id= ref->frame_num;

  4058.                     }else{

  4059.                         pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx

  4060.                         ref = h->long_ref[pic_id];

  4061.                         ref->pic_id= pic_id;

  4062.                         assert(ref->reference == 3);

  4063.                         assert(ref->long_ref);

  4064.                         i=0;

  4065.                     }

  4066. 

  4067.                     if (i < 0) {

  4068.                         av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");

  4069.                         memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME

  4070.                     } else {

  4071.                         for(i=index; i+1<h->ref_count[list]; i++){

  4072.                             if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)

  4073.                                 break;

  4074.                         }

  4075.                         for(; i > index; i--){

  4076.                             h->ref_list[list][i]= h->ref_list[list][i-1];

  4077.                         }

  4078.                         h->ref_list[list][index]= *ref;

  4079.                     }

  4080.                 }else{

  4081.                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");

  4082.                     return -1;

  4083.                 }

  4084.             }

  4085.         }

  4086. 

  4087.         if(h->slice_type!=B_TYPE) break;

  4088.     }

  4089.     for(list=0; list<2; list++){

  4090.         for(index= 0; index < h->ref_count[list]; index++){

  4091.             if(!h->ref_list[list][index].data[0])

  4092.                 h->ref_list[list][index]= s->current_picture;

  4093.         }

  4094.         if(h->slice_type!=B_TYPE) break;

  4095.     }

  4096. 

  4097.     if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)

  4098.         direct_dist_scale_factor(h);

  4099.     direct_ref_list_init(h);

  4100.     return 0;

  4101. }

  4102. 

  4103. static void fill_mbaff_ref_list(H264Context *h){

  4104.     int list, i, j;

  4105.     for(list=0; list<2; list++){

  4106.         for(i=0; i<h->ref_count[list]; i++){

  4107.             Picture *frame = &h->ref_list[list][i];

  4108.             Picture *field = &h->ref_list[list][16+2*i];

  4109.             field[0] = *frame;

  4110.             for(j=0; j<3; j++)

  4111.                 field[0].linesize[j] <<= 1;

  4112.             field[1] = field[0];

  4113.             for(j=0; j<3; j++)

  4114.                 field[1].data[j] += frame->linesize[j];

  4115. 

  4116.             h->luma_weight[list][16+2*i] = h->luma_weight[list][16+2*i+1] = h->luma_weight[list][i];

  4117.             h->luma_offset[list][16+2*i] = h->luma_offset[list][16+2*i+1] = h->luma_offset[list][i];

  4118.             for(j=0; j<2; j++){

  4119.                 h->chroma_weight[list][16+2*i][j] = h->chroma_weight[list][16+2*i+1][j] = h->chroma_weight[list][i][j];

  4120.                 h->chroma_offset[list][16+2*i][j] = h->chroma_offset[list][16+2*i+1][j] = h->chroma_offset[list][i][j];

  4121.             }

  4122.         }

  4123.     }

  4124.     for(j=0; j<h->ref_count[1]; j++){

  4125.         for(i=0; i<h->ref_count[0]; i++)

  4126.             h->implicit_weight[j][16+2*i] = h->implicit_weight[j][16+2*i+1] = h->implicit_weight[j][i];

  4127.         memcpy(h->implicit_weight[16+2*j],   h->implicit_weight[j], sizeof(*h->implicit_weight));

  4128.         memcpy(h->implicit_weight[16+2*j+1], h->implicit_weight[j], sizeof(*h->implicit_weight));

  4129.     }

  4130. }

  4131. 

  4132. static int pred_weight_table(H264Context *h){

  4133.     MpegEncContext * const s = &h->s;

  4134.     int list, i;

  4135.     int luma_def, chroma_def;

  4136. 

  4137.     h->use_weight= 0;

  4138.     h->use_weight_chroma= 0;

  4139.     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);

  4140.     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);

  4141.     luma_def = 1<<h->luma_log2_weight_denom;

  4142.     chroma_def = 1<<h->chroma_log2_weight_denom;

  4143. 

  4144.     for(list=0; list<2; list++){

  4145.         for(i=0; i<h->ref_count[list]; i++){

  4146.             int luma_weight_flag, chroma_weight_flag;

  4147. 

  4148.             luma_weight_flag= get_bits1(&s->gb);

  4149.             if(luma_weight_flag){

  4150.                 h->luma_weight[list][i]= get_se_golomb(&s->gb);

  4151.                 h->luma_offset[list][i]= get_se_golomb(&s->gb);

  4152.                 if(   h->luma_weight[list][i] != luma_def

  4153.                    || h->luma_offset[list][i] != 0)

  4154.                     h->use_weight= 1;

  4155.             }else{

  4156.                 h->luma_weight[list][i]= luma_def;

  4157.                 h->luma_offset[list][i]= 0;

  4158.             }

  4159. 

  4160.             chroma_weight_flag= get_bits1(&s->gb);

  4161.             if(chroma_weight_flag){

  4162.                 int j;

  4163.                 for(j=0; j<2; j++){

  4164.                     h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);

  4165.                     h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);

  4166.                     if(   h->chroma_weight[list][i][j] != chroma_def

  4167.                        || h->chroma_offset[list][i][j] != 0)

  4168.                         h->use_weight_chroma= 1;

  4169.                 }

  4170.             }else{

  4171.                 int j;

  4172.                 for(j=0; j<2; j++){

  4173.                     h->chroma_weight[list][i][j]= chroma_def;

  4174.                     h->chroma_offset[list][i][j]= 0;

  4175.                 }

  4176.             }

  4177.         }

  4178.         if(h->slice_type != B_TYPE) break;

  4179.     }

  4180.     h->use_weight= h->use_weight || h->use_weight_chroma;

  4181.     return 0;

  4182. }

  4183. 

  4184. static void implicit_weight_table(H264Context *h){

  4185.     MpegEncContext * const s = &h->s;

  4186.     int ref0, ref1;

  4187.     int cur_poc = s->current_picture_ptr->poc;

  4188. 

  4189.     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1

  4190.        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){

  4191.         h->use_weight= 0;

  4192.         h->use_weight_chroma= 0;

  4193.         return;

  4194.     }

  4195. 

  4196.     h->use_weight= 2;

  4197.     h->use_weight_chroma= 2;

  4198.     h->luma_log2_weight_denom= 5;

  4199.     h->chroma_log2_weight_denom= 5;

  4200. 

  4201.     for(ref0=0; ref0 < h->ref_count[0]; ref0++){

  4202.         int poc0 = h->ref_list[0][ref0].poc;

  4203.         for(ref1=0; ref1 < h->ref_count[1]; ref1++){

  4204.             int poc1 = h->ref_list[1][ref1].poc;

  4205.             int td = clip(poc1 - poc0, -128, 127);

  4206.             if(td){

  4207.                 int tb = clip(cur_poc - poc0, -128, 127);

  4208.                 int tx = (16384 + (ABS(td) >> 1)) / td;

  4209.                 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;

  4210.                 if(dist_scale_factor < -64 || dist_scale_factor > 128)

  4211.                     h->implicit_weight[ref0][ref1] = 32;

  4212.                 else

  4213.                     h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;

  4214.             }else

  4215.                 h->implicit_weight[ref0][ref1] = 32;

  4216.         }

  4217.     }

  4218. }

  4219. 

  4220. static inline void unreference_pic(H264Context *h, Picture *pic){

  4221.     int i;

  4222.     pic->reference=0;

  4223.     if(pic == h->delayed_output_pic)

  4224.         pic->reference=1;

  4225.     else{

  4226.         for(i = 0; h->delayed_pic[i]; i++)

  4227.             if(pic == h->delayed_pic[i]){

  4228.                 pic->reference=1;

  4229.                 break;

  4230.             }

  4231.     }

  4232. }

  4233. 

  4234. /**

  4235.  * instantaneous decoder refresh.

  4236.  */

  4237. static void idr(H264Context *h){

  4238.     int i;

  4239. 

  4240.     for(i=0; i<16; i++){

  4241.         if (h->long_ref[i] != NULL) {

  4242.             unreference_pic(h, h->long_ref[i]);

  4243.             h->long_ref[i]= NULL;

  4244.         }

  4245.     }

  4246.     h->long_ref_count=0;

  4247. 

  4248.     for(i=0; i<h->short_ref_count; i++){

  4249.         unreference_pic(h, h->short_ref[i]);

  4250.         h->short_ref[i]= NULL;

  4251.     }

  4252.     h->short_ref_count=0;

  4253. }

  4254. 

  4255. /* forget old pics after a seek */

  4256. static void flush_dpb(AVCodecContext *avctx){

  4257.     H264Context *h= avctx->priv_data;

  4258.     int i;

  4259.     for(i=0; i<16; i++) {

  4260.         if(h->delayed_pic[i])

  4261.             h->delayed_pic[i]->reference= 0;

  4262.         h->delayed_pic[i]= NULL;

  4263.     }

  4264.     if(h->delayed_output_pic)

  4265.         h->delayed_output_pic->reference= 0;

  4266.     h->delayed_output_pic= NULL;

  4267.     idr(h);

  4268.     if(h->s.current_picture_ptr)

  4269.         h->s.current_picture_ptr->reference= 0;

  4270. }

  4271. 

  4272. /**

  4273.  *

  4274.  * @return the removed picture or NULL if an error occurs

  4275.  */

  4276. static Picture * remove_short(H264Context *h, int frame_num){

  4277.     MpegEncContext * const s = &h->s;

  4278.     int i;

  4279. 

  4280.     if(s->avctx->debug&FF_DEBUG_MMCO)

  4281.         av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);

  4282. 

  4283.     for(i=0; i<h->short_ref_count; i++){

  4284.         Picture *pic= h->short_ref[i];

  4285.         if(s->avctx->debug&FF_DEBUG_MMCO)

  4286.             av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);

  4287.         if(pic->frame_num == frame_num){

  4288.             h->short_ref[i]= NULL;

  4289.             memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));

  4290.             h->short_ref_count--;

  4291.             return pic;

  4292.         }

  4293.     }

  4294.     return NULL;

  4295. }

  4296. 

  4297. /**

  4298.  *

  4299.  * @return the removed picture or NULL if an error occurs

  4300.  */

  4301. static Picture * remove_long(H264Context *h, int i){

  4302.     Picture *pic;

  4303. 

  4304.     pic= h->long_ref[i];

  4305.     h->long_ref[i]= NULL;

  4306.     if(pic) h->long_ref_count--;

  4307. 

  4308.     return pic;

  4309. }

  4310. 

  4311. /**

  4312.  * print short term list

  4313.  */

  4314. static void print_short_term(H264Context *h) {

  4315.     uint32_t i;

  4316.     if(h->s.avctx->debug&FF_DEBUG_MMCO) {

  4317.         av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");

  4318.         for(i=0; i<h->short_ref_count; i++){

  4319.             Picture *pic= h->short_ref[i];

  4320.             av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);

  4321.         }

  4322.     }

  4323. }

  4324. 

  4325. /**

  4326.  * print long term list

  4327.  */

  4328. static void print_long_term(H264Context *h) {

  4329.     uint32_t i;

  4330.     if(h->s.avctx->debug&FF_DEBUG_MMCO) {

  4331.         av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");

  4332.         for(i = 0; i < 16; i++){

  4333.             Picture *pic= h->long_ref[i];

  4334.             if (pic) {

  4335.                 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);

  4336.             }

  4337.         }

  4338.     }

  4339. }

  4340. 

  4341. /**

  4342.  * Executes the reference picture marking (memory management control operations).

  4343.  */

  4344. static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){

  4345.     MpegEncContext * const s = &h->s;

  4346.     int i, j;

  4347.     int current_is_long=0;

  4348.     Picture *pic;

  4349. 

  4350.     if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)

  4351.         av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");

  4352. 

  4353.     for(i=0; i<mmco_count; i++){

  4354.         if(s->avctx->debug&FF_DEBUG_MMCO)

  4355.             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);

  4356. 

  4357.         switch(mmco[i].opcode){

  4358.         case MMCO_SHORT2UNUSED:

  4359.             pic= remove_short(h, mmco[i].short_frame_num);

  4360.             if(pic)

  4361.                 unreference_pic(h, pic);

  4362.             else if(s->avctx->debug&FF_DEBUG_MMCO)

  4363.                 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_short() failure\n");

  4364.             break;

  4365.         case MMCO_SHORT2LONG:

  4366.             pic= remove_long(h, mmco[i].long_index);

  4367.             if(pic) unreference_pic(h, pic);

  4368. 

  4369.             h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);

  4370.             h->long_ref[ mmco[i].long_index ]->long_ref=1;

  4371.             h->long_ref_count++;

  4372.             break;

  4373.         case MMCO_LONG2UNUSED:

  4374.             pic= remove_long(h, mmco[i].long_index);

  4375.             if(pic)

  4376.                 unreference_pic(h, pic);

  4377.             else if(s->avctx->debug&FF_DEBUG_MMCO)

  4378.                 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco: remove_long() failure\n");

  4379.             break;

  4380.         case MMCO_LONG:

  4381.             pic= remove_long(h, mmco[i].long_index);

  4382.             if(pic) unreference_pic(h, pic);

  4383. 

  4384.             h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;

  4385.             h->long_ref[ mmco[i].long_index ]->long_ref=1;

  4386.             h->long_ref_count++;

  4387. 

  4388.             current_is_long=1;

  4389.             break;

  4390.         case MMCO_SET_MAX_LONG:

  4391.             assert(mmco[i].long_index <= 16);

  4392.             // just remove the long term which index is greater than new max

  4393.             for(j = mmco[i].long_index; j<16; j++){

  4394.                 pic = remove_long(h, j);

  4395.                 if (pic) unreference_pic(h, pic);

  4396.             }

  4397.             break;

  4398.         case MMCO_RESET:

  4399.             while(h->short_ref_count){

  4400.                 pic= remove_short(h, h->short_ref[0]->frame_num);

  4401.                 unreference_pic(h, pic);

  4402.             }

  4403.             for(j = 0; j < 16; j++) {

  4404.                 pic= remove_long(h, j);

  4405.                 if(pic) unreference_pic(h, pic);

  4406.             }

  4407.             break;

  4408.         default: assert(0);

  4409.         }

  4410.     }

  4411. 

  4412.     if(!current_is_long){

  4413.         pic= remove_short(h, s->current_picture_ptr->frame_num);

  4414.         if(pic){

  4415.             unreference_pic(h, pic);

  4416.             av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");

  4417.         }

  4418. 

  4419.         if(h->short_ref_count)

  4420.             memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));

  4421. 

  4422.         h->short_ref[0]= s->current_picture_ptr;

  4423.         h->short_ref[0]->long_ref=0;

  4424.         h->short_ref_count++;

  4425.     }

  4426. 

  4427.     print_short_term(h);

  4428.     print_long_term(h);

  4429.     return 0;

  4430. }

  4431. 

  4432. static int decode_ref_pic_marking(H264Context *h){

  4433.     MpegEncContext * const s = &h->s;

  4434.     int i;

  4435. 

  4436.     if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields

  4437.         s->broken_link= get_bits1(&s->gb) -1;

  4438.         h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx

  4439.         if(h->mmco[0].long_index == -1)

  4440.             h->mmco_index= 0;

  4441.         else{

  4442.             h->mmco[0].opcode= MMCO_LONG;

  4443.             h->mmco_index= 1;

  4444.         }

  4445.     }else{

  4446.         if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag

  4447.             for(i= 0; i<MAX_MMCO_COUNT; i++) {

  4448.                 MMCOOpcode opcode= get_ue_golomb(&s->gb);;

  4449. 

  4450.                 h->mmco[i].opcode= opcode;

  4451.                 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){

  4452.                     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

  4453. /*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){

  4454.                         av_log(s->avctx, AV_LOG_ERROR, "illegal short ref in memory management control operation %d\n", mmco);

  4455.                         return -1;

  4456.                     }*/

  4457.                 }

  4458.                 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){

  4459.                     h->mmco[i].long_index= get_ue_golomb(&s->gb);

  4460.                     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){

  4461.                         av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);

  4462.                         return -1;

  4463.                     }

  4464.                 }

  4465. 

  4466.                 if(opcode > MMCO_LONG){

  4467.                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);

  4468.                     return -1;

  4469.                 }

  4470.                 if(opcode == MMCO_END)

  4471.                     break;

  4472.             }

  4473.             h->mmco_index= i;

  4474.         }else{

  4475.             assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);

  4476. 

  4477.             if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields

  4478.                 h->mmco[0].opcode= MMCO_SHORT2UNUSED;

  4479.                 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;

  4480.                 h->mmco_index= 1;

  4481.             }else

  4482.                 h->mmco_index= 0;

  4483.         }

  4484.     }

  4485. 

  4486.     return 0;

  4487. }

  4488. 

  4489. static int init_poc(H264Context *h){

  4490.     MpegEncContext * const s = &h->s;

  4491.     const int max_frame_num= 1<<h->sps.log2_max_frame_num;

  4492.     int field_poc[2];

  4493. 

  4494.     if(h->nal_unit_type == NAL_IDR_SLICE){

  4495.         h->frame_num_offset= 0;

  4496.     }else{

  4497.         if(h->frame_num < h->prev_frame_num)

  4498.             h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;

  4499.         else

  4500.             h->frame_num_offset= h->prev_frame_num_offset;

  4501.     }

  4502. 

  4503.     if(h->sps.poc_type==0){

  4504.         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;

  4505. 

  4506.         if(h->nal_unit_type == NAL_IDR_SLICE){

  4507.              h->prev_poc_msb=

  4508.              h->prev_poc_lsb= 0;

  4509.         }

  4510. 

  4511.         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)

  4512.             h->poc_msb = h->prev_poc_msb + max_poc_lsb;

  4513.         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)

  4514.             h->poc_msb = h->prev_poc_msb - max_poc_lsb;

  4515.         else

  4516.             h->poc_msb = h->prev_poc_msb;

  4517. //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);

  4518.         field_poc[0] =

  4519.         field_poc[1] = h->poc_msb + h->poc_lsb;

  4520.         if(s->picture_structure == PICT_FRAME)

  4521.             field_poc[1] += h->delta_poc_bottom;

  4522.     }else if(h->sps.poc_type==1){

  4523.         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;

  4524.         int i;

  4525. 

  4526.         if(h->sps.poc_cycle_length != 0)

  4527.             abs_frame_num = h->frame_num_offset + h->frame_num;

  4528.         else

  4529.             abs_frame_num = 0;

  4530. 

  4531.         if(h->nal_ref_idc==0 && abs_frame_num > 0)

  4532.             abs_frame_num--;

  4533. 

  4534.         expected_delta_per_poc_cycle = 0;

  4535.         for(i=0; i < h->sps.poc_cycle_length; i++)

  4536.             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse

  4537. 

  4538.         if(abs_frame_num > 0){

  4539.             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;

  4540.             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;

  4541. 

  4542.             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;

  4543.             for(i = 0; i <= frame_num_in_poc_cycle; i++)

  4544.                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];

  4545.         } else

  4546.             expectedpoc = 0;

  4547. 

  4548.         if(h->nal_ref_idc == 0)

  4549.             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;

  4550. 

  4551.         field_poc[0] = expectedpoc + h->delta_poc[0];

  4552.         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;

  4553. 

  4554.         if(s->picture_structure == PICT_FRAME)

  4555.             field_poc[1] += h->delta_poc[1];

  4556.     }else{

  4557.         int poc;

  4558.         if(h->nal_unit_type == NAL_IDR_SLICE){

  4559.             poc= 0;

  4560.         }else{

  4561.             if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);

  4562.             else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;

  4563.         }

  4564.         field_poc[0]= poc;

  4565.         field_poc[1]= poc;

  4566.     }

  4567. 

  4568.     if(s->picture_structure != PICT_BOTTOM_FIELD)

  4569.         s->current_picture_ptr->field_poc[0]= field_poc[0];

  4570.     if(s->picture_structure != PICT_TOP_FIELD)

  4571.         s->current_picture_ptr->field_poc[1]= field_poc[1];

  4572.     if(s->picture_structure == PICT_FRAME) // FIXME field pix?

  4573.         s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);

  4574. 

  4575.     return 0;

  4576. }

  4577. 

  4578. /**

  4579.  * decodes a slice header.

  4580.  * this will allso call MPV_common_init() and frame_start() as needed

  4581.  */

  4582. static int decode_slice_header(H264Context *h){

  4583.     MpegEncContext * const s = &h->s;

  4584.     int first_mb_in_slice, pps_id;

  4585.     int num_ref_idx_active_override_flag;

  4586.     static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};

  4587.     int slice_type;

  4588.     int default_ref_list_done = 0;

  4589. 

  4590.     s->current_picture.reference= h->nal_ref_idc != 0;

  4591.     s->dropable= h->nal_ref_idc == 0;

  4592. 

  4593.     first_mb_in_slice= get_ue_golomb(&s->gb);

  4594. 

  4595.     slice_type= get_ue_golomb(&s->gb);

  4596.     if(slice_type > 9){

  4597.         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);

  4598.         return -1;

  4599.     }

  4600.     if(slice_type > 4){

  4601.         slice_type -= 5;

  4602.         h->slice_type_fixed=1;

  4603.     }else

  4604.         h->slice_type_fixed=0;

  4605. 

  4606.     slice_type= slice_type_map[ slice_type ];

  4607.     if (slice_type == I_TYPE

  4608.         || (h->slice_num != 0 && slice_type == h->slice_type) ) {

  4609.         default_ref_list_done = 1;

  4610.     }

  4611.     h->slice_type= slice_type;

  4612. 

  4613.     s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though

  4614. 

  4615.     pps_id= get_ue_golomb(&s->gb);

  4616.     if(pps_id>255){

  4617.         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");

  4618.         return -1;

  4619.     }

  4620.     h->pps= h->pps_buffer[pps_id];

  4621.     if(h->pps.slice_group_count == 0){

  4622.         av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");

  4623.         return -1;

  4624.     }

  4625. 

  4626.     h->sps= h->sps_buffer[ h->pps.sps_id ];

  4627.     if(h->sps.log2_max_frame_num == 0){

  4628.         av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");

  4629.         return -1;

  4630.     }

  4631. 

  4632.     if(h->dequant_coeff_pps != pps_id){

  4633.         h->dequant_coeff_pps = pps_id;

  4634.         init_dequant_tables(h);

  4635.     }

  4636. 

  4637.     s->mb_width= h->sps.mb_width;

  4638.     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);

  4639. 

  4640.     h->b_stride=  s->mb_width*4;

  4641.     h->b8_stride= s->mb_width*2;

  4642. 

  4643.     s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );

  4644.     if(h->sps.frame_mbs_only_flag)

  4645.         s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);

  4646.     else

  4647.         s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck

  4648. 

  4649.     if (s->context_initialized

  4650.         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {

  4651.         free_tables(h);

  4652.         MPV_common_end(s);

  4653.     }

  4654.     if (!s->context_initialized) {

  4655.         if (MPV_common_init(s) < 0)

  4656.             return -1;

  4657. 

  4658.         if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly

  4659.             memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));

  4660.             memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));

  4661.         }else{

  4662.             int i;

  4663.             for(i=0; i<16; i++){

  4664. #define T(x) (x>>2) | ((x<<2) & 0xF)

  4665.                 h->zigzag_scan[i] = T(zigzag_scan[i]);

  4666.                 h-> field_scan[i] = T( field_scan[i]);

  4667. #undef T

  4668.             }

  4669.         }

  4670.         if(s->dsp.h264_idct8_add == ff_h264_idct8_add_c){

  4671.             memcpy(h->zigzag_scan8x8,       zigzag_scan8x8,       64*sizeof(uint8_t));

  4672.             memcpy(h->zigzag_scan8x8_cavlc, zigzag_scan8x8_cavlc, 64*sizeof(uint8_t));

  4673.             memcpy(h->field_scan8x8,        field_scan8x8,        64*sizeof(uint8_t));

  4674.             memcpy(h->field_scan8x8_cavlc,  field_scan8x8_cavlc,  64*sizeof(uint8_t));

  4675.         }else{

  4676.             int i;

  4677.             for(i=0; i<64; i++){

  4678. #define T(x) (x>>3) | ((x&7)<<3)

  4679.                 h->zigzag_scan8x8[i]       = T(zigzag_scan8x8[i]);

  4680.                 h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);

  4681.                 h->field_scan8x8[i]        = T(field_scan8x8[i]);

  4682.                 h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);

  4683. #undef T

  4684.             }

  4685.         }

  4686.         if(h->sps.transform_bypass){ //FIXME same ugly

  4687.             h->zigzag_scan_q0          = zigzag_scan;

  4688.             h->zigzag_scan8x8_q0       = zigzag_scan8x8;

  4689.             h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;

  4690.             h->field_scan_q0           = field_scan;

  4691.             h->field_scan8x8_q0        = field_scan8x8;

  4692.             h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;

  4693.         }else{

  4694.             h->zigzag_scan_q0          = h->zigzag_scan;

  4695.             h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;

  4696.             h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;

  4697.             h->field_scan_q0           = h->field_scan;

  4698.             h->field_scan8x8_q0        = h->field_scan8x8;

  4699.             h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;

  4700.         }

  4701. 

  4702.         alloc_tables(h);

  4703. 

  4704.         s->avctx->width = s->width;

  4705.         s->avctx->height = s->height;

  4706.         s->avctx->sample_aspect_ratio= h->sps.sar;

  4707.         if(!s->avctx->sample_aspect_ratio.den)

  4708.             s->avctx->sample_aspect_ratio.den = 1;

  4709. 

  4710.         if(h->sps.timing_info_present_flag){

  4711.             s->avctx->time_base= (AVRational){h->sps.num_units_in_tick * 2, h->sps.time_scale};

  4712.             if(h->x264_build > 0 && h->x264_build < 44)

  4713.                 s->avctx->time_base.den *= 2;

  4714.             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,

  4715.                       s->avctx->time_base.num, s->avctx->time_base.den, 1<<30);

  4716.         }

  4717.     }

  4718. 

  4719.     if(h->slice_num == 0){

  4720.         if(frame_start(h) < 0)

  4721.             return -1;

  4722.     }

  4723. 

  4724.     s->current_picture_ptr->frame_num= //FIXME frame_num cleanup

  4725.     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);

  4726. 

  4727.     h->mb_mbaff = 0;

  4728.     h->mb_aff_frame = 0;

  4729.     if(h->sps.frame_mbs_only_flag){

  4730.         s->picture_structure= PICT_FRAME;

  4731.     }else{

  4732.         if(get_bits1(&s->gb)) { //field_pic_flag

  4733.             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag

  4734.             av_log(h->s.avctx, AV_LOG_ERROR, "PAFF interlacing is not implemented\n");

  4735.         } else {

  4736.             s->picture_structure= PICT_FRAME;

  4737.             h->mb_aff_frame = h->sps.mb_aff;

  4738.         }

  4739.     }

  4740. 

  4741.     s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;

  4742.     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << h->mb_aff_frame;

  4743.     if(s->mb_y >= s->mb_height){

  4744.         return -1;

  4745.     }

  4746. 

  4747.     if(s->picture_structure==PICT_FRAME){

  4748.         h->curr_pic_num=   h->frame_num;

  4749.         h->max_pic_num= 1<< h->sps.log2_max_frame_num;

  4750.     }else{

  4751.         h->curr_pic_num= 2*h->frame_num;

  4752.         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);

  4753.     }

  4754. 

  4755.     if(h->nal_unit_type == NAL_IDR_SLICE){

  4756.         get_ue_golomb(&s->gb); /* idr_pic_id */

  4757.     }

  4758. 

  4759.     if(h->sps.poc_type==0){

  4760.         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);

  4761. 

  4762.         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){

  4763.             h->delta_poc_bottom= get_se_golomb(&s->gb);

  4764.         }

  4765.     }

  4766. 

  4767.     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){

  4768.         h->delta_poc[0]= get_se_golomb(&s->gb);

  4769. 

  4770.         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)

  4771.             h->delta_poc[1]= get_se_golomb(&s->gb);

  4772.     }

  4773. 

  4774.     init_poc(h);

  4775. 

  4776.     if(h->pps.redundant_pic_cnt_present){

  4777.         h->redundant_pic_count= get_ue_golomb(&s->gb);

  4778.     }

  4779. 

  4780.     //set defaults, might be overriden a few line later

  4781.     h->ref_count[0]= h->pps.ref_count[0];

  4782.     h->ref_count[1]= h->pps.ref_count[1];

  4783. 

  4784.     if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){

  4785.         if(h->slice_type == B_TYPE){

  4786.             h->direct_spatial_mv_pred= get_bits1(&s->gb);

  4787.             if(h->sps.mb_aff && h->direct_spatial_mv_pred)

  4788.                 av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + spatial direct mode is not implemented\n");

  4789.         }

  4790.         num_ref_idx_active_override_flag= get_bits1(&s->gb);

  4791. 

  4792.         if(num_ref_idx_active_override_flag){

  4793.             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;

  4794.             if(h->slice_type==B_TYPE)

  4795.                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;

  4796. 

  4797.             if(h->ref_count[0] > 32 || h->ref_count[1] > 32){

  4798.                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");

  4799.                 return -1;

  4800.             }

  4801.         }

  4802.     }

  4803. 

  4804.     if(!default_ref_list_done){

  4805.         fill_default_ref_list(h);

  4806.     }

  4807. 

  4808.     if(decode_ref_pic_list_reordering(h) < 0)

  4809.         return -1;

  4810. 

  4811.     if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))

  4812.        || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )

  4813.         pred_weight_table(h);

  4814.     else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)

  4815.         implicit_weight_table(h);

  4816.     else

  4817.         h->use_weight = 0;

  4818. 

  4819.     if(s->current_picture.reference)

  4820.         decode_ref_pic_marking(h);

  4821. 

  4822.     if(FRAME_MBAFF)

  4823.         fill_mbaff_ref_list(h);

  4824. 

  4825.     if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )

  4826.         h->cabac_init_idc = get_ue_golomb(&s->gb);

  4827. 

  4828.     h->last_qscale_diff = 0;

  4829.     s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);

  4830.     if(s->qscale<0 || s->qscale>51){

  4831.         av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);

  4832.         return -1;

  4833.     }

  4834.     h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);

  4835.     //FIXME qscale / qp ... stuff

  4836.     if(h->slice_type == SP_TYPE){

  4837.         get_bits1(&s->gb); /* sp_for_switch_flag */

  4838.     }

  4839.     if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){

  4840.         get_se_golomb(&s->gb); /* slice_qs_delta */

  4841.     }

  4842. 

  4843.     h->deblocking_filter = 1;

  4844.     h->slice_alpha_c0_offset = 0;

  4845.     h->slice_beta_offset = 0;

  4846.     if( h->pps.deblocking_filter_parameters_present ) {

  4847.         h->deblocking_filter= get_ue_golomb(&s->gb);

  4848.         if(h->deblocking_filter < 2)

  4849.             h->deblocking_filter^= 1; // 1<->0

  4850. 

  4851.         if( h->deblocking_filter ) {

  4852.             h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;

  4853.             h->slice_beta_offset = get_se_golomb(&s->gb) << 1;

  4854.         }

  4855.     }

  4856.     if(   s->avctx->skip_loop_filter >= AVDISCARD_ALL

  4857.        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONKEY && h->slice_type != I_TYPE)

  4858.        ||(s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  && h->slice_type == B_TYPE)

  4859.        ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))

  4860.         h->deblocking_filter= 0;

  4861. 

  4862. #if 0 //FMO

  4863.     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)

  4864.         slice_group_change_cycle= get_bits(&s->gb, ?);

  4865. #endif

  4866. 

  4867.     h->slice_num++;

  4868. 

  4869.     h->emu_edge_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;

  4870.     h->emu_edge_height= FRAME_MBAFF ? 0 : h->emu_edge_width;

  4871. 

  4872.     if(s->avctx->debug&FF_DEBUG_PICT_INFO){

  4873.         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",

  4874.                h->slice_num,

  4875.                (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),

  4876.                first_mb_in_slice,

  4877.                av_get_pict_type_char(h->slice_type),

  4878.                pps_id, h->frame_num,

  4879.                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],

  4880.                h->ref_count[0], h->ref_count[1],

  4881.                s->qscale,

  4882.                h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,

  4883.                h->use_weight,

  4884.                h->use_weight==1 && h->use_weight_chroma ? "c" : ""

  4885.                );

  4886.     }

  4887. 

  4888.     if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !s->current_picture.reference){

  4889.         s->me.qpel_put= s->dsp.put_2tap_qpel_pixels_tab;

  4890.         s->me.qpel_avg= s->dsp.avg_2tap_qpel_pixels_tab;

  4891.     }else{

  4892.         s->me.qpel_put= s->dsp.put_h264_qpel_pixels_tab;

  4893.         s->me.qpel_avg= s->dsp.avg_h264_qpel_pixels_tab;

  4894.     }

  4895. 

  4896.     return 0;

  4897. }

  4898. 

  4899. /**

  4900.  *

  4901.  */

  4902. static inline int get_level_prefix(GetBitContext *gb){

  4903.     unsigned int buf;

  4904.     int log;

  4905. 

  4906.     OPEN_READER(re, gb);

  4907.     UPDATE_CACHE(re, gb);

  4908.     buf=GET_CACHE(re, gb);

  4909. 

  4910.     log= 32 - av_log2(buf);

  4911. #ifdef TRACE

  4912.     print_bin(buf>>(32-log), log);

  4913.     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__);

  4914. #endif

  4915. 

  4916.     LAST_SKIP_BITS(re, gb, log);

  4917.     CLOSE_READER(re, gb);

  4918. 

  4919.     return log-1;

  4920. }

  4921. 

  4922. static inline int get_dct8x8_allowed(H264Context *h){

  4923.     int i;

  4924.     for(i=0; i<4; i++){

  4925.         if(!IS_SUB_8X8(h->sub_mb_type[i])

  4926.            || (!h->sps.direct_8x8_inference_flag && IS_DIRECT(h->sub_mb_type[i])))

  4927.             return 0;

  4928.     }

  4929.     return 1;

  4930. }

  4931. 

  4932. /**

  4933.  * decodes a residual block.

  4934.  * @param n block index

  4935.  * @param scantable scantable

  4936.  * @param max_coeff number of coefficients in the block

  4937.  * @return <0 if an error occured

  4938.  */

  4939. static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff){

  4940.     MpegEncContext * const s = &h->s;

  4941.     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};

  4942.     int level[16];

  4943.     int zeros_left, coeff_num, coeff_token, total_coeff, i, j, trailing_ones, run_before;

  4944. 

  4945.     //FIXME put trailing_onex into the context

  4946. 

  4947.     if(n == CHROMA_DC_BLOCK_INDEX){

  4948.         coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);

  4949.         total_coeff= coeff_token>>2;

  4950.     }else{

  4951.         if(n == LUMA_DC_BLOCK_INDEX){

  4952.             total_coeff= pred_non_zero_count(h, 0);

  4953.             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);

  4954.             total_coeff= coeff_token>>2;

  4955.         }else{

  4956.             total_coeff= pred_non_zero_count(h, n);

  4957.             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);

  4958.             total_coeff= coeff_token>>2;

  4959.             h->non_zero_count_cache[ scan8[n] ]= total_coeff;

  4960.         }

  4961.     }

  4962. 

  4963.     //FIXME set last_non_zero?

  4964. 

  4965.     if(total_coeff==0)

  4966.         return 0;

  4967. 

  4968.     trailing_ones= coeff_token&3;

  4969.     tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);

  4970.     assert(total_coeff<=16);

  4971. 

  4972.     for(i=0; i<trailing_ones; i++){

  4973.         level[i]= 1 - 2*get_bits1(gb);

  4974.     }

  4975. 

  4976.     if(i<total_coeff) {

  4977.         int level_code, mask;

  4978.         int suffix_length = total_coeff > 10 && trailing_ones < 3;

  4979.         int prefix= get_level_prefix(gb);

  4980. 

  4981.         //first coefficient has suffix_length equal to 0 or 1

  4982.         if(prefix<14){ //FIXME try to build a large unified VLC table for all this

  4983.             if(suffix_length)

  4984.                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part

  4985.             else

  4986.                 level_code= (prefix<<suffix_length); //part

  4987.         }else if(prefix==14){

  4988.             if(suffix_length)

  4989.                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part

  4990.             else

  4991.                 level_code= prefix + get_bits(gb, 4); //part

  4992.         }else if(prefix==15){

  4993.             level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part

  4994.             if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense

  4995.         }else{

  4996.             av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);

  4997.             return -1;

  4998.         }

  4999. 

  5000.         if(trailing_ones < 3) level_code += 2;

  5001. 

  5002.         suffix_length = 1;

  5003.         if(level_code > 5)

  5004.             suffix_length++;

  5005.         mask= -(level_code&1);

  5006.         level[i]= (((2+level_code)>>1) ^ mask) - mask;

  5007.         i++;

  5008. 

  5009.         //remaining coefficients have suffix_length > 0

  5010.         for(;i<total_coeff;i++) {

  5011.             static const int suffix_limit[7] = {0,5,11,23,47,95,INT_MAX };

  5012.             prefix = get_level_prefix(gb);

  5013.             if(prefix<15){

  5014.                 level_code = (prefix<<suffix_length) + get_bits(gb, suffix_length);

  5015.             }else if(prefix==15){

  5016.                 level_code =  (prefix<<suffix_length) + get_bits(gb, 12);

  5017.             }else{

  5018.                 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);

  5019.                 return -1;

  5020.             }

  5021.             mask= -(level_code&1);

  5022.             level[i]= (((2+level_code)>>1) ^ mask) - mask;

  5023.             if(level_code > suffix_limit[suffix_length])

  5024.                 suffix_length++;

  5025.         }

  5026.     }

  5027. 

  5028.     if(total_coeff == max_coeff)

  5029.         zeros_left=0;

  5030.     else{

  5031.         if(n == CHROMA_DC_BLOCK_INDEX)

  5032.             zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);

  5033.         else

  5034.             zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);

  5035.     }

  5036. 

  5037.     coeff_num = zeros_left + total_coeff - 1;

  5038.     j = scantable[coeff_num];

  5039.     if(n > 24){

  5040.         block[j] = level[0];

  5041.         for(i=1;i<total_coeff;i++) {

  5042.             if(zeros_left <= 0)

  5043.                 run_before = 0;

  5044.             else if(zeros_left < 7){

  5045.                 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);

  5046.             }else{

  5047.                 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);

  5048.             }

  5049.             zeros_left -= run_before;

  5050.             coeff_num -= 1 + run_before;

  5051.             j= scantable[ coeff_num ];

  5052. 

  5053.             block[j]= level[i];

  5054.         }

  5055.     }else{

  5056.         block[j] = (level[0] * qmul[j] + 32)>>6;

  5057.         for(i=1;i<total_coeff;i++) {

  5058.             if(zeros_left <= 0)

  5059.                 run_before = 0;

  5060.             else if(zeros_left < 7){

  5061.                 run_before= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);

  5062.             }else{

  5063.                 run_before= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);

  5064.             }

  5065.             zeros_left -= run_before;

  5066.             coeff_num -= 1 + run_before;

  5067.             j= scantable[ coeff_num ];

  5068. 

  5069.             block[j]= (level[i] * qmul[j] + 32)>>6;

  5070.         }

  5071.     }

  5072. 

  5073.     if(zeros_left<0){

  5074.         av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);

  5075.         return -1;

  5076.     }

  5077. 

  5078.     return 0;

  5079. }

  5080. 

  5081. static void predict_field_decoding_flag(H264Context *h){

  5082.     MpegEncContext * const s = &h->s;

  5083.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  5084.     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)

  5085.                 ? s->current_picture.mb_type[mb_xy-1]

  5086.                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)

  5087.                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]

  5088.                 : 0;

  5089.     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;

  5090. }

  5091. 

  5092. /**

  5093.  * decodes a P_SKIP or B_SKIP macroblock

  5094.  */

  5095. static void decode_mb_skip(H264Context *h){

  5096.     MpegEncContext * const s = &h->s;

  5097.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  5098.     int mb_type=0;

  5099. 

  5100.     memset(h->non_zero_count[mb_xy], 0, 16);

  5101.     memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui

  5102. 

  5103.     if(MB_FIELD)

  5104.         mb_type|= MB_TYPE_INTERLACED;

  5105. 

  5106.     if( h->slice_type == B_TYPE )

  5107.     {

  5108.         // just for fill_caches. pred_direct_motion will set the real mb_type

  5109.         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;

  5110. 

  5111.         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...

  5112.         pred_direct_motion(h, &mb_type);

  5113.         if(h->pps.cabac){

  5114.             fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);

  5115.             fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);

  5116.         }

  5117.     }

  5118.     else

  5119.     {

  5120.         int mx, my;

  5121.         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;

  5122. 

  5123.         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...

  5124.         pred_pskip_motion(h, &mx, &my);

  5125.         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);

  5126.         fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);

  5127.         if(h->pps.cabac)

  5128.             fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);

  5129.     }

  5130. 

  5131.     write_back_motion(h, mb_type);

  5132.     s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;

  5133.     s->current_picture.qscale_table[mb_xy]= s->qscale;

  5134.     h->slice_table[ mb_xy ]= h->slice_num;

  5135.     h->prev_mb_skipped= 1;

  5136. }

  5137. 

  5138. /**

  5139.  * decodes a macroblock

  5140.  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed

  5141.  */

  5142. static int decode_mb_cavlc(H264Context *h){

  5143.     MpegEncContext * const s = &h->s;

  5144.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  5145.     int mb_type, partition_count, cbp;

  5146.     int dct8x8_allowed= h->pps.transform_8x8_mode;

  5147. 

  5148.     s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?

  5149. 

  5150.     tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);

  5151.     cbp = 0; /* avoid warning. FIXME: find a solution without slowing

  5152.                 down the code */

  5153.     if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){

  5154.         if(s->mb_skip_run==-1)

  5155.             s->mb_skip_run= get_ue_golomb(&s->gb);

  5156. 

  5157.         if (s->mb_skip_run--) {

  5158.             if(FRAME_MBAFF && (s->mb_y&1) == 0){

  5159.                 if(s->mb_skip_run==0)

  5160.                     h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);

  5161.                 else

  5162.                     predict_field_decoding_flag(h);

  5163.             }

  5164.             decode_mb_skip(h);

  5165.             return 0;

  5166.         }

  5167.     }

  5168.     if(FRAME_MBAFF){

  5169.         if( (s->mb_y&1) == 0 )

  5170.             h->mb_mbaff = h->mb_field_decoding_flag = get_bits1(&s->gb);

  5171.     }else

  5172.         h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);

  5173. 

  5174.     h->prev_mb_skipped= 0;

  5175. 

  5176.     mb_type= get_ue_golomb(&s->gb);

  5177.     if(h->slice_type == B_TYPE){

  5178.         if(mb_type < 23){

  5179.             partition_count= b_mb_type_info[mb_type].partition_count;

  5180.             mb_type=         b_mb_type_info[mb_type].type;

  5181.         }else{

  5182.             mb_type -= 23;

  5183.             goto decode_intra_mb;

  5184.         }

  5185.     }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){

  5186.         if(mb_type < 5){

  5187.             partition_count= p_mb_type_info[mb_type].partition_count;

  5188.             mb_type=         p_mb_type_info[mb_type].type;

  5189.         }else{

  5190.             mb_type -= 5;

  5191.             goto decode_intra_mb;

  5192.         }

  5193.     }else{

  5194.        assert(h->slice_type == I_TYPE);

  5195. decode_intra_mb:

  5196.         if(mb_type > 25){

  5197.             av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice to large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);

  5198.             return -1;

  5199.         }

  5200.         partition_count=0;

  5201.         cbp= i_mb_type_info[mb_type].cbp;

  5202.         h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;

  5203.         mb_type= i_mb_type_info[mb_type].type;

  5204.     }

  5205. 

  5206.     if(MB_FIELD)

  5207.         mb_type |= MB_TYPE_INTERLACED;

  5208. 

  5209.     h->slice_table[ mb_xy ]= h->slice_num;

  5210. 

  5211.     if(IS_INTRA_PCM(mb_type)){

  5212.         unsigned int x, y;

  5213. 

  5214.         // we assume these blocks are very rare so we dont optimize it

  5215.         align_get_bits(&s->gb);

  5216. 

  5217.         // The pixels are stored in the same order as levels in h->mb array.

  5218.         for(y=0; y<16; y++){

  5219.             const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);

  5220.             for(x=0; x<16; x++){

  5221.                 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));

  5222.                 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);

  5223.             }

  5224.         }

  5225.         for(y=0; y<8; y++){

  5226.             const int index= 256 + 4*(y&3) + 32*(y>>2);

  5227.             for(x=0; x<8; x++){

  5228.                 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));

  5229.                 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);

  5230.             }

  5231.         }

  5232.         for(y=0; y<8; y++){

  5233.             const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);

  5234.             for(x=0; x<8; x++){

  5235.                 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));

  5236.                 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);

  5237.             }

  5238.         }

  5239. 

  5240.         // In deblocking, the quantizer is 0

  5241.         s->current_picture.qscale_table[mb_xy]= 0;

  5242.         h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);

  5243.         // All coeffs are present

  5244.         memset(h->non_zero_count[mb_xy], 16, 16);

  5245. 

  5246.         s->current_picture.mb_type[mb_xy]= mb_type;

  5247.         return 0;

  5248.     }

  5249. 

  5250.     if(MB_MBAFF){

  5251.         h->ref_count[0] <<= 1;

  5252.         h->ref_count[1] <<= 1;

  5253.     }

  5254. 

  5255.     fill_caches(h, mb_type, 0);

  5256. 

  5257.     //mb_pred

  5258.     if(IS_INTRA(mb_type)){

  5259. //            init_top_left_availability(h);

  5260.             if(IS_INTRA4x4(mb_type)){

  5261.                 int i;

  5262.                 int di = 1;

  5263.                 if(dct8x8_allowed && get_bits1(&s->gb)){

  5264.                     mb_type |= MB_TYPE_8x8DCT;

  5265.                     di = 4;

  5266.                 }

  5267. 

  5268. //                fill_intra4x4_pred_table(h);

  5269.                 for(i=0; i<16; i+=di){

  5270.                     int mode= pred_intra_mode(h, i);

  5271. 

  5272.                     if(!get_bits1(&s->gb)){

  5273.                         const int rem_mode= get_bits(&s->gb, 3);

  5274.                         mode = rem_mode + (rem_mode >= mode);

  5275.                     }

  5276. 

  5277.                     if(di==4)

  5278.                         fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );

  5279.                     else

  5280.                         h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;

  5281.                 }

  5282.                 write_back_intra_pred_mode(h);

  5283.                 if( check_intra4x4_pred_mode(h) < 0)

  5284.                     return -1;

  5285.             }else{

  5286.                 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);

  5287.                 if(h->intra16x16_pred_mode < 0)

  5288.                     return -1;

  5289.             }

  5290.             h->chroma_pred_mode= get_ue_golomb(&s->gb);

  5291. 

  5292.             h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);

  5293.             if(h->chroma_pred_mode < 0)

  5294.                 return -1;

  5295.     }else if(partition_count==4){

  5296.         int i, j, sub_partition_count[4], list, ref[2][4];

  5297. 

  5298.         if(h->slice_type == B_TYPE){

  5299.             for(i=0; i<4; i++){

  5300.                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);

  5301.                 if(h->sub_mb_type[i] >=13){

  5302.                     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);

  5303.                     return -1;

  5304.                 }

  5305.                 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;

  5306.                 h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;

  5307.             }

  5308.             if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])

  5309.                || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {

  5310.                 pred_direct_motion(h, &mb_type);

  5311.                 h->ref_cache[0][scan8[4]] =

  5312.                 h->ref_cache[1][scan8[4]] =

  5313.                 h->ref_cache[0][scan8[12]] =

  5314.                 h->ref_cache[1][scan8[12]] = PART_NOT_AVAILABLE;

  5315.             }

  5316.         }else{

  5317.             assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?

  5318.             for(i=0; i<4; i++){

  5319.                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);

  5320.                 if(h->sub_mb_type[i] >=4){

  5321.                     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);

  5322.                     return -1;

  5323.                 }

  5324.                 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;

  5325.                 h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;

  5326.             }

  5327.         }

  5328. 

  5329.         for(list=0; list<2; list++){

  5330.             int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];

  5331.             if(ref_count == 0) continue;

  5332.             for(i=0; i<4; i++){

  5333.                 if(IS_DIRECT(h->sub_mb_type[i])) continue;

  5334.                 if(IS_DIR(h->sub_mb_type[i], 0, list)){

  5335.                     ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?

  5336.                 }else{

  5337.                  //FIXME

  5338.                     ref[list][i] = -1;

  5339.                 }

  5340.             }

  5341.         }

  5342. 

  5343.         if(dct8x8_allowed)

  5344.             dct8x8_allowed = get_dct8x8_allowed(h);

  5345. 

  5346.         for(list=0; list<2; list++){

  5347.             const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];

  5348.             if(ref_count == 0) continue;

  5349. 

  5350.             for(i=0; i<4; i++){

  5351.                 if(IS_DIRECT(h->sub_mb_type[i])) {

  5352.                     h->ref_cache[list][ scan8[4*i] ] = h->ref_cache[list][ scan8[4*i]+1 ];

  5353.                     continue;

  5354.                 }

  5355.                 h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=

  5356.                 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];

  5357. 

  5358.                 if(IS_DIR(h->sub_mb_type[i], 0, list)){

  5359.                     const int sub_mb_type= h->sub_mb_type[i];

  5360.                     const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;

  5361.                     for(j=0; j<sub_partition_count[i]; j++){

  5362.                         int mx, my;

  5363.                         const int index= 4*i + block_width*j;

  5364.                         int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];

  5365.                         pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);

  5366.                         mx += get_se_golomb(&s->gb);

  5367.                         my += get_se_golomb(&s->gb);

  5368.                         tprintf("final mv:%d %d\n", mx, my);

  5369. 

  5370.                         if(IS_SUB_8X8(sub_mb_type)){

  5371.                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=

  5372.                             mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;

  5373.                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=

  5374.                             mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;

  5375.                         }else if(IS_SUB_8X4(sub_mb_type)){

  5376.                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;

  5377.                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;

  5378.                         }else if(IS_SUB_4X8(sub_mb_type)){

  5379.                             mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;

  5380.                             mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;

  5381.                         }else{

  5382.                             assert(IS_SUB_4X4(sub_mb_type));

  5383.                             mv_cache[ 0 ][0]= mx;

  5384.                             mv_cache[ 0 ][1]= my;

  5385.                         }

  5386.                     }

  5387.                 }else{

  5388.                     uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];

  5389.                     p[0] = p[1]=

  5390.                     p[8] = p[9]= 0;

  5391.                 }

  5392.             }

  5393.         }

  5394.     }else if(IS_DIRECT(mb_type)){

  5395.         pred_direct_motion(h, &mb_type);

  5396.         dct8x8_allowed &= h->sps.direct_8x8_inference_flag;

  5397.     }else{

  5398.         int list, mx, my, i;

  5399.          //FIXME we should set ref_idx_l? to 0 if we use that later ...

  5400.         if(IS_16X16(mb_type)){

  5401.             for(list=0; list<2; list++){

  5402.                 if(h->ref_count[list]>0){

  5403.                     if(IS_DIR(mb_type, 0, list)){

  5404.                         const int val= get_te0_golomb(&s->gb, h->ref_count[list]);

  5405.                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);

  5406.                     }else

  5407.                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1);

  5408.                 }

  5409.             }

  5410.             for(list=0; list<2; list++){

  5411.                 if(IS_DIR(mb_type, 0, list)){

  5412.                     pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);

  5413.                     mx += get_se_golomb(&s->gb);

  5414.                     my += get_se_golomb(&s->gb);

  5415.                     tprintf("final mv:%d %d\n", mx, my);

  5416. 

  5417.                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);

  5418.                 }else

  5419.                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);

  5420.             }

  5421.         }

  5422.         else if(IS_16X8(mb_type)){

  5423.             for(list=0; list<2; list++){

  5424.                 if(h->ref_count[list]>0){

  5425.                     for(i=0; i<2; i++){

  5426.                         if(IS_DIR(mb_type, i, list)){

  5427.                             const int val= get_te0_golomb(&s->gb, h->ref_count[list]);

  5428.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);

  5429.                         }else

  5430.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);

  5431.                     }

  5432.                 }

  5433.             }

  5434.             for(list=0; list<2; list++){

  5435.                 for(i=0; i<2; i++){

  5436.                     if(IS_DIR(mb_type, i, list)){

  5437.                         pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);

  5438.                         mx += get_se_golomb(&s->gb);

  5439.                         my += get_se_golomb(&s->gb);

  5440.                         tprintf("final mv:%d %d\n", mx, my);

  5441. 

  5442.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);

  5443.                     }else

  5444.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);

  5445.                 }

  5446.             }

  5447.         }else{

  5448.             assert(IS_8X16(mb_type));

  5449.             for(list=0; list<2; list++){

  5450.                 if(h->ref_count[list]>0){

  5451.                     for(i=0; i<2; i++){

  5452.                         if(IS_DIR(mb_type, i, list)){ //FIXME optimize

  5453.                             const int val= get_te0_golomb(&s->gb, h->ref_count[list]);

  5454.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);

  5455.                         }else

  5456.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);

  5457.                     }

  5458.                 }

  5459.             }

  5460.             for(list=0; list<2; list++){

  5461.                 for(i=0; i<2; i++){

  5462.                     if(IS_DIR(mb_type, i, list)){

  5463.                         pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);

  5464.                         mx += get_se_golomb(&s->gb);

  5465.                         my += get_se_golomb(&s->gb);

  5466.                         tprintf("final mv:%d %d\n", mx, my);

  5467. 

  5468.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);

  5469.                     }else

  5470.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);

  5471.                 }

  5472.             }

  5473.         }

  5474.     }

  5475. 

  5476.     if(IS_INTER(mb_type))

  5477.         write_back_motion(h, mb_type);

  5478. 

  5479.     if(!IS_INTRA16x16(mb_type)){

  5480.         cbp= get_ue_golomb(&s->gb);

  5481.         if(cbp > 47){

  5482.             av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);

  5483.             return -1;

  5484.         }

  5485. 

  5486.         if(IS_INTRA4x4(mb_type))

  5487.             cbp= golomb_to_intra4x4_cbp[cbp];

  5488.         else

  5489.             cbp= golomb_to_inter_cbp[cbp];

  5490.     }

  5491.     h->cbp = cbp;

  5492. 

  5493.     if(dct8x8_allowed && (cbp&15) && !IS_INTRA(mb_type)){

  5494.         if(get_bits1(&s->gb))

  5495.             mb_type |= MB_TYPE_8x8DCT;

  5496.     }

  5497.     s->current_picture.mb_type[mb_xy]= mb_type;

  5498. 

  5499.     if(cbp || IS_INTRA16x16(mb_type)){

  5500.         int i8x8, i4x4, chroma_idx;

  5501.         int chroma_qp, dquant;

  5502.         GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;

  5503.         const uint8_t *scan, *scan8x8, *dc_scan;

  5504. 

  5505. //        fill_non_zero_count_cache(h);

  5506. 

  5507.         if(IS_INTERLACED(mb_type)){

  5508.             scan8x8= s->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;

  5509.             scan= s->qscale ? h->field_scan : h->field_scan_q0;

  5510.             dc_scan= luma_dc_field_scan;

  5511.         }else{

  5512.             scan8x8= s->qscale ? h->zigzag_scan8x8_cavlc : h->zigzag_scan8x8_cavlc_q0;

  5513.             scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;

  5514.             dc_scan= luma_dc_zigzag_scan;

  5515.         }

  5516. 

  5517.         dquant= get_se_golomb(&s->gb);

  5518. 

  5519.         if( dquant > 25 || dquant < -26 ){

  5520.             av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);

  5521.             return -1;

  5522.         }

  5523. 

  5524.         s->qscale += dquant;

  5525.         if(((unsigned)s->qscale) > 51){

  5526.             if(s->qscale<0) s->qscale+= 52;

  5527.             else            s->qscale-= 52;

  5528.         }

  5529. 

  5530.         h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);

  5531.         if(IS_INTRA16x16(mb_type)){

  5532.             if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, h->dequant4_coeff[0][s->qscale], 16) < 0){

  5533.                 return -1; //FIXME continue if partitioned and other return -1 too

  5534.             }

  5535. 

  5536.             assert((cbp&15) == 0 || (cbp&15) == 15);

  5537. 

  5538.             if(cbp&15){

  5539.                 for(i8x8=0; i8x8<4; i8x8++){

  5540.                     for(i4x4=0; i4x4<4; i4x4++){

  5541.                         const int index= i4x4 + 4*i8x8;

  5542.                         if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 ){

  5543.                             return -1;

  5544.                         }

  5545.                     }

  5546.                 }

  5547.             }else{

  5548.                 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);

  5549.             }

  5550.         }else{

  5551.             for(i8x8=0; i8x8<4; i8x8++){

  5552.                 if(cbp & (1<<i8x8)){

  5553.                     if(IS_8x8DCT(mb_type)){

  5554.                         DCTELEM *buf = &h->mb[64*i8x8];

  5555.                         uint8_t *nnz;

  5556.                         for(i4x4=0; i4x4<4; i4x4++){

  5557.                             if( decode_residual(h, gb, buf, i4x4+4*i8x8, scan8x8+16*i4x4,

  5558.                                                 h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 16) <0 )

  5559.                                 return -1;

  5560.                         }

  5561.                         nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];

  5562.                         nnz[0] += nnz[1] + nnz[8] + nnz[9];

  5563.                     }else{

  5564.                         for(i4x4=0; i4x4<4; i4x4++){

  5565.                             const int index= i4x4 + 4*i8x8;

  5566. 

  5567.                             if( decode_residual(h, gb, h->mb + 16*index, index, scan, h->dequant4_coeff[IS_INTRA( mb_type ) ? 0:3][s->qscale], 16) <0 ){

  5568.                                 return -1;

  5569.                             }

  5570.                         }

  5571.                     }

  5572.                 }else{

  5573.                     uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];

  5574.                     nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;

  5575.                 }

  5576.             }

  5577.         }

  5578. 

  5579.         if(cbp&0x30){

  5580.             for(chroma_idx=0; chroma_idx<2; chroma_idx++)

  5581.                 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, NULL, 4) < 0){

  5582.                     return -1;

  5583.                 }

  5584.         }

  5585. 

  5586.         if(cbp&0x20){

  5587.             for(chroma_idx=0; chroma_idx<2; chroma_idx++){

  5588.                 for(i4x4=0; i4x4<4; i4x4++){

  5589.                     const int index= 16 + 4*chroma_idx + i4x4;

  5590.                     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){

  5591.                         return -1;

  5592.                     }

  5593.                 }

  5594.             }

  5595.         }else{

  5596.             uint8_t * const nnz= &h->non_zero_count_cache[0];

  5597.             nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =

  5598.             nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;

  5599.         }

  5600.     }else{

  5601.         uint8_t * const nnz= &h->non_zero_count_cache[0];

  5602.         fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);

  5603.         nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =

  5604.         nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;

  5605.     }

  5606.     s->current_picture.qscale_table[mb_xy]= s->qscale;

  5607.     write_back_non_zero_count(h);

  5608. 

  5609.     if(MB_MBAFF){

  5610.         h->ref_count[0] >>= 1;

  5611.         h->ref_count[1] >>= 1;

  5612.     }

  5613. 

  5614.     return 0;

  5615. }

  5616. 

  5617. static int decode_cabac_field_decoding_flag(H264Context *h) {

  5618.     MpegEncContext * const s = &h->s;

  5619.     const int mb_x = s->mb_x;

  5620.     const int mb_y = s->mb_y & ~1;

  5621.     const int mba_xy = mb_x - 1 +  mb_y   *s->mb_stride;

  5622.     const int mbb_xy = mb_x     + (mb_y-2)*s->mb_stride;

  5623. 

  5624.     unsigned int ctx = 0;

  5625. 

  5626.     if( h->slice_table[mba_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) {

  5627.         ctx += 1;

  5628.     }

  5629.     if( h->slice_table[mbb_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) {

  5630.         ctx += 1;

  5631.     }

  5632. 

  5633.     return get_cabac( &h->cabac, &h->cabac_state[70 + ctx] );

  5634. }

  5635. 

  5636. static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {

  5637.     uint8_t *state= &h->cabac_state[ctx_base];

  5638.     int mb_type;

  5639. 

  5640.     if(intra_slice){

  5641.         MpegEncContext * const s = &h->s;

  5642.         const int mba_xy = h->left_mb_xy[0];

  5643.         const int mbb_xy = h->top_mb_xy;

  5644.         int ctx=0;

  5645.         if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )

  5646.             ctx++;

  5647.         if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )

  5648.             ctx++;

  5649.         if( get_cabac( &h->cabac, &state[ctx] ) == 0 )

  5650.             return 0;   /* I4x4 */

  5651.         state += 2;

  5652.     }else{

  5653.         if( get_cabac( &h->cabac, &state[0] ) == 0 )

  5654.             return 0;   /* I4x4 */

  5655.     }

  5656. 

  5657.     if( get_cabac_terminate( &h->cabac ) )

  5658.         return 25;  /* PCM */

  5659. 

  5660.     mb_type = 1; /* I16x16 */

  5661.     mb_type += 12 * get_cabac( &h->cabac, &state[1] ); /* cbp_luma != 0 */

  5662.     if( get_cabac( &h->cabac, &state[2] ) ) /* cbp_chroma */

  5663.         mb_type += 4 + 4 * get_cabac( &h->cabac, &state[2+intra_slice] );

  5664.     mb_type += 2 * get_cabac( &h->cabac, &state[3+intra_slice] );

  5665.     mb_type += 1 * get_cabac( &h->cabac, &state[3+2*intra_slice] );

  5666.     return mb_type;

  5667. }

  5668. 

  5669. static int decode_cabac_mb_type( H264Context *h ) {

  5670.     MpegEncContext * const s = &h->s;

  5671. 

  5672.     if( h->slice_type == I_TYPE ) {

  5673.         return decode_cabac_intra_mb_type(h, 3, 1);

  5674.     } else if( h->slice_type == P_TYPE ) {

  5675.         if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {

  5676.             /* P-type */

  5677.             if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {

  5678.                 /* P_L0_D16x16, P_8x8 */

  5679.                 return 3 * get_cabac( &h->cabac, &h->cabac_state[16] );

  5680.             } else {

  5681.                 /* P_L0_D8x16, P_L0_D16x8 */

  5682.                 return 2 - get_cabac( &h->cabac, &h->cabac_state[17] );

  5683.             }

  5684.         } else {

  5685.             return decode_cabac_intra_mb_type(h, 17, 0) + 5;

  5686.         }

  5687.     } else if( h->slice_type == B_TYPE ) {

  5688.         const int mba_xy = h->left_mb_xy[0];

  5689.         const int mbb_xy = h->top_mb_xy;

  5690.         int ctx = 0;

  5691.         int bits;

  5692. 

  5693.         if( h->slice_table[mba_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )

  5694.             ctx++;

  5695.         if( h->slice_table[mbb_xy] == h->slice_num && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )

  5696.             ctx++;

  5697. 

  5698.         if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )

  5699.             return 0; /* B_Direct_16x16 */

  5700. 

  5701.         if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {

  5702.             return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */

  5703.         }

  5704. 

  5705.         bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;

  5706.         bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;

  5707.         bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;

  5708.         bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );

  5709.         if( bits < 8 )

  5710.             return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */

  5711.         else if( bits == 13 ) {

  5712.             return decode_cabac_intra_mb_type(h, 32, 0) + 23;

  5713.         } else if( bits == 14 )

  5714.             return 11; /* B_L1_L0_8x16 */

  5715.         else if( bits == 15 )

  5716.             return 22; /* B_8x8 */

  5717. 

  5718.         bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );

  5719.         return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */

  5720.     } else {

  5721.         /* TODO SI/SP frames? */

  5722.         return -1;

  5723.     }

  5724. }

  5725. 

  5726. static int decode_cabac_mb_skip( H264Context *h, int mb_x, int mb_y ) {

  5727.     MpegEncContext * const s = &h->s;

  5728.     int mba_xy, mbb_xy;

  5729.     int ctx = 0;

  5730. 

  5731.     if(FRAME_MBAFF){ //FIXME merge with the stuff in fill_caches?

  5732.         int mb_xy = mb_x + (mb_y&~1)*s->mb_stride;

  5733.         mba_xy = mb_xy - 1;

  5734.         if( (mb_y&1)

  5735.             && h->slice_table[mba_xy] == h->slice_num

  5736.             && MB_FIELD == !!IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) )

  5737.             mba_xy += s->mb_stride;

  5738.         if( MB_FIELD ){

  5739.             mbb_xy = mb_xy - s->mb_stride;

  5740.             if( !(mb_y&1)

  5741.                 && h->slice_table[mbb_xy] == h->slice_num

  5742.                 && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) )

  5743.                 mbb_xy -= s->mb_stride;

  5744.         }else

  5745.             mbb_xy = mb_x + (mb_y-1)*s->mb_stride;

  5746.     }else{

  5747.         int mb_xy = mb_x + mb_y*s->mb_stride;

  5748.         mba_xy = mb_xy - 1;

  5749.         mbb_xy = mb_xy - s->mb_stride;

  5750.     }

  5751. 

  5752.     if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))

  5753.         ctx++;

  5754.     if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))

  5755.         ctx++;

  5756. 

  5757.     if( h->slice_type == B_TYPE )

  5758.         ctx += 13;

  5759.     return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );

  5760. }

  5761. 

  5762. static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {

  5763.     int mode = 0;

  5764. 

  5765.     if( get_cabac( &h->cabac, &h->cabac_state[68] ) )

  5766.         return pred_mode;

  5767. 

  5768.     mode += 1 * get_cabac( &h->cabac, &h->cabac_state[69] );

  5769.     mode += 2 * get_cabac( &h->cabac, &h->cabac_state[69] );

  5770.     mode += 4 * get_cabac( &h->cabac, &h->cabac_state[69] );

  5771. 

  5772.     if( mode >= pred_mode )

  5773.         return mode + 1;

  5774.     else

  5775.         return mode;

  5776. }

  5777. 

  5778. static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {

  5779.     const int mba_xy = h->left_mb_xy[0];

  5780.     const int mbb_xy = h->top_mb_xy;

  5781. 

  5782.     int ctx = 0;

  5783. 

  5784.     /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */

  5785.     if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )

  5786.         ctx++;

  5787. 

  5788.     if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )

  5789.         ctx++;

  5790. 

  5791.     if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )

  5792.         return 0;

  5793. 

  5794.     if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )

  5795.         return 1;

  5796.     if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )

  5797.         return 2;

  5798.     else

  5799.         return 3;

  5800. }

  5801. 

  5802. static const uint8_t block_idx_x[16] = {

  5803.     0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3

  5804. };

  5805. static const uint8_t block_idx_y[16] = {

  5806.     0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3

  5807. };

  5808. static const uint8_t block_idx_xy[4][4] = {

  5809.     { 0, 2, 8,  10},

  5810.     { 1, 3, 9,  11},

  5811.     { 4, 6, 12, 14},

  5812.     { 5, 7, 13, 15}

  5813. };

  5814. 

  5815. static int decode_cabac_mb_cbp_luma( H264Context *h) {

  5816.     int cbp = 0;

  5817.     int cbp_b = -1;

  5818.     int i8x8;

  5819. 

  5820.     if( h->slice_table[h->top_mb_xy] == h->slice_num ) {

  5821.         cbp_b = h->top_cbp;

  5822.         tprintf("cbp_b = top_cbp = %x\n", cbp_b);

  5823.     }

  5824. 

  5825.     for( i8x8 = 0; i8x8 < 4; i8x8++ ) {

  5826.         int cbp_a = -1;

  5827.         int x, y;

  5828.         int ctx = 0;

  5829. 

  5830.         x = block_idx_x[4*i8x8];

  5831.         y = block_idx_y[4*i8x8];

  5832. 

  5833.         if( x > 0 )

  5834.             cbp_a = cbp;

  5835.         else if( h->slice_table[h->left_mb_xy[0]] == h->slice_num ) {

  5836.             cbp_a = h->left_cbp;

  5837.             tprintf("cbp_a = left_cbp = %x\n", cbp_a);

  5838.         }

  5839. 

  5840.         if( y > 0 )

  5841.             cbp_b = cbp;

  5842. 

  5843.         /* No need to test for skip as we put 0 for skip block */

  5844.         /* No need to test for IPCM as we put 1 for IPCM block */

  5845.         if( cbp_a >= 0 ) {

  5846.             int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;

  5847.             if( ((cbp_a >> i8x8a)&0x01) == 0 )

  5848.                 ctx++;

  5849.         }

  5850. 

  5851.         if( cbp_b >= 0 ) {

  5852.             int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;

  5853.             if( ((cbp_b >> i8x8b)&0x01) == 0 )

  5854.                 ctx += 2;

  5855.         }

  5856. 

  5857.         if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {

  5858.             cbp |= 1 << i8x8;

  5859.         }

  5860.     }

  5861.     return cbp;

  5862. }

  5863. static int decode_cabac_mb_cbp_chroma( H264Context *h) {

  5864.     int ctx;

  5865.     int cbp_a, cbp_b;

  5866. 

  5867.     cbp_a = (h->left_cbp>>4)&0x03;

  5868.     cbp_b = (h-> top_cbp>>4)&0x03;

  5869. 

  5870.     ctx = 0;

  5871.     if( cbp_a > 0 ) ctx++;

  5872.     if( cbp_b > 0 ) ctx += 2;

  5873.     if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )

  5874.         return 0;

  5875. 

  5876.     ctx = 4;

  5877.     if( cbp_a == 2 ) ctx++;

  5878.     if( cbp_b == 2 ) ctx += 2;

  5879.     return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );

  5880. }

  5881. static int decode_cabac_mb_dqp( H264Context *h) {

  5882.     MpegEncContext * const s = &h->s;

  5883.     int mbn_xy;

  5884.     int   ctx = 0;

  5885.     int   val = 0;

  5886. 

  5887.     if( s->mb_x > 0 )

  5888.         mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;

  5889.     else

  5890.         mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;

  5891. 

  5892.     if( h->last_qscale_diff != 0 )

  5893.         ctx++;

  5894. 

  5895.     while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {

  5896.         if( ctx < 2 )

  5897.             ctx = 2;

  5898.         else

  5899.             ctx = 3;

  5900.         val++;

  5901.         if(val > 102) //prevent infinite loop

  5902.             return INT_MIN;

  5903.     }

  5904. 

  5905.     if( val&0x01 )

  5906.         return (val + 1)/2;

  5907.     else

  5908.         return -(val + 1)/2;

  5909. }

  5910. static int decode_cabac_p_mb_sub_type( H264Context *h ) {

  5911.     if( get_cabac( &h->cabac, &h->cabac_state[21] ) )

  5912.         return 0;   /* 8x8 */

  5913.     if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )

  5914.         return 1;   /* 8x4 */

  5915.     if( get_cabac( &h->cabac, &h->cabac_state[23] ) )

  5916.         return 2;   /* 4x8 */

  5917.     return 3;       /* 4x4 */

  5918. }

  5919. static int decode_cabac_b_mb_sub_type( H264Context *h ) {

  5920.     int type;

  5921.     if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )

  5922.         return 0;   /* B_Direct_8x8 */

  5923.     if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )

  5924.         return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */

  5925.     type = 3;

  5926.     if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {

  5927.         if( get_cabac( &h->cabac, &h->cabac_state[39] ) )

  5928.             return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */

  5929.         type += 4;

  5930.     }

  5931.     type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );

  5932.     type +=   get_cabac( &h->cabac, &h->cabac_state[39] );

  5933.     return type;

  5934. }

  5935. 

  5936. static inline int decode_cabac_mb_transform_size( H264Context *h ) {

  5937.     return get_cabac( &h->cabac, &h->cabac_state[399 + h->neighbor_transform_size] );

  5938. }

  5939. 

  5940. static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {

  5941.     int refa = h->ref_cache[list][scan8[n] - 1];

  5942.     int refb = h->ref_cache[list][scan8[n] - 8];

  5943.     int ref  = 0;

  5944.     int ctx  = 0;

  5945. 

  5946.     if( h->slice_type == B_TYPE) {

  5947.         if( refa > 0 && !h->direct_cache[scan8[n] - 1] )

  5948.             ctx++;

  5949.         if( refb > 0 && !h->direct_cache[scan8[n] - 8] )

  5950.             ctx += 2;

  5951.     } else {

  5952.         if( refa > 0 )

  5953.             ctx++;

  5954.         if( refb > 0 )

  5955.             ctx += 2;

  5956.     }

  5957. 

  5958.     while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {

  5959.         ref++;

  5960.         if( ctx < 4 )

  5961.             ctx = 4;

  5962.         else

  5963.             ctx = 5;

  5964.     }

  5965.     return ref;

  5966. }

  5967. 

  5968. static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {

  5969.     int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +

  5970.                abs( h->mvd_cache[list][scan8[n] - 8][l] );

  5971.     int ctxbase = (l == 0) ? 40 : 47;

  5972.     int ctx, mvd;

  5973. 

  5974.     if( amvd < 3 )

  5975.         ctx = 0;

  5976.     else if( amvd > 32 )

  5977.         ctx = 2;

  5978.     else

  5979.         ctx = 1;

  5980. 

  5981.     if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))

  5982.         return 0;

  5983. 

  5984.     mvd= 1;

  5985.     ctx= 3;

  5986.     while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {

  5987.         mvd++;

  5988.         if( ctx < 6 )

  5989.             ctx++;

  5990.     }

  5991. 

  5992.     if( mvd >= 9 ) {

  5993.         int k = 3;

  5994.         while( get_cabac_bypass( &h->cabac ) ) {

  5995.             mvd += 1 << k;

  5996.             k++;

  5997.         }

  5998.         while( k-- ) {

  5999.             if( get_cabac_bypass( &h->cabac ) )

  6000.                 mvd += 1 << k;

  6001.         }

  6002.     }

  6003.     if( get_cabac_bypass( &h->cabac ) )  return -mvd;

  6004.     else                                 return  mvd;

  6005. }

  6006. 

  6007. static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {

  6008.     int nza, nzb;

  6009.     int ctx = 0;

  6010. 

  6011.     if( cat == 0 ) {

  6012.         nza = h->left_cbp&0x100;

  6013.         nzb = h-> top_cbp&0x100;

  6014.     } else if( cat == 1 || cat == 2 ) {

  6015.         nza = h->non_zero_count_cache[scan8[idx] - 1];

  6016.         nzb = h->non_zero_count_cache[scan8[idx] - 8];

  6017.     } else if( cat == 3 ) {

  6018.         nza = (h->left_cbp>>(6+idx))&0x01;

  6019.         nzb = (h-> top_cbp>>(6+idx))&0x01;

  6020.     } else {

  6021.         assert(cat == 4);

  6022.         nza = h->non_zero_count_cache[scan8[16+idx] - 1];

  6023.         nzb = h->non_zero_count_cache[scan8[16+idx] - 8];

  6024.     }

  6025. 

  6026.     if( nza > 0 )

  6027.         ctx++;

  6028. 

  6029.     if( nzb > 0 )

  6030.         ctx += 2;

  6031. 

  6032.     return ctx + 4 * cat;

  6033. }

  6034. 

  6035. static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, const uint32_t *qmul, int max_coeff) {

  6036.     const int mb_xy  = h->s.mb_x + h->s.mb_y*h->s.mb_stride;

  6037.     static const int significant_coeff_flag_offset[2][6] = {

  6038.       { 105+0, 105+15, 105+29, 105+44, 105+47, 402 },

  6039.       { 277+0, 277+15, 277+29, 277+44, 277+47, 436 }

  6040.     };

  6041.     static const int last_coeff_flag_offset[2][6] = {

  6042.       { 166+0, 166+15, 166+29, 166+44, 166+47, 417 },

  6043.       { 338+0, 338+15, 338+29, 338+44, 338+47, 451 }

  6044.     };

  6045.     static const int coeff_abs_level_m1_offset[6] = {

  6046.         227+0, 227+10, 227+20, 227+30, 227+39, 426

  6047.     };

  6048.     static const int significant_coeff_flag_offset_8x8[2][63] = {

  6049.       { 0, 1, 2, 3, 4, 5, 5, 4, 4, 3, 3, 4, 4, 4, 5, 5,

  6050.         4, 4, 4, 4, 3, 3, 6, 7, 7, 7, 8, 9,10, 9, 8, 7,

  6051.         7, 6,11,12,13,11, 6, 7, 8, 9,14,10, 9, 8, 6,11,

  6052.        12,13,11, 6, 9,14,10, 9,11,12,13,11,14,10,12 },

  6053.       { 0, 1, 1, 2, 2, 3, 3, 4, 5, 6, 7, 7, 7, 8, 4, 5,

  6054.         6, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,11,12,11,

  6055.         9, 9,10,10, 8,11,12,11, 9, 9,10,10, 8,13,13, 9,

  6056.         9,10,10, 8,13,13, 9, 9,10,10,14,14,14,14,14 }

  6057.     };

  6058.     static const int last_coeff_flag_offset_8x8[63] = {

  6059.         0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,

  6060.         2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  6061.         3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,

  6062.         5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8

  6063.     };

  6064. 

  6065.     int index[64];

  6066. 

  6067.     int i, last;

  6068.     int coeff_count = 0;

  6069. 

  6070.     int abslevel1 = 1;

  6071.     int abslevelgt1 = 0;

  6072. 

  6073.     uint8_t *significant_coeff_ctx_base;

  6074.     uint8_t *last_coeff_ctx_base;

  6075.     uint8_t *abs_level_m1_ctx_base;

  6076. 

  6077.     /* cat: 0-> DC 16x16  n = 0

  6078.      *      1-> AC 16x16  n = luma4x4idx

  6079.      *      2-> Luma4x4   n = luma4x4idx

  6080.      *      3-> DC Chroma n = iCbCr

  6081.      *      4-> AC Chroma n = 4 * iCbCr + chroma4x4idx

  6082.      *      5-> Luma8x8   n = 4 * luma8x8idx

  6083.      */

  6084. 

  6085.     /* read coded block flag */

  6086.     if( cat != 5 ) {

  6087.         if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {

  6088.             if( cat == 1 || cat == 2 )

  6089.                 h->non_zero_count_cache[scan8[n]] = 0;

  6090.             else if( cat == 4 )

  6091.                 h->non_zero_count_cache[scan8[16+n]] = 0;

  6092. 

  6093.             return 0;

  6094.         }

  6095.     }

  6096. 

  6097.     significant_coeff_ctx_base = h->cabac_state

  6098.         + significant_coeff_flag_offset[MB_FIELD][cat];

  6099.     last_coeff_ctx_base = h->cabac_state

  6100.         + last_coeff_flag_offset[MB_FIELD][cat];

  6101.     abs_level_m1_ctx_base = h->cabac_state

  6102.         + coeff_abs_level_m1_offset[cat];

  6103. 

  6104.     if( cat == 5 ) {

  6105. #define DECODE_SIGNIFICANCE( coefs, sig_off, last_off ) \

  6106.         for(last= 0; last < coefs; last++) { \

  6107.             uint8_t *sig_ctx = significant_coeff_ctx_base + sig_off; \

  6108.             if( get_cabac( &h->cabac, sig_ctx )) { \

  6109.                 uint8_t *last_ctx = last_coeff_ctx_base + last_off; \

  6110.                 index[coeff_count++] = last; \

  6111.                 if( get_cabac( &h->cabac, last_ctx ) ) { \

  6112.                     last= max_coeff; \

  6113.                     break; \

  6114.                 } \

  6115.             } \

  6116.         }

  6117.         const int *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD];

  6118.         DECODE_SIGNIFICANCE( 63, sig_off[last], last_coeff_flag_offset_8x8[last] );

  6119.     } else {

  6120.         DECODE_SIGNIFICANCE( max_coeff - 1, last, last );

  6121.     }

  6122.     if( last == max_coeff -1 ) {

  6123.         index[coeff_count++] = last;

  6124.     }

  6125.     assert(coeff_count > 0);

  6126. 

  6127.     if( cat == 0 )

  6128.         h->cbp_table[mb_xy] |= 0x100;

  6129.     else if( cat == 1 || cat == 2 )

  6130.         h->non_zero_count_cache[scan8[n]] = coeff_count;

  6131.     else if( cat == 3 )

  6132.         h->cbp_table[mb_xy] |= 0x40 << n;

  6133.     else if( cat == 4 )

  6134.         h->non_zero_count_cache[scan8[16+n]] = coeff_count;

  6135.     else {

  6136.         assert( cat == 5 );

  6137.         fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, coeff_count, 1);

  6138.     }

  6139. 

  6140.     for( i = coeff_count - 1; i >= 0; i-- ) {

  6141.         uint8_t *ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + abs_level_m1_ctx_base;

  6142.         int j= scantable[index[i]];

  6143. 

  6144.         if( get_cabac( &h->cabac, ctx ) == 0 ) {

  6145.             if( !qmul ) {

  6146.                 if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;

  6147.                 else                                block[j] =  1;

  6148.             }else{

  6149.                 if( get_cabac_bypass( &h->cabac ) ) block[j] = (-qmul[j] + 32) >> 6;

  6150.                 else                                block[j] = ( qmul[j] + 32) >> 6;

  6151.             }

  6152. 

  6153.             abslevel1++;

  6154.         } else {

  6155.             int coeff_abs = 2;

  6156.             ctx = 5 + FFMIN( 4, abslevelgt1 ) + abs_level_m1_ctx_base;

  6157.             while( coeff_abs < 15 && get_cabac( &h->cabac, ctx ) ) {

  6158.                 coeff_abs++;

  6159.             }

  6160. 

  6161.             if( coeff_abs >= 15 ) {

  6162.                 int j = 0;

  6163.                 while( get_cabac_bypass( &h->cabac ) ) {

  6164.                     coeff_abs += 1 << j;

  6165.                     j++;

  6166.                 }

  6167. 

  6168.                 while( j-- ) {

  6169.                     if( get_cabac_bypass( &h->cabac ) )

  6170.                         coeff_abs += 1 << j ;

  6171.                 }

  6172.             }

  6173. 

  6174.             if( !qmul ) {

  6175.                 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;

  6176.                 else                                block[j] =  coeff_abs;

  6177.             }else{

  6178.                 if( get_cabac_bypass( &h->cabac ) ) block[j] = (-coeff_abs * qmul[j] + 32) >> 6;

  6179.                 else                                block[j] = ( coeff_abs * qmul[j] + 32) >> 6;

  6180.             }

  6181. 

  6182.             abslevelgt1++;

  6183.         }

  6184.     }

  6185.     return 0;

  6186. }

  6187. 

  6188. static void inline compute_mb_neighbors(H264Context *h)

  6189. {

  6190.     MpegEncContext * const s = &h->s;

  6191.     const int mb_xy  = s->mb_x + s->mb_y*s->mb_stride;

  6192.     h->top_mb_xy     = mb_xy - s->mb_stride;

  6193.     h->left_mb_xy[0] = mb_xy - 1;

  6194.     if(FRAME_MBAFF){

  6195.         const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;

  6196.         const int top_pair_xy      = pair_xy     - s->mb_stride;

  6197.         const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);

  6198.         const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);

  6199.         const int curr_mb_frame_flag = !MB_FIELD;

  6200.         const int bottom = (s->mb_y & 1);

  6201.         if (bottom

  6202.                 ? !curr_mb_frame_flag // bottom macroblock

  6203.                 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock

  6204.                 ) {

  6205.             h->top_mb_xy -= s->mb_stride;

  6206.         }

  6207.         if (left_mb_frame_flag != curr_mb_frame_flag) {

  6208.             h->left_mb_xy[0] = pair_xy - 1;

  6209.         }

  6210.     }

  6211.     return;

  6212. }

  6213. 

  6214. /**

  6215.  * decodes a macroblock

  6216.  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed

  6217.  */

  6218. static int decode_mb_cabac(H264Context *h) {

  6219.     MpegEncContext * const s = &h->s;

  6220.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  6221.     int mb_type, partition_count, cbp = 0;

  6222.     int dct8x8_allowed= h->pps.transform_8x8_mode;

  6223. 

  6224.     s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?)

  6225. 

  6226.     tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);

  6227.     if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {

  6228.         int skip;

  6229.         /* a skipped mb needs the aff flag from the following mb */

  6230.         if( FRAME_MBAFF && s->mb_x==0 && (s->mb_y&1)==0 )

  6231.             predict_field_decoding_flag(h);

  6232.         if( FRAME_MBAFF && (s->mb_y&1)==1 && h->prev_mb_skipped )

  6233.             skip = h->next_mb_skipped;

  6234.         else

  6235.             skip = decode_cabac_mb_skip( h, s->mb_x, s->mb_y );

  6236.         /* read skip flags */

  6237.         if( skip ) {

  6238.             if( FRAME_MBAFF && (s->mb_y&1)==0 ){

  6239.                 s->current_picture.mb_type[mb_xy] = MB_TYPE_SKIP;

  6240.                 h->next_mb_skipped = decode_cabac_mb_skip( h, s->mb_x, s->mb_y+1 );

  6241.                 if(h->next_mb_skipped)

  6242.                     predict_field_decoding_flag(h);

  6243.                 else

  6244.                     h->mb_mbaff = h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);

  6245.             }

  6246. 

  6247.             decode_mb_skip(h);

  6248. 

  6249.             h->cbp_table[mb_xy] = 0;

  6250.             h->chroma_pred_mode_table[mb_xy] = 0;

  6251.             h->last_qscale_diff = 0;

  6252. 

  6253.             return 0;

  6254. 

  6255.         }

  6256.     }

  6257.     if(FRAME_MBAFF){

  6258.         if( (s->mb_y&1) == 0 )

  6259.             h->mb_mbaff =

  6260.             h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);

  6261.     }else

  6262.         h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);

  6263. 

  6264.     h->prev_mb_skipped = 0;

  6265. 

  6266.     compute_mb_neighbors(h);

  6267.     if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {

  6268.         av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );

  6269.         return -1;

  6270.     }

  6271. 

  6272.     if( h->slice_type == B_TYPE ) {

  6273.         if( mb_type < 23 ){

  6274.             partition_count= b_mb_type_info[mb_type].partition_count;

  6275.             mb_type=         b_mb_type_info[mb_type].type;

  6276.         }else{

  6277.             mb_type -= 23;

  6278.             goto decode_intra_mb;

  6279.         }

  6280.     } else if( h->slice_type == P_TYPE ) {

  6281.         if( mb_type < 5) {

  6282.             partition_count= p_mb_type_info[mb_type].partition_count;

  6283.             mb_type=         p_mb_type_info[mb_type].type;

  6284.         } else {

  6285.             mb_type -= 5;

  6286.             goto decode_intra_mb;

  6287.         }

  6288.     } else {

  6289.        assert(h->slice_type == I_TYPE);

  6290. decode_intra_mb:

  6291.         partition_count = 0;

  6292.         cbp= i_mb_type_info[mb_type].cbp;

  6293.         h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;

  6294.         mb_type= i_mb_type_info[mb_type].type;

  6295.     }

  6296.     if(MB_FIELD)

  6297.         mb_type |= MB_TYPE_INTERLACED;

  6298. 

  6299.     h->slice_table[ mb_xy ]= h->slice_num;

  6300. 

  6301.     if(IS_INTRA_PCM(mb_type)) {

  6302.         const uint8_t *ptr;

  6303.         unsigned int x, y;

  6304. 

  6305.         // We assume these blocks are very rare so we dont optimize it.

  6306.         // FIXME The two following lines get the bitstream position in the cabac

  6307.         // decode, I think it should be done by a function in cabac.h (or cabac.c).

  6308.         ptr= h->cabac.bytestream;

  6309.         if (h->cabac.low&0x1) ptr-=CABAC_BITS/8;

  6310. 

  6311.         // The pixels are stored in the same order as levels in h->mb array.

  6312.         for(y=0; y<16; y++){

  6313.             const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);

  6314.             for(x=0; x<16; x++){

  6315.                 tprintf("LUMA ICPM LEVEL (%3d)\n", *ptr);

  6316.                 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++;

  6317.             }

  6318.         }

  6319.         for(y=0; y<8; y++){

  6320.             const int index= 256 + 4*(y&3) + 32*(y>>2);

  6321.             for(x=0; x<8; x++){

  6322.                 tprintf("CHROMA U ICPM LEVEL (%3d)\n", *ptr);

  6323.                 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;

  6324.             }

  6325.         }

  6326.         for(y=0; y<8; y++){

  6327.             const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);

  6328.             for(x=0; x<8; x++){

  6329.                 tprintf("CHROMA V ICPM LEVEL (%3d)\n", *ptr);

  6330.                 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;

  6331.             }

  6332.         }

  6333. 

  6334.         ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);

  6335. 

  6336.         // All blocks are present

  6337.         h->cbp_table[mb_xy] = 0x1ef;

  6338.         h->chroma_pred_mode_table[mb_xy] = 0;

  6339.         // In deblocking, the quantizer is 0

  6340.         s->current_picture.qscale_table[mb_xy]= 0;

  6341.         h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);

  6342.         // All coeffs are present

  6343.         memset(h->non_zero_count[mb_xy], 16, 16);

  6344.         s->current_picture.mb_type[mb_xy]= mb_type;

  6345.         return 0;

  6346.     }

  6347. 

  6348.     if(MB_MBAFF){

  6349.         h->ref_count[0] <<= 1;

  6350.         h->ref_count[1] <<= 1;

  6351.     }

  6352. 

  6353.     fill_caches(h, mb_type, 0);

  6354. 

  6355.     if( IS_INTRA( mb_type ) ) {

  6356.         int i;

  6357.         if( IS_INTRA4x4( mb_type ) ) {

  6358.             if( dct8x8_allowed && decode_cabac_mb_transform_size( h ) ) {

  6359.                 mb_type |= MB_TYPE_8x8DCT;

  6360.                 for( i = 0; i < 16; i+=4 ) {

  6361.                     int pred = pred_intra_mode( h, i );

  6362.                     int mode = decode_cabac_mb_intra4x4_pred_mode( h, pred );

  6363.                     fill_rectangle( &h->intra4x4_pred_mode_cache[ scan8[i] ], 2, 2, 8, mode, 1 );

  6364.                 }

  6365.             } else {

  6366.                 for( i = 0; i < 16; i++ ) {

  6367.                     int pred = pred_intra_mode( h, i );

  6368.                     h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );

  6369. 

  6370.                 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );

  6371.                 }

  6372.             }

  6373.             write_back_intra_pred_mode(h);

  6374.             if( check_intra4x4_pred_mode(h) < 0 ) return -1;

  6375.         } else {

  6376.             h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );

  6377.             if( h->intra16x16_pred_mode < 0 ) return -1;

  6378.         }

  6379.         h->chroma_pred_mode_table[mb_xy] =

  6380.             h->chroma_pred_mode          = decode_cabac_mb_chroma_pre_mode( h );

  6381. 

  6382.         h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );

  6383.         if( h->chroma_pred_mode < 0 ) return -1;

  6384.     } else if( partition_count == 4 ) {

  6385.         int i, j, sub_partition_count[4], list, ref[2][4];

  6386. 

  6387.         if( h->slice_type == B_TYPE ) {

  6388.             for( i = 0; i < 4; i++ ) {

  6389.                 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );

  6390.                 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;

  6391.                 h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;

  6392.             }

  6393.             if( IS_DIRECT(h->sub_mb_type[0] | h->sub_mb_type[1] |

  6394.                           h->sub_mb_type[2] | h->sub_mb_type[3]) ) {

  6395.                 pred_direct_motion(h, &mb_type);

  6396.                 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {

  6397.                     for( i = 0; i < 4; i++ )

  6398.                         if( IS_DIRECT(h->sub_mb_type[i]) )

  6399.                             fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );

  6400.                 }

  6401.             }

  6402.         } else {

  6403.             for( i = 0; i < 4; i++ ) {

  6404.                 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );

  6405.                 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;

  6406.                 h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;

  6407.             }

  6408.         }

  6409. 

  6410.         for( list = 0; list < 2; list++ ) {

  6411.             if( h->ref_count[list] > 0 ) {

  6412.                 for( i = 0; i < 4; i++ ) {

  6413.                     if(IS_DIRECT(h->sub_mb_type[i])) continue;

  6414.                     if(IS_DIR(h->sub_mb_type[i], 0, list)){

  6415.                         if( h->ref_count[list] > 1 )

  6416.                             ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );

  6417.                         else

  6418.                             ref[list][i] = 0;

  6419.                     } else {

  6420.                         ref[list][i] = -1;

  6421.                     }

  6422.                                                        h->ref_cache[list][ scan8[4*i]+1 ]=

  6423.                     h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];

  6424.                 }

  6425.             }

  6426.         }

  6427. 

  6428.         if(dct8x8_allowed)

  6429.             dct8x8_allowed = get_dct8x8_allowed(h);

  6430. 

  6431.         for(list=0; list<2; list++){

  6432.             for(i=0; i<4; i++){

  6433.                 if(IS_DIRECT(h->sub_mb_type[i])){

  6434.                     fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);

  6435.                     continue;

  6436.                 }

  6437.                 h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ];

  6438. 

  6439.                 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){

  6440.                     const int sub_mb_type= h->sub_mb_type[i];

  6441.                     const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;

  6442.                     for(j=0; j<sub_partition_count[i]; j++){

  6443.                         int mpx, mpy;

  6444.                         int mx, my;

  6445.                         const int index= 4*i + block_width*j;

  6446.                         int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];

  6447.                         int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];

  6448.                         pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);

  6449. 

  6450.                         mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );

  6451.                         my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );

  6452.                         tprintf("final mv:%d %d\n", mx, my);

  6453. 

  6454.                         if(IS_SUB_8X8(sub_mb_type)){

  6455.                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=

  6456.                             mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;

  6457.                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=

  6458.                             mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;

  6459. 

  6460.                             mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=

  6461.                             mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;

  6462.                             mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=

  6463.                             mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;

  6464.                         }else if(IS_SUB_8X4(sub_mb_type)){

  6465.                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;

  6466.                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;

  6467. 

  6468.                             mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;

  6469.                             mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;

  6470.                         }else if(IS_SUB_4X8(sub_mb_type)){

  6471.                             mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;

  6472.                             mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;

  6473. 

  6474.                             mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;

  6475.                             mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;

  6476.                         }else{

  6477.                             assert(IS_SUB_4X4(sub_mb_type));

  6478.                             mv_cache[ 0 ][0]= mx;

  6479.                             mv_cache[ 0 ][1]= my;

  6480. 

  6481.                             mvd_cache[ 0 ][0]= mx - mpx;

  6482.                             mvd_cache[ 0 ][1]= my - mpy;

  6483.                         }

  6484.                     }

  6485.                 }else{

  6486.                     uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];

  6487.                     uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];

  6488.                     p[0] = p[1] = p[8] = p[9] = 0;

  6489.                     pd[0]= pd[1]= pd[8]= pd[9]= 0;

  6490.                 }

  6491.             }

  6492.         }

  6493.     } else if( IS_DIRECT(mb_type) ) {

  6494.         pred_direct_motion(h, &mb_type);

  6495.         fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);

  6496.         fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);

  6497.         dct8x8_allowed &= h->sps.direct_8x8_inference_flag;

  6498.     } else {

  6499.         int list, mx, my, i, mpx, mpy;

  6500.         if(IS_16X16(mb_type)){

  6501.             for(list=0; list<2; list++){

  6502.                 if(IS_DIR(mb_type, 0, list)){

  6503.                     if(h->ref_count[list] > 0 ){

  6504.                         const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;

  6505.                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);

  6506.                     }

  6507.                 }else

  6508.                     fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);

  6509.             }

  6510.             for(list=0; list<2; list++){

  6511.                 if(IS_DIR(mb_type, 0, list)){

  6512.                     pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);

  6513. 

  6514.                     mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );

  6515.                     my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );

  6516.                     tprintf("final mv:%d %d\n", mx, my);

  6517. 

  6518.                     fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);

  6519.                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);

  6520.                 }else

  6521.                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);

  6522.             }

  6523.         }

  6524.         else if(IS_16X8(mb_type)){

  6525.             for(list=0; list<2; list++){

  6526.                 if(h->ref_count[list]>0){

  6527.                     for(i=0; i<2; i++){

  6528.                         if(IS_DIR(mb_type, i, list)){

  6529.                             const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;

  6530.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);

  6531.                         }else

  6532.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);

  6533.                     }

  6534.                 }

  6535.             }

  6536.             for(list=0; list<2; list++){

  6537.                 for(i=0; i<2; i++){

  6538.                     if(IS_DIR(mb_type, i, list)){

  6539.                         pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);

  6540.                         mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );

  6541.                         my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );

  6542.                         tprintf("final mv:%d %d\n", mx, my);

  6543. 

  6544.                         fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);

  6545.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);

  6546.                     }else{

  6547.                         fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);

  6548.                         fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);

  6549.                     }

  6550.                 }

  6551.             }

  6552.         }else{

  6553.             assert(IS_8X16(mb_type));

  6554.             for(list=0; list<2; list++){

  6555.                 if(h->ref_count[list]>0){

  6556.                     for(i=0; i<2; i++){

  6557.                         if(IS_DIR(mb_type, i, list)){ //FIXME optimize

  6558.                             const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;

  6559.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);

  6560.                         }else

  6561.                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);

  6562.                     }

  6563.                 }

  6564.             }

  6565.             for(list=0; list<2; list++){

  6566.                 for(i=0; i<2; i++){

  6567.                     if(IS_DIR(mb_type, i, list)){

  6568.                         pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);

  6569.                         mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );

  6570.                         my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );

  6571. 

  6572.                         tprintf("final mv:%d %d\n", mx, my);

  6573.                         fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);

  6574.                         fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);

  6575.                     }else{

  6576.                         fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);

  6577.                         fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);

  6578.                     }

  6579.                 }

  6580.             }

  6581.         }

  6582.     }

  6583. 

  6584.    if( IS_INTER( mb_type ) ) {

  6585.         h->chroma_pred_mode_table[mb_xy] = 0;

  6586.         write_back_motion( h, mb_type );

  6587.    }

  6588. 

  6589.     if( !IS_INTRA16x16( mb_type ) ) {

  6590.         cbp  = decode_cabac_mb_cbp_luma( h );

  6591.         cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;

  6592.     }

  6593. 

  6594.     h->cbp_table[mb_xy] = h->cbp = cbp;

  6595. 

  6596.     if( dct8x8_allowed && (cbp&15) && !IS_INTRA( mb_type ) ) {

  6597.         if( decode_cabac_mb_transform_size( h ) )

  6598.             mb_type |= MB_TYPE_8x8DCT;

  6599.     }

  6600.     s->current_picture.mb_type[mb_xy]= mb_type;

  6601. 

  6602.     if( cbp || IS_INTRA16x16( mb_type ) ) {

  6603.         const uint8_t *scan, *scan8x8, *dc_scan;

  6604.         int dqp;

  6605. 

  6606.         if(IS_INTERLACED(mb_type)){

  6607.             scan8x8= s->qscale ? h->field_scan8x8 : h->field_scan8x8_q0;

  6608.             scan= s->qscale ? h->field_scan : h->field_scan_q0;

  6609.             dc_scan= luma_dc_field_scan;

  6610.         }else{

  6611.             scan8x8= s->qscale ? h->zigzag_scan8x8 : h->zigzag_scan8x8_q0;

  6612.             scan= s->qscale ? h->zigzag_scan : h->zigzag_scan_q0;

  6613.             dc_scan= luma_dc_zigzag_scan;

  6614.         }

  6615. 

  6616.         h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );

  6617.         if( dqp == INT_MIN ){

  6618.             av_log(h->s.avctx, AV_LOG_ERROR, "cabac decode of qscale diff failed at %d %d\n", s->mb_x, s->mb_y);

  6619.             return -1;

  6620.         }

  6621.         s->qscale += dqp;

  6622.         if(((unsigned)s->qscale) > 51){

  6623.             if(s->qscale<0) s->qscale+= 52;

  6624.             else            s->qscale-= 52;

  6625.         }

  6626.         h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);

  6627. 

  6628.         if( IS_INTRA16x16( mb_type ) ) {

  6629.             int i;

  6630.             //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );

  6631.             if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, NULL, 16) < 0)

  6632.                 return -1;

  6633.             if( cbp&15 ) {

  6634.                 for( i = 0; i < 16; i++ ) {

  6635.                     //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );

  6636.                     if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, h->dequant4_coeff[0][s->qscale], 15) < 0 )

  6637.                         return -1;

  6638.                 }

  6639.             } else {

  6640.                 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);

  6641.             }

  6642.         } else {

  6643.             int i8x8, i4x4;

  6644.             for( i8x8 = 0; i8x8 < 4; i8x8++ ) {

  6645.                 if( cbp & (1<<i8x8) ) {

  6646.                     if( IS_8x8DCT(mb_type) ) {

  6647.                         if( decode_cabac_residual(h, h->mb + 64*i8x8, 5, 4*i8x8,

  6648.                             scan8x8, h->dequant8_coeff[IS_INTRA( mb_type ) ? 0:1][s->qscale], 64) < 0 )

  6649.                             return -1;

  6650.                     } else

  6651.                     for( i4x4 = 0; i4x4 < 4; i4x4++ ) {

  6652.                         const int index = 4*i8x8 + i4x4;

  6653.                         //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );

  6654.                         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 )

  6655.                             return -1;

  6656.                     }

  6657.                 } else {

  6658.                     uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];

  6659.                     nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;

  6660.                 }

  6661.             }

  6662.         }

  6663. 

  6664.         if( cbp&0x30 ){

  6665.             int c;

  6666.             for( c = 0; c < 2; c++ ) {

  6667.                 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );

  6668.                 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, NULL, 4) < 0)

  6669.                     return -1;

  6670.             }

  6671.         }

  6672. 

  6673.         if( cbp&0x20 ) {

  6674.             int c, i;

  6675.             for( c = 0; c < 2; c++ ) {

  6676.                 for( i = 0; i < 4; i++ ) {

  6677.                     const int index = 16 + 4 * c + i;

  6678.                     //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );

  6679.                     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)

  6680.                         return -1;

  6681.                 }

  6682.             }

  6683.         } else {

  6684.             uint8_t * const nnz= &h->non_zero_count_cache[0];

  6685.             nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =

  6686.             nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;

  6687.         }

  6688.     } else {

  6689.         uint8_t * const nnz= &h->non_zero_count_cache[0];

  6690.         fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);

  6691.         nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =

  6692.         nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;

  6693.         h->last_qscale_diff = 0;

  6694.     }

  6695. 

  6696.     s->current_picture.qscale_table[mb_xy]= s->qscale;

  6697.     write_back_non_zero_count(h);

  6698. 

  6699.     if(MB_MBAFF){

  6700.         h->ref_count[0] >>= 1;

  6701.         h->ref_count[1] >>= 1;

  6702.     }

  6703. 

  6704.     return 0;

  6705. }

  6706. 

  6707. 

  6708. static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {

  6709.     int i, d;

  6710.     const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );

  6711.     const int alpha = alpha_table[index_a];

  6712.     const int beta  = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];

  6713. 

  6714.     if( bS[0] < 4 ) {

  6715.         int8_t tc[4];

  6716.         for(i=0; i<4; i++)

  6717.             tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] : -1;

  6718.         h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);

  6719.     } else {

  6720.         /* 16px edge length, because bS=4 is triggered by being at

  6721.          * the edge of an intra MB, so all 4 bS are the same */

  6722.             for( d = 0; d < 16; d++ ) {

  6723.                 const int p0 = pix[-1];

  6724.                 const int p1 = pix[-2];

  6725.                 const int p2 = pix[-3];

  6726. 

  6727.                 const int q0 = pix[0];

  6728.                 const int q1 = pix[1];

  6729.                 const int q2 = pix[2];

  6730. 

  6731.                 if( ABS( p0 - q0 ) < alpha &&

  6732.                     ABS( p1 - p0 ) < beta &&

  6733.                     ABS( q1 - q0 ) < beta ) {

  6734. 

  6735.                     if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){

  6736.                         if( ABS( p2 - p0 ) < beta)

  6737.                         {

  6738.                             const int p3 = pix[-4];

  6739.                             /* p0', p1', p2' */

  6740.                             pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;

  6741.                             pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;

  6742.                             pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;

  6743.                         } else {

  6744.                             /* p0' */

  6745.                             pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6746.                         }

  6747.                         if( ABS( q2 - q0 ) < beta)

  6748.                         {

  6749.                             const int q3 = pix[3];

  6750.                             /* q0', q1', q2' */

  6751.                             pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;

  6752.                             pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;

  6753.                             pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;

  6754.                         } else {

  6755.                             /* q0' */

  6756.                             pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6757.                         }

  6758.                     }else{

  6759.                         /* p0', q0' */

  6760.                         pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6761.                         pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6762.                     }

  6763.                     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]);

  6764.                 }

  6765.                 pix += stride;

  6766.             }

  6767.     }

  6768. }

  6769. static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {

  6770.     int i;

  6771.     const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );

  6772.     const int alpha = alpha_table[index_a];

  6773.     const int beta  = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];

  6774. 

  6775.     if( bS[0] < 4 ) {

  6776.         int8_t tc[4];

  6777.         for(i=0; i<4; i++)

  6778.             tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] + 1 : 0;

  6779.         h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);

  6780.     } else {

  6781.         h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);

  6782.     }

  6783. }

  6784. 

  6785. static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) {

  6786.     int i;

  6787.     for( i = 0; i < 16; i++, pix += stride) {

  6788.         int index_a;

  6789.         int alpha;

  6790.         int beta;

  6791. 

  6792.         int qp_index;

  6793.         int bS_index = (i >> 1);

  6794.         if (!MB_FIELD) {

  6795.             bS_index &= ~1;

  6796.             bS_index |= (i & 1);

  6797.         }

  6798. 

  6799.         if( bS[bS_index] == 0 ) {

  6800.             continue;

  6801.         }

  6802. 

  6803.         qp_index = MB_FIELD ? (i >> 3) : (i & 1);

  6804.         index_a = clip( qp[qp_index] + h->slice_alpha_c0_offset, 0, 51 );

  6805.         alpha = alpha_table[index_a];

  6806.         beta  = beta_table[clip( qp[qp_index] + h->slice_beta_offset, 0, 51 )];

  6807. 

  6808.         if( bS[bS_index] < 4 ) {

  6809.             const int tc0 = tc0_table[index_a][bS[bS_index] - 1];

  6810.             const int p0 = pix[-1];

  6811.             const int p1 = pix[-2];

  6812.             const int p2 = pix[-3];

  6813.             const int q0 = pix[0];

  6814.             const int q1 = pix[1];

  6815.             const int q2 = pix[2];

  6816. 

  6817.             if( ABS( p0 - q0 ) < alpha &&

  6818.                 ABS( p1 - p0 ) < beta &&

  6819.                 ABS( q1 - q0 ) < beta ) {

  6820.                 int tc = tc0;

  6821.                 int i_delta;

  6822. 

  6823.                 if( ABS( p2 - p0 ) < beta ) {

  6824.                     pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );

  6825.                     tc++;

  6826.                 }

  6827.                 if( ABS( q2 - q0 ) < beta ) {

  6828.                     pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );

  6829.                     tc++;

  6830.                 }

  6831. 

  6832.                 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );

  6833.                 pix[-1] = clip_uint8( p0 + i_delta );    /* p0' */

  6834.                 pix[0]  = clip_uint8( q0 - i_delta );    /* q0' */

  6835.                 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);

  6836.             }

  6837.         }else{

  6838.             const int p0 = pix[-1];

  6839.             const int p1 = pix[-2];

  6840.             const int p2 = pix[-3];

  6841. 

  6842.             const int q0 = pix[0];

  6843.             const int q1 = pix[1];

  6844.             const int q2 = pix[2];

  6845. 

  6846.             if( ABS( p0 - q0 ) < alpha &&

  6847.                 ABS( p1 - p0 ) < beta &&

  6848.                 ABS( q1 - q0 ) < beta ) {

  6849. 

  6850.                 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){

  6851.                     if( ABS( p2 - p0 ) < beta)

  6852.                     {

  6853.                         const int p3 = pix[-4];

  6854.                         /* p0', p1', p2' */

  6855.                         pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;

  6856.                         pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;

  6857.                         pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;

  6858.                     } else {

  6859.                         /* p0' */

  6860.                         pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6861.                     }

  6862.                     if( ABS( q2 - q0 ) < beta)

  6863.                     {

  6864.                         const int q3 = pix[3];

  6865.                         /* q0', q1', q2' */

  6866.                         pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;

  6867.                         pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;

  6868.                         pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;

  6869.                     } else {

  6870.                         /* q0' */

  6871.                         pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6872.                     }

  6873.                 }else{

  6874.                     /* p0', q0' */

  6875.                     pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6876.                     pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6877.                 }

  6878.                 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]);

  6879.             }

  6880.         }

  6881.     }

  6882. }

  6883. static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[8], int qp[2] ) {

  6884.     int i;

  6885.     for( i = 0; i < 8; i++, pix += stride) {

  6886.         int index_a;

  6887.         int alpha;

  6888.         int beta;

  6889. 

  6890.         int qp_index;

  6891.         int bS_index = i;

  6892. 

  6893.         if( bS[bS_index] == 0 ) {

  6894.             continue;

  6895.         }

  6896. 

  6897.         qp_index = MB_FIELD ? (i >> 2) : (i & 1);

  6898.         index_a = clip( qp[qp_index] + h->slice_alpha_c0_offset, 0, 51 );

  6899.         alpha = alpha_table[index_a];

  6900.         beta  = beta_table[clip( qp[qp_index] + h->slice_beta_offset, 0, 51 )];

  6901. 

  6902.         if( bS[bS_index] < 4 ) {

  6903.             const int tc = tc0_table[index_a][bS[bS_index] - 1] + 1;

  6904.             const int p0 = pix[-1];

  6905.             const int p1 = pix[-2];

  6906.             const int q0 = pix[0];

  6907.             const int q1 = pix[1];

  6908. 

  6909.             if( ABS( p0 - q0 ) < alpha &&

  6910.                 ABS( p1 - p0 ) < beta &&

  6911.                 ABS( q1 - q0 ) < beta ) {

  6912.                 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );

  6913. 

  6914.                 pix[-1] = clip_uint8( p0 + i_delta );    /* p0' */

  6915.                 pix[0]  = clip_uint8( q0 - i_delta );    /* q0' */

  6916.                 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);

  6917.             }

  6918.         }else{

  6919.             const int p0 = pix[-1];

  6920.             const int p1 = pix[-2];

  6921.             const int q0 = pix[0];

  6922.             const int q1 = pix[1];

  6923. 

  6924.             if( ABS( p0 - q0 ) < alpha &&

  6925.                 ABS( p1 - p0 ) < beta &&

  6926.                 ABS( q1 - q0 ) < beta ) {

  6927. 

  6928.                 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;   /* p0' */

  6929.                 pix[0]  = ( 2*q1 + q0 + p1 + 2 ) >> 2;   /* q0' */

  6930.                 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]);

  6931.             }

  6932.         }

  6933.     }

  6934. }

  6935. 

  6936. static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {

  6937.     int i, d;

  6938.     const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );

  6939.     const int alpha = alpha_table[index_a];

  6940.     const int beta  = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];

  6941.     const int pix_next  = stride;

  6942. 

  6943.     if( bS[0] < 4 ) {

  6944.         int8_t tc[4];

  6945.         for(i=0; i<4; i++)

  6946.             tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] : -1;

  6947.         h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);

  6948.     } else {

  6949.         /* 16px edge length, see filter_mb_edgev */

  6950.             for( d = 0; d < 16; d++ ) {

  6951.                 const int p0 = pix[-1*pix_next];

  6952.                 const int p1 = pix[-2*pix_next];

  6953.                 const int p2 = pix[-3*pix_next];

  6954.                 const int q0 = pix[0];

  6955.                 const int q1 = pix[1*pix_next];

  6956.                 const int q2 = pix[2*pix_next];

  6957. 

  6958.                 if( ABS( p0 - q0 ) < alpha &&

  6959.                     ABS( p1 - p0 ) < beta &&

  6960.                     ABS( q1 - q0 ) < beta ) {

  6961. 

  6962.                     const int p3 = pix[-4*pix_next];

  6963.                     const int q3 = pix[ 3*pix_next];

  6964. 

  6965.                     if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){

  6966.                         if( ABS( p2 - p0 ) < beta) {

  6967.                             /* p0', p1', p2' */

  6968.                             pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;

  6969.                             pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;

  6970.                             pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;

  6971.                         } else {

  6972.                             /* p0' */

  6973.                             pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6974.                         }

  6975.                         if( ABS( q2 - q0 ) < beta) {

  6976.                             /* q0', q1', q2' */

  6977.                             pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;

  6978.                             pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;

  6979.                             pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;

  6980.                         } else {

  6981.                             /* q0' */

  6982.                             pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6983.                         }

  6984.                     }else{

  6985.                         /* p0', q0' */

  6986.                         pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;

  6987.                         pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;

  6988.                     }

  6989.                     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]);

  6990.                 }

  6991.                 pix++;

  6992.             }

  6993.     }

  6994. }

  6995. 

  6996. static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {

  6997.     int i;

  6998.     const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );

  6999.     const int alpha = alpha_table[index_a];

  7000.     const int beta  = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];

  7001. 

  7002.     if( bS[0] < 4 ) {

  7003.         int8_t tc[4];

  7004.         for(i=0; i<4; i++)

  7005.             tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] + 1 : 0;

  7006.         h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);

  7007.     } else {

  7008.         h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);

  7009.     }

  7010. }

  7011. 

  7012. 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) {

  7013.     MpegEncContext * const s = &h->s;

  7014.     int mb_xy, mb_type;

  7015.     int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;

  7016. 

  7017.     if(mb_x==0 || mb_y==0 || !s->dsp.h264_loop_filter_strength) {

  7018.         filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);

  7019.         return;

  7020.     }

  7021.     assert(!FRAME_MBAFF);

  7022. 

  7023.     mb_xy = mb_x + mb_y*s->mb_stride;

  7024.     mb_type = s->current_picture.mb_type[mb_xy];

  7025.     qp = s->current_picture.qscale_table[mb_xy];

  7026.     qp0 = s->current_picture.qscale_table[mb_xy-1];

  7027.     qp1 = s->current_picture.qscale_table[h->top_mb_xy];

  7028.     qpc = get_chroma_qp( h->pps.chroma_qp_index_offset, qp );

  7029.     qpc0 = get_chroma_qp( h->pps.chroma_qp_index_offset, qp0 );

  7030.     qpc1 = get_chroma_qp( h->pps.chroma_qp_index_offset, qp1 );

  7031.     qp0 = (qp + qp0 + 1) >> 1;

  7032.     qp1 = (qp + qp1 + 1) >> 1;

  7033.     qpc0 = (qpc + qpc0 + 1) >> 1;

  7034.     qpc1 = (qpc + qpc1 + 1) >> 1;

  7035.     qp_thresh = 15 - h->slice_alpha_c0_offset;

  7036.     if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&

  7037.        qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)

  7038.         return;

  7039. 

  7040.     if( IS_INTRA(mb_type) ) {

  7041.         int16_t bS4[4] = {4,4,4,4};

  7042.         int16_t bS3[4] = {3,3,3,3};

  7043.         if( IS_8x8DCT(mb_type) ) {

  7044.             filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );

  7045.             filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );

  7046.             filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bS4, qp1 );

  7047.             filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );

  7048.         } else {

  7049.             filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );

  7050.             filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp );

  7051.             filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );

  7052.             filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp );

  7053.             filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bS4, qp1 );

  7054.             filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp );

  7055.             filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );

  7056.             filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp );

  7057.         }

  7058.         filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 );

  7059.         filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc );

  7060.         filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 );

  7061.         filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc );

  7062.         filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bS4, qpc1 );

  7063.         filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc );

  7064.         filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bS4, qpc1 );

  7065.         filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc );

  7066.         return;

  7067.     } else {

  7068.         DECLARE_ALIGNED_8(int16_t, bS[2][4][4]);

  7069.         uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;

  7070.         int edges;

  7071.         if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {

  7072.             edges = 4;

  7073.             bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;

  7074.         } else {

  7075.             int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 :

  7076.                              (mb_type & MB_TYPE_16x8) ? 1 : 0;

  7077.             int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16))

  7078.                              && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16))

  7079.                              ? 3 : 0;

  7080.             int step = IS_8x8DCT(mb_type) ? 2 : 1;

  7081.             edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;

  7082.             s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,

  7083.                                               (h->slice_type == B_TYPE), edges, step, mask_edge0, mask_edge1 );

  7084.         }

  7085.         if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) )

  7086.             bSv[0][0] = 0x0004000400040004ULL;

  7087.         if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) )

  7088.             bSv[1][0] = 0x0004000400040004ULL;

  7089. 

  7090. #define FILTER(hv,dir,edge)\

  7091.         if(bSv[dir][edge]) {\

  7092.             filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\

  7093.             if(!(edge&1)) {\

  7094.                 filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\

  7095.                 filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\

  7096.             }\

  7097.         }

  7098.         if( edges == 1 ) {

  7099.             FILTER(v,0,0);

  7100.             FILTER(h,1,0);

  7101.         } else if( IS_8x8DCT(mb_type) ) {

  7102.             FILTER(v,0,0);

  7103.             FILTER(v,0,2);

  7104.             FILTER(h,1,0);

  7105.             FILTER(h,1,2);

  7106.         } else {

  7107.             FILTER(v,0,0);

  7108.             FILTER(v,0,1);

  7109.             FILTER(v,0,2);

  7110.             FILTER(v,0,3);

  7111.             FILTER(h,1,0);

  7112.             FILTER(h,1,1);

  7113.             FILTER(h,1,2);

  7114.             FILTER(h,1,3);

  7115.         }

  7116. #undef FILTER

  7117.     }

  7118. }

  7119. 

  7120. 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) {

  7121.     MpegEncContext * const s = &h->s;

  7122.     const int mb_xy= mb_x + mb_y*s->mb_stride;

  7123.     const int mb_type = s->current_picture.mb_type[mb_xy];

  7124.     const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;

  7125.     int first_vertical_edge_done = 0;

  7126.     int dir;

  7127.     /* FIXME: A given frame may occupy more than one position in

  7128.      * the reference list. So ref2frm should be populated with

  7129.      * frame numbers, not indices. */

  7130.     static const int ref2frm[34] = {-1,-1,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,

  7131.                                     16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31};

  7132. 

  7133.     //for sufficiently low qp, filtering wouldn't do anything

  7134.     //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp

  7135.     if(!FRAME_MBAFF){

  7136.         int qp_thresh = 15 - h->slice_alpha_c0_offset - FFMAX(0, h->pps.chroma_qp_index_offset);

  7137.         int qp = s->current_picture.qscale_table[mb_xy];

  7138.         if(qp <= qp_thresh

  7139.            && (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh)

  7140.            && (mb_y == 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){

  7141.             return;

  7142.         }

  7143.     }

  7144. 

  7145.     if (FRAME_MBAFF

  7146.             // left mb is in picture

  7147.             && h->slice_table[mb_xy-1] != 255

  7148.             // and current and left pair do not have the same interlaced type

  7149.             && (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))

  7150.             // and left mb is in the same slice if deblocking_filter == 2

  7151.             && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {

  7152.         /* First vertical edge is different in MBAFF frames

  7153.          * There are 8 different bS to compute and 2 different Qp

  7154.          */

  7155.         const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;

  7156.         const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };

  7157.         int16_t bS[8];

  7158.         int qp[2];

  7159.         int chroma_qp[2];

  7160.         int mb_qp, mbn0_qp, mbn1_qp;

  7161.         int i;

  7162.         first_vertical_edge_done = 1;

  7163. 

  7164.         if( IS_INTRA(mb_type) )

  7165.             bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4;

  7166.         else {

  7167.             for( i = 0; i < 8; i++ ) {

  7168.                 int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];

  7169. 

  7170.                 if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )

  7171.                     bS[i] = 4;

  7172.                 else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||

  7173.                          /* FIXME: with 8x8dct + cavlc, should check cbp instead of nnz */

  7174.                          h->non_zero_count[mbn_xy][MB_FIELD ? i&3 : (i>>2)+(mb_y&1)*2] )

  7175.                     bS[i] = 2;

  7176.                 else

  7177.                     bS[i] = 1;

  7178.             }

  7179.         }

  7180. 

  7181.         mb_qp = s->current_picture.qscale_table[mb_xy];

  7182.         mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];

  7183.         mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];

  7184.         qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;

  7185.         chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) +

  7186.                          get_chroma_qp( h->pps.chroma_qp_index_offset, mbn0_qp ) + 1 ) >> 1;

  7187.         qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;

  7188.         chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, mb_qp ) +

  7189.                          get_chroma_qp( h->pps.chroma_qp_index_offset, mbn1_qp ) + 1 ) >> 1;

  7190. 

  7191.         /* Filter edge */

  7192.         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);

  7193.         { int i; for (i = 0; i < 8; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }

  7194.         filter_mb_mbaff_edgev ( h, &img_y [0], linesize,   bS, qp );

  7195.         filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp );

  7196.         filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp );

  7197.     }

  7198.     /* dir : 0 -> vertical edge, 1 -> horizontal edge */

  7199.     for( dir = 0; dir < 2; dir++ )

  7200.     {

  7201.         int edge;

  7202.         const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;

  7203.         const int mbm_type = s->current_picture.mb_type[mbm_xy];

  7204.         int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;

  7205. 

  7206.         const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP))

  7207.                                   == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4;

  7208.         // how often to recheck mv-based bS when iterating between edges

  7209.         const int mask_edge = (mb_type & (MB_TYPE_16x16 | (MB_TYPE_16x8 << dir))) ? 3 :

  7210.                               (mb_type & (MB_TYPE_8x16 >> dir)) ? 1 : 0;

  7211.         // how often to recheck mv-based bS when iterating along each edge

  7212.         const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));

  7213. 

  7214.         if (first_vertical_edge_done) {

  7215.             start = 1;

  7216.             first_vertical_edge_done = 0;

  7217.         }

  7218. 

  7219.         if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])

  7220.             start = 1;

  7221. 

  7222.         if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0

  7223.             && !IS_INTERLACED(mb_type)

  7224.             && IS_INTERLACED(mbm_type)

  7225.             ) {

  7226.             // This is a special case in the norm where the filtering must

  7227.             // be done twice (one each of the field) even if we are in a

  7228.             // frame macroblock.

  7229.             //

  7230.             static const int nnz_idx[4] = {4,5,6,3};

  7231.             unsigned int tmp_linesize   = 2 *   linesize;

  7232.             unsigned int tmp_uvlinesize = 2 * uvlinesize;

  7233.             int mbn_xy = mb_xy - 2 * s->mb_stride;

  7234.             int qp, chroma_qp;

  7235.             int i, j;

  7236.             int16_t bS[4];

  7237. 

  7238.             for(j=0; j<2; j++, mbn_xy += s->mb_stride){

  7239.                 if( IS_INTRA(mb_type) ||

  7240.                     IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) {

  7241.                     bS[0] = bS[1] = bS[2] = bS[3] = 3;

  7242.                 } else {

  7243.                     const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy];

  7244.                     for( i = 0; i < 4; i++ ) {

  7245.                         if( h->non_zero_count_cache[scan8[0]+i] != 0 ||

  7246.                             mbn_nnz[nnz_idx[i]] != 0 )

  7247.                             bS[i] = 2;

  7248.                         else

  7249.                             bS[i] = 1;

  7250.                     }

  7251.                 }

  7252.                 // Do not use s->qscale as luma quantizer because it has not the same

  7253.                 // value in IPCM macroblocks.

  7254.                 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;

  7255.                 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);

  7256.                 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }

  7257.                 filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp );

  7258.                 chroma_qp = ( h->chroma_qp +

  7259.                               get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;

  7260.                 filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp );

  7261.                 filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp );

  7262.             }

  7263. 

  7264.             start = 1;

  7265.         }

  7266. 

  7267.         /* Calculate bS */

  7268.         for( edge = start; edge < edges; edge++ ) {

  7269.             /* mbn_xy: neighbor macroblock */

  7270.             const int mbn_xy = edge > 0 ? mb_xy : mbm_xy;

  7271.             const int mbn_type = s->current_picture.mb_type[mbn_xy];

  7272.             int16_t bS[4];

  7273.             int qp;

  7274. 

  7275.             if( (edge&1) && IS_8x8DCT(mb_type) )

  7276.                 continue;

  7277. 

  7278.             if( IS_INTRA(mb_type) ||

  7279.                 IS_INTRA(mbn_type) ) {

  7280.                 int value;

  7281.                 if (edge == 0) {

  7282.                     if (   (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type))

  7283.                         || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))

  7284.                     ) {

  7285.                         value = 4;

  7286.                     } else {

  7287.                         value = 3;

  7288.                     }

  7289.                 } else {

  7290.                     value = 3;

  7291.                 }

  7292.                 bS[0] = bS[1] = bS[2] = bS[3] = value;

  7293.             } else {

  7294.                 int i, l;

  7295.                 int mv_done;

  7296. 

  7297.                 if( edge & mask_edge ) {

  7298.                     bS[0] = bS[1] = bS[2] = bS[3] = 0;

  7299.                     mv_done = 1;

  7300.                 }

  7301.                 else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) {

  7302.                     bS[0] = bS[1] = bS[2] = bS[3] = 1;

  7303.                     mv_done = 1;

  7304.                 }

  7305.                 else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {

  7306.                     int b_idx= 8 + 4 + edge * (dir ? 8:1);

  7307.                     int bn_idx= b_idx - (dir ? 8:1);

  7308.                     int v = 0;

  7309.                     for( l = 0; !v && l < 1 + (h->slice_type == B_TYPE); l++ ) {

  7310.                         v |= ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||

  7311.                              ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||

  7312.                              ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;

  7313.                     }

  7314.                     bS[0] = bS[1] = bS[2] = bS[3] = v;

  7315.                     mv_done = 1;

  7316.                 }

  7317.                 else

  7318.                     mv_done = 0;

  7319. 

  7320.                 for( i = 0; i < 4; i++ ) {

  7321.                     int x = dir == 0 ? edge : i;

  7322.                     int y = dir == 0 ? i    : edge;

  7323.                     int b_idx= 8 + 4 + x + 8*y;

  7324.                     int bn_idx= b_idx - (dir ? 8:1);

  7325. 

  7326.                     if( h->non_zero_count_cache[b_idx] != 0 ||

  7327.                         h->non_zero_count_cache[bn_idx] != 0 ) {

  7328.                         bS[i] = 2;

  7329.                     }

  7330.                     else if(!mv_done)

  7331.                     {

  7332.                         bS[i] = 0;

  7333.                         for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {

  7334.                             if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||

  7335.                                 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||

  7336.                                 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {

  7337.                                 bS[i] = 1;

  7338.                                 break;

  7339.                             }

  7340.                         }

  7341.                     }

  7342.                 }

  7343. 

  7344.                 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)

  7345.                     continue;

  7346.             }

  7347. 

  7348.             /* Filter edge */

  7349.             // Do not use s->qscale as luma quantizer because it has not the same

  7350.             // value in IPCM macroblocks.

  7351.             qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;

  7352.             //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]);

  7353.             tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);

  7354.             { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }

  7355.             if( dir == 0 ) {

  7356.                 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );

  7357.                 if( (edge&1) == 0 ) {

  7358.                     int chroma_qp = ( h->chroma_qp +

  7359.                                       get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;

  7360.                     filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );

  7361.                     filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );

  7362.                 }

  7363.             } else {

  7364.                 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );

  7365.                 if( (edge&1) == 0 ) {

  7366.                     int chroma_qp = ( h->chroma_qp +

  7367.                                       get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;

  7368.                     filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );

  7369.                     filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );

  7370.                 }

  7371.             }

  7372.         }

  7373.     }

  7374. }

  7375. 

  7376. static int decode_slice(H264Context *h){

  7377.     MpegEncContext * const s = &h->s;

  7378.     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;

  7379. 

  7380.     s->mb_skip_run= -1;

  7381. 

  7382.     if( h->pps.cabac ) {

  7383.         int i;

  7384. 

  7385.         /* realign */

  7386.         align_get_bits( &s->gb );

  7387. 

  7388.         /* init cabac */

  7389.         ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );

  7390.         ff_init_cabac_decoder( &h->cabac,

  7391.                                s->gb.buffer + get_bits_count(&s->gb)/8,

  7392.                                ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);

  7393.         /* calculate pre-state */

  7394.         for( i= 0; i < 460; i++ ) {

  7395.             int pre;

  7396.             if( h->slice_type == I_TYPE )

  7397.                 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );

  7398.             else

  7399.                 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 );

  7400. 

  7401.             if( pre <= 63 )

  7402.                 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;

  7403.             else

  7404.                 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;

  7405.         }

  7406. 

  7407.         for(;;){

  7408.             int ret = decode_mb_cabac(h);

  7409.             int eos;

  7410. 

  7411.             if(ret>=0) hl_decode_mb(h);

  7412. 

  7413.             if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?

  7414.                 s->mb_y++;

  7415. 

  7416.                 if(ret>=0) ret = decode_mb_cabac(h);

  7417. 

  7418.                 if(ret>=0) hl_decode_mb(h);

  7419.                 s->mb_y--;

  7420.             }

  7421.             eos = get_cabac_terminate( &h->cabac );

  7422. 

  7423.             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {

  7424.                 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);

  7425.                 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);

  7426.                 return -1;

  7427.             }

  7428. 

  7429.             if( ++s->mb_x >= s->mb_width ) {

  7430.                 s->mb_x = 0;

  7431.                 ff_draw_horiz_band(s, 16*s->mb_y, 16);

  7432.                 ++s->mb_y;

  7433.                 if(FRAME_MBAFF) {

  7434.                     ++s->mb_y;

  7435.                 }

  7436.             }

  7437. 

  7438.             if( eos || s->mb_y >= s->mb_height ) {

  7439.                 tprintf("slice end %d %d\n", 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.                 return 0;

  7442.             }

  7443.         }

  7444. 

  7445.     } else {

  7446.         for(;;){

  7447.             int ret = decode_mb_cavlc(h);

  7448. 

  7449.             if(ret>=0) hl_decode_mb(h);

  7450. 

  7451.             if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?

  7452.                 s->mb_y++;

  7453.                 ret = decode_mb_cavlc(h);

  7454. 

  7455.                 if(ret>=0) hl_decode_mb(h);

  7456.                 s->mb_y--;

  7457.             }

  7458. 

  7459.             if(ret<0){

  7460.                 av_log(h->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.                 ff_draw_horiz_band(s, 16*s->mb_y, 16);

  7469.                 ++s->mb_y;

  7470.                 if(FRAME_MBAFF) {

  7471.                     ++s->mb_y;

  7472.                 }

  7473.                 if(s->mb_y >= s->mb_height){

  7474.                     tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);

  7475. 

  7476.                     if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {

  7477.                         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);

  7478. 

  7479.                         return 0;

  7480.                     }else{

  7481.                         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);

  7482. 

  7483.                         return -1;

  7484.                     }

  7485.                 }

  7486.             }

  7487. 

  7488.             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){

  7489.                 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);

  7490.                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){

  7491.                     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);

  7492. 

  7493.                     return 0;

  7494.                 }else{

  7495.                     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);

  7496. 

  7497.                     return -1;

  7498.                 }

  7499.             }

  7500.         }

  7501.     }

  7502. 

  7503. #if 0

  7504.     for(;s->mb_y < s->mb_height; s->mb_y++){

  7505.         for(;s->mb_x < s->mb_width; s->mb_x++){

  7506.             int ret= decode_mb(h);

  7507. 

  7508.             hl_decode_mb(h);

  7509. 

  7510.             if(ret<0){

  7511.                 av_log(s->avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  7512.                 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);

  7513. 

  7514.                 return -1;

  7515.             }

  7516. 

  7517.             if(++s->mb_x >= s->mb_width){

  7518.                 s->mb_x=0;

  7519.                 if(++s->mb_y >= s->mb_height){

  7520.                     if(get_bits_count(s->gb) == s->gb.size_in_bits){

  7521.                         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);

  7522. 

  7523.                         return 0;

  7524.                     }else{

  7525.                         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);

  7526. 

  7527.                         return -1;

  7528.                     }

  7529.                 }

  7530.             }

  7531. 

  7532.             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){

  7533.                 if(get_bits_count(s->gb) == s->gb.size_in_bits){

  7534.                     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);

  7535. 

  7536.                     return 0;

  7537.                 }else{

  7538.                     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);

  7539. 

  7540.                     return -1;

  7541.                 }

  7542.             }

  7543.         }

  7544.         s->mb_x=0;

  7545.         ff_draw_horiz_band(s, 16*s->mb_y, 16);

  7546.     }

  7547. #endif

  7548.     return -1; //not reached

  7549. }

  7550. 

  7551. static int decode_unregistered_user_data(H264Context *h, int size){

  7552.     MpegEncContext * const s = &h->s;

  7553.     uint8_t user_data[16+256];

  7554.     int e, build, i;

  7555. 

  7556.     if(size<16)

  7557.         return -1;

  7558. 

  7559.     for(i=0; i<sizeof(user_data)-1 && i<size; i++){

  7560.         user_data[i]= get_bits(&s->gb, 8);

  7561.     }

  7562. 

  7563.     user_data[i]= 0;

  7564.     e= sscanf(user_data+16, "x264 - core %d"/*%s - H.264/MPEG-4 AVC codec - Copyleft 2005 - http://www.videolan.org/x264.html*/, &build);

  7565.     if(e==1 && build>=0)

  7566.         h->x264_build= build;

  7567. 

  7568.     if(s->avctx->debug & FF_DEBUG_BUGS)

  7569.         av_log(s->avctx, AV_LOG_DEBUG, "user data:\"%s\"\n", user_data+16);

  7570. 

  7571.     for(; i<size; i++)

  7572.         skip_bits(&s->gb, 8);

  7573. 

  7574.     return 0;

  7575. }

  7576. 

  7577. static int decode_sei(H264Context *h){

  7578.     MpegEncContext * const s = &h->s;

  7579. 

  7580.     while(get_bits_count(&s->gb) + 16 < s->gb.size_in_bits){

  7581.         int size, type;

  7582. 

  7583.         type=0;

  7584.         do{

  7585.             type+= show_bits(&s->gb, 8);

  7586.         }while(get_bits(&s->gb, 8) == 255);

  7587. 

  7588.         size=0;

  7589.         do{

  7590.             size+= show_bits(&s->gb, 8);

  7591.         }while(get_bits(&s->gb, 8) == 255);

  7592. 

  7593.         switch(type){

  7594.         case 5:

  7595.             if(decode_unregistered_user_data(h, size) < 0)

  7596.                 return -1;

  7597.             break;

  7598.         default:

  7599.             skip_bits(&s->gb, 8*size);

  7600.         }

  7601. 

  7602.         //FIXME check bits here

  7603.         align_get_bits(&s->gb);

  7604.     }

  7605. 

  7606.     return 0;

  7607. }

  7608. 

  7609. static inline void decode_hrd_parameters(H264Context *h, SPS *sps){

  7610.     MpegEncContext * const s = &h->s;

  7611.     int cpb_count, i;

  7612.     cpb_count = get_ue_golomb(&s->gb) + 1;

  7613.     get_bits(&s->gb, 4); /* bit_rate_scale */

  7614.     get_bits(&s->gb, 4); /* cpb_size_scale */

  7615.     for(i=0; i<cpb_count; i++){

  7616.         get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */

  7617.         get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */

  7618.         get_bits1(&s->gb);     /* cbr_flag */

  7619.     }

  7620.     get_bits(&s->gb, 5); /* initial_cpb_removal_delay_length_minus1 */

  7621.     get_bits(&s->gb, 5); /* cpb_removal_delay_length_minus1 */

  7622.     get_bits(&s->gb, 5); /* dpb_output_delay_length_minus1 */

  7623.     get_bits(&s->gb, 5); /* time_offset_length */

  7624. }

  7625. 

  7626. static inline int decode_vui_parameters(H264Context *h, SPS *sps){

  7627.     MpegEncContext * const s = &h->s;

  7628.     int aspect_ratio_info_present_flag, aspect_ratio_idc;

  7629.     int nal_hrd_parameters_present_flag, vcl_hrd_parameters_present_flag;

  7630. 

  7631.     aspect_ratio_info_present_flag= get_bits1(&s->gb);

  7632. 

  7633.     if( aspect_ratio_info_present_flag ) {

  7634.         aspect_ratio_idc= get_bits(&s->gb, 8);

  7635.         if( aspect_ratio_idc == EXTENDED_SAR ) {

  7636.             sps->sar.num= get_bits(&s->gb, 16);

  7637.             sps->sar.den= get_bits(&s->gb, 16);

  7638.         }else if(aspect_ratio_idc < 14){

  7639.             sps->sar=  pixel_aspect[aspect_ratio_idc];

  7640.         }else{

  7641.             av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");

  7642.             return -1;

  7643.         }

  7644.     }else{

  7645.         sps->sar.num=

  7646.         sps->sar.den= 0;

  7647.     }

  7648. //            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);

  7649. 

  7650.     if(get_bits1(&s->gb)){      /* overscan_info_present_flag */

  7651.         get_bits1(&s->gb);      /* overscan_appropriate_flag */

  7652.     }

  7653. 

  7654.     if(get_bits1(&s->gb)){      /* video_signal_type_present_flag */

  7655.         get_bits(&s->gb, 3);    /* video_format */

  7656.         get_bits1(&s->gb);      /* video_full_range_flag */

  7657.         if(get_bits1(&s->gb)){  /* colour_description_present_flag */

  7658.             get_bits(&s->gb, 8); /* colour_primaries */

  7659.             get_bits(&s->gb, 8); /* transfer_characteristics */

  7660.             get_bits(&s->gb, 8); /* matrix_coefficients */

  7661.         }

  7662.     }

  7663. 

  7664.     if(get_bits1(&s->gb)){      /* chroma_location_info_present_flag */

  7665.         get_ue_golomb(&s->gb);  /* chroma_sample_location_type_top_field */

  7666.         get_ue_golomb(&s->gb);  /* chroma_sample_location_type_bottom_field */

  7667.     }

  7668. 

  7669.     sps->timing_info_present_flag = get_bits1(&s->gb);

  7670.     if(sps->timing_info_present_flag){

  7671.         sps->num_units_in_tick = get_bits_long(&s->gb, 32);

  7672.         sps->time_scale = get_bits_long(&s->gb, 32);

  7673.         sps->fixed_frame_rate_flag = get_bits1(&s->gb);

  7674.     }

  7675. 

  7676.     nal_hrd_parameters_present_flag = get_bits1(&s->gb);

  7677.     if(nal_hrd_parameters_present_flag)

  7678.         decode_hrd_parameters(h, sps);

  7679.     vcl_hrd_parameters_present_flag = get_bits1(&s->gb);

  7680.     if(vcl_hrd_parameters_present_flag)

  7681.         decode_hrd_parameters(h, sps);

  7682.     if(nal_hrd_parameters_present_flag || vcl_hrd_parameters_present_flag)

  7683.         get_bits1(&s->gb);     /* low_delay_hrd_flag */

  7684.     get_bits1(&s->gb);         /* pic_struct_present_flag */

  7685. 

  7686.     sps->bitstream_restriction_flag = get_bits1(&s->gb);

  7687.     if(sps->bitstream_restriction_flag){

  7688.         get_bits1(&s->gb);     /* motion_vectors_over_pic_boundaries_flag */

  7689.         get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */

  7690.         get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */

  7691.         get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */

  7692.         get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */

  7693.         sps->num_reorder_frames = get_ue_golomb(&s->gb);

  7694.         get_ue_golomb(&s->gb); /* max_dec_frame_buffering */

  7695.     }

  7696. 

  7697.     return 0;

  7698. }

  7699. 

  7700. static void decode_scaling_list(H264Context *h, uint8_t *factors, int size,

  7701.                                 const uint8_t *jvt_list, const uint8_t *fallback_list){

  7702.     MpegEncContext * const s = &h->s;

  7703.     int i, last = 8, next = 8;

  7704.     const uint8_t *scan = size == 16 ? zigzag_scan : zigzag_scan8x8;

  7705.     if(!get_bits1(&s->gb)) /* matrix not written, we use the predicted one */

  7706.         memcpy(factors, fallback_list, size*sizeof(uint8_t));

  7707.     else

  7708.     for(i=0;i<size;i++){

  7709.         if(next)

  7710.             next = (last + get_se_golomb(&s->gb)) & 0xff;

  7711.         if(!i && !next){ /* matrix not written, we use the preset one */

  7712.             memcpy(factors, jvt_list, size*sizeof(uint8_t));

  7713.             break;

  7714.         }

  7715.         last = factors[scan[i]] = next ? next : last;

  7716.     }

  7717. }

  7718. 

  7719. static void decode_scaling_matrices(H264Context *h, SPS *sps, PPS *pps, int is_sps,

  7720.                                    uint8_t (*scaling_matrix4)[16], uint8_t (*scaling_matrix8)[64]){

  7721.     MpegEncContext * const s = &h->s;

  7722.     int fallback_sps = !is_sps && sps->scaling_matrix_present;

  7723.     const uint8_t *fallback[4] = {

  7724.         fallback_sps ? sps->scaling_matrix4[0] : default_scaling4[0],

  7725.         fallback_sps ? sps->scaling_matrix4[3] : default_scaling4[1],

  7726.         fallback_sps ? sps->scaling_matrix8[0] : default_scaling8[0],

  7727.         fallback_sps ? sps->scaling_matrix8[1] : default_scaling8[1]

  7728.     };

  7729.     if(get_bits1(&s->gb)){

  7730.         sps->scaling_matrix_present |= is_sps;

  7731.         decode_scaling_list(h,scaling_matrix4[0],16,default_scaling4[0],fallback[0]); // Intra, Y

  7732.         decode_scaling_list(h,scaling_matrix4[1],16,default_scaling4[0],scaling_matrix4[0]); // Intra, Cr

  7733.         decode_scaling_list(h,scaling_matrix4[2],16,default_scaling4[0],scaling_matrix4[1]); // Intra, Cb

  7734.         decode_scaling_list(h,scaling_matrix4[3],16,default_scaling4[1],fallback[1]); // Inter, Y

  7735.         decode_scaling_list(h,scaling_matrix4[4],16,default_scaling4[1],scaling_matrix4[3]); // Inter, Cr

  7736.         decode_scaling_list(h,scaling_matrix4[5],16,default_scaling4[1],scaling_matrix4[4]); // Inter, Cb

  7737.         if(is_sps || pps->transform_8x8_mode){

  7738.             decode_scaling_list(h,scaling_matrix8[0],64,default_scaling8[0],fallback[2]);  // Intra, Y

  7739.             decode_scaling_list(h,scaling_matrix8[1],64,default_scaling8[1],fallback[3]);  // Inter, Y

  7740.         }

  7741.     } else if(fallback_sps) {

  7742.         memcpy(scaling_matrix4, sps->scaling_matrix4, 6*16*sizeof(uint8_t));

  7743.         memcpy(scaling_matrix8, sps->scaling_matrix8, 2*64*sizeof(uint8_t));

  7744.     }

  7745. }

  7746. 

  7747. static inline int decode_seq_parameter_set(H264Context *h){

  7748.     MpegEncContext * const s = &h->s;

  7749.     int profile_idc, level_idc;

  7750.     int sps_id, i;

  7751.     SPS *sps;

  7752. 

  7753.     profile_idc= get_bits(&s->gb, 8);

  7754.     get_bits1(&s->gb);   //constraint_set0_flag

  7755.     get_bits1(&s->gb);   //constraint_set1_flag

  7756.     get_bits1(&s->gb);   //constraint_set2_flag

  7757.     get_bits1(&s->gb);   //constraint_set3_flag

  7758.     get_bits(&s->gb, 4); // reserved

  7759.     level_idc= get_bits(&s->gb, 8);

  7760.     sps_id= get_ue_golomb(&s->gb);

  7761. 

  7762.     sps= &h->sps_buffer[ sps_id ];

  7763.     sps->profile_idc= profile_idc;

  7764.     sps->level_idc= level_idc;

  7765. 

  7766.     if(sps->profile_idc >= 100){ //high profile

  7767.         if(get_ue_golomb(&s->gb) == 3) //chroma_format_idc

  7768.             get_bits1(&s->gb);  //residual_color_transform_flag

  7769.         get_ue_golomb(&s->gb);  //bit_depth_luma_minus8

  7770.         get_ue_golomb(&s->gb);  //bit_depth_chroma_minus8

  7771.         sps->transform_bypass = get_bits1(&s->gb);

  7772.         decode_scaling_matrices(h, sps, NULL, 1, sps->scaling_matrix4, sps->scaling_matrix8);

  7773.     }else

  7774.         sps->scaling_matrix_present = 0;

  7775. 

  7776.     sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;

  7777.     sps->poc_type= get_ue_golomb(&s->gb);

  7778. 

  7779.     if(sps->poc_type == 0){ //FIXME #define

  7780.         sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;

  7781.     } else if(sps->poc_type == 1){//FIXME #define

  7782.         sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);

  7783.         sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);

  7784.         sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);

  7785.         sps->poc_cycle_length= get_ue_golomb(&s->gb);

  7786. 

  7787.         for(i=0; i<sps->poc_cycle_length; i++)

  7788.             sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);

  7789.     }

  7790.     if(sps->poc_type > 2){

  7791.         av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);

  7792.         return -1;

  7793.     }

  7794. 

  7795.     sps->ref_frame_count= get_ue_golomb(&s->gb);

  7796.     if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){

  7797.         av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");

  7798.     }

  7799.     sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);

  7800.     sps->mb_width= get_ue_golomb(&s->gb) + 1;

  7801.     sps->mb_height= get_ue_golomb(&s->gb) + 1;

  7802.     if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 ||

  7803.        avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height))

  7804.         return -1;

  7805. 

  7806.     sps->frame_mbs_only_flag= get_bits1(&s->gb);

  7807.     if(!sps->frame_mbs_only_flag)

  7808.         sps->mb_aff= get_bits1(&s->gb);

  7809.     else

  7810.         sps->mb_aff= 0;

  7811. 

  7812.     sps->direct_8x8_inference_flag= get_bits1(&s->gb);

  7813. 

  7814. #ifndef ALLOW_INTERLACE

  7815.     if(sps->mb_aff)

  7816.         av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF support not included; enable it compilation time\n");

  7817. #endif

  7818.     if(!sps->direct_8x8_inference_flag && sps->mb_aff)

  7819.         av_log(h->s.avctx, AV_LOG_ERROR, "MBAFF + !direct_8x8_inference is not implemented\n");

  7820. 

  7821.     sps->crop= get_bits1(&s->gb);

  7822.     if(sps->crop){

  7823.         sps->crop_left  = get_ue_golomb(&s->gb);

  7824.         sps->crop_right = get_ue_golomb(&s->gb);

  7825.         sps->crop_top   = get_ue_golomb(&s->gb);

  7826.         sps->crop_bottom= get_ue_golomb(&s->gb);

  7827.         if(sps->crop_left || sps->crop_top){

  7828.             av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completely supported, this could look slightly wrong ...\n");

  7829.         }

  7830.     }else{

  7831.         sps->crop_left  =

  7832.         sps->crop_right =

  7833.         sps->crop_top   =

  7834.         sps->crop_bottom= 0;

  7835.     }

  7836. 

  7837.     sps->vui_parameters_present_flag= get_bits1(&s->gb);

  7838.     if( sps->vui_parameters_present_flag )

  7839.         decode_vui_parameters(h, sps);

  7840. 

  7841.     if(s->avctx->debug&FF_DEBUG_PICT_INFO){

  7842.         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",

  7843.                sps_id, sps->profile_idc, sps->level_idc,

  7844.                sps->poc_type,

  7845.                sps->ref_frame_count,

  7846.                sps->mb_width, sps->mb_height,

  7847.                sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),

  7848.                sps->direct_8x8_inference_flag ? "8B8" : "",

  7849.                sps->crop_left, sps->crop_right,

  7850.                sps->crop_top, sps->crop_bottom,

  7851.                sps->vui_parameters_present_flag ? "VUI" : ""

  7852.                );

  7853.     }

  7854.     return 0;

  7855. }

  7856. 

  7857. static inline int decode_picture_parameter_set(H264Context *h, int bit_length){

  7858.     MpegEncContext * const s = &h->s;

  7859.     int pps_id= get_ue_golomb(&s->gb);

  7860.     PPS *pps= &h->pps_buffer[pps_id];

  7861. 

  7862.     pps->sps_id= get_ue_golomb(&s->gb);

  7863.     pps->cabac= get_bits1(&s->gb);

  7864.     pps->pic_order_present= get_bits1(&s->gb);

  7865.     pps->slice_group_count= get_ue_golomb(&s->gb) + 1;

  7866.     if(pps->slice_group_count > 1 ){

  7867.         pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);

  7868.         av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");

  7869.         switch(pps->mb_slice_group_map_type){

  7870.         case 0:

  7871. #if 0

  7872. |   for( i = 0; i <= num_slice_groups_minus1; i++ ) |   |        |

  7873. |    run_length[ i ]                                |1  |ue(v)   |

  7874. #endif

  7875.             break;

  7876.         case 2:

  7877. #if 0

  7878. |   for( i = 0; i < num_slice_groups_minus1; i++ )  |   |        |

  7879. |{                                                  |   |        |

  7880. |    top_left_mb[ i ]                               |1  |ue(v)   |

  7881. |    bottom_right_mb[ i ]                           |1  |ue(v)   |

  7882. |   }                                               |   |        |

  7883. #endif

  7884.             break;

  7885.         case 3:

  7886.         case 4:

  7887.         case 5:

  7888. #if 0

  7889. |   slice_group_change_direction_flag               |1  |u(1)    |

  7890. |   slice_group_change_rate_minus1                  |1  |ue(v)   |

  7891. #endif

  7892.             break;

  7893.         case 6:

  7894. #if 0

  7895. |   slice_group_id_cnt_minus1                       |1  |ue(v)   |

  7896. |   for( i = 0; i <= slice_group_id_cnt_minus1; i++ |   |        |

  7897. |)                                                  |   |        |

  7898. |    slice_group_id[ i ]                            |1  |u(v)    |

  7899. #endif

  7900.             break;

  7901.         }

  7902.     }

  7903.     pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;

  7904.     pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;

  7905.     if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){

  7906.         av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");

  7907.         return -1;

  7908.     }

  7909. 

  7910.     pps->weighted_pred= get_bits1(&s->gb);

  7911.     pps->weighted_bipred_idc= get_bits(&s->gb, 2);

  7912.     pps->init_qp= get_se_golomb(&s->gb) + 26;

  7913.     pps->init_qs= get_se_golomb(&s->gb) + 26;

  7914.     pps->chroma_qp_index_offset= get_se_golomb(&s->gb);

  7915.     pps->deblocking_filter_parameters_present= get_bits1(&s->gb);

  7916.     pps->constrained_intra_pred= get_bits1(&s->gb);

  7917.     pps->redundant_pic_cnt_present = get_bits1(&s->gb);

  7918. 

  7919.     pps->transform_8x8_mode= 0;

  7920.     h->dequant_coeff_pps= -1; //contents of sps/pps can change even if id doesn't, so reinit

  7921.     memset(pps->scaling_matrix4, 16, 6*16*sizeof(uint8_t));

  7922.     memset(pps->scaling_matrix8, 16, 2*64*sizeof(uint8_t));

  7923. 

  7924.     if(get_bits_count(&s->gb) < bit_length){

  7925.         pps->transform_8x8_mode= get_bits1(&s->gb);

  7926.         decode_scaling_matrices(h, &h->sps_buffer[pps->sps_id], pps, 0, pps->scaling_matrix4, pps->scaling_matrix8);

  7927.         get_se_golomb(&s->gb);  //second_chroma_qp_index_offset

  7928.     }

  7929. 

  7930.     if(s->avctx->debug&FF_DEBUG_PICT_INFO){

  7931.         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",

  7932.                pps_id, pps->sps_id,

  7933.                pps->cabac ? "CABAC" : "CAVLC",

  7934.                pps->slice_group_count,

  7935.                pps->ref_count[0], pps->ref_count[1],

  7936.                pps->weighted_pred ? "weighted" : "",

  7937.                pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,

  7938.                pps->deblocking_filter_parameters_present ? "LPAR" : "",

  7939.                pps->constrained_intra_pred ? "CONSTR" : "",

  7940.                pps->redundant_pic_cnt_present ? "REDU" : "",

  7941.                pps->transform_8x8_mode ? "8x8DCT" : ""

  7942.                );

  7943.     }

  7944. 

  7945.     return 0;

  7946. }

  7947. 

  7948. /**

  7949.  * finds the end of the current frame in the bitstream.

  7950.  * @return the position of the first byte of the next frame, or -1

  7951.  */

  7952. static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){

  7953.     int i;

  7954.     uint32_t state;

  7955.     ParseContext *pc = &(h->s.parse_context);

  7956. //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);

  7957. //    mb_addr= pc->mb_addr - 1;

  7958.     state= pc->state;

  7959.     for(i=0; i<=buf_size; i++){

  7960.         if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){

  7961.             tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);

  7962.             if(pc->frame_start_found){

  7963.                 // If there isn't one more byte in the buffer

  7964.                 // the test on first_mb_in_slice cannot be done yet

  7965.                 // do it at next call.

  7966.                 if (i >= buf_size) break;

  7967.                 if (buf[i] & 0x80) {

  7968.                     // first_mb_in_slice is 0, probably the first nal of a new

  7969.                     // slice

  7970.                     tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);

  7971.                     pc->state=-1;

  7972.                     pc->frame_start_found= 0;

  7973.                     return i-4;

  7974.                 }

  7975.             }

  7976.             pc->frame_start_found = 1;

  7977.         }

  7978.         if((state&0xFFFFFF1F) == 0x107 || (state&0xFFFFFF1F) == 0x108 || (state&0xFFFFFF1F) == 0x109){

  7979.            if(pc->frame_start_found){

  7980.                 pc->state=-1;

  7981.                 pc->frame_start_found= 0;

  7982.                 return i-4;

  7983.            }

  7984.         }

  7985.         if (i<buf_size)

  7986.             state= (state<<8) | buf[i];

  7987.     }

  7988. 

  7989.     pc->state= state;

  7990.     return END_NOT_FOUND;

  7991. }

  7992. 

  7993. #ifdef CONFIG_H264_PARSER

  7994. static int h264_parse(AVCodecParserContext *s,

  7995.                       AVCodecContext *avctx,

  7996.                       uint8_t **poutbuf, int *poutbuf_size,

  7997.                       const uint8_t *buf, int buf_size)

  7998. {

  7999.     H264Context *h = s->priv_data;

  8000.     ParseContext *pc = &h->s.parse_context;

  8001.     int next;

  8002. 

  8003.     next= find_frame_end(h, buf, buf_size);

  8004. 

  8005.     if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {

  8006.         *poutbuf = NULL;

  8007.         *poutbuf_size = 0;

  8008.         return buf_size;

  8009.     }

  8010. 

  8011.     *poutbuf = (uint8_t *)buf;

  8012.     *poutbuf_size = buf_size;

  8013.     return next;

  8014. }

  8015. 

  8016. static int h264_split(AVCodecContext *avctx,

  8017.                       const uint8_t *buf, int buf_size)

  8018. {

  8019.     int i;

  8020.     uint32_t state = -1;

  8021.     int has_sps= 0;

  8022. 

  8023.     for(i=0; i<=buf_size; i++){

  8024.         if((state&0xFFFFFF1F) == 0x107)

  8025.             has_sps=1;

  8026. /*        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){

  8027.         }*/

  8028.         if((state&0xFFFFFF00) == 0x100 && (state&0xFFFFFF1F) != 0x107 && (state&0xFFFFFF1F) != 0x108 && (state&0xFFFFFF1F) != 0x109){

  8029.             if(has_sps){

  8030.                 while(i>4 && buf[i-5]==0) i--;

  8031.                 return i-4;

  8032.             }

  8033.         }

  8034.         if (i<buf_size)

  8035.             state= (state<<8) | buf[i];

  8036.     }

  8037.     return 0;

  8038. }

  8039. #endif /* CONFIG_H264_PARSER */

  8040. 

  8041. static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){

  8042.     MpegEncContext * const s = &h->s;

  8043.     AVCodecContext * const avctx= s->avctx;

  8044.     int buf_index=0;

  8045. #if 0

  8046.     int i;

  8047.     for(i=0; i<50; i++){

  8048.         av_log(NULL, AV_LOG_ERROR,"%02X ", buf[i]);

  8049.     }

  8050. #endif

  8051.     h->slice_num = 0;

  8052.     s->current_picture_ptr= NULL;

  8053.     for(;;){

  8054.         int consumed;

  8055.         int dst_length;

  8056.         int bit_length;

  8057.         uint8_t *ptr;

  8058.         int i, nalsize = 0;

  8059. 

  8060.       if(h->is_avc) {

  8061.         if(buf_index >= buf_size) break;

  8062.         nalsize = 0;

  8063.         for(i = 0; i < h->nal_length_size; i++)

  8064.             nalsize = (nalsize << 8) | buf[buf_index++];

  8065.         if(nalsize <= 1){

  8066.             if(nalsize == 1){

  8067.                 buf_index++;

  8068.                 continue;

  8069.             }else{

  8070.                 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);

  8071.                 break;

  8072.             }

  8073.         }

  8074.       } else {

  8075.         // start code prefix search

  8076.         for(; buf_index + 3 < buf_size; buf_index++){

  8077.             // this should allways succeed in the first iteration

  8078.             if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)

  8079.                 break;

  8080.         }

  8081. 

  8082.         if(buf_index+3 >= buf_size) break;

  8083. 

  8084.         buf_index+=3;

  8085.       }

  8086. 

  8087.         ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);

  8088.         while(ptr[dst_length - 1] == 0 && dst_length > 1)

  8089.             dst_length--;

  8090.         bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);

  8091. 

  8092.         if(s->avctx->debug&FF_DEBUG_STARTCODE){

  8093.             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);

  8094.         }

  8095. 

  8096.         if (h->is_avc && (nalsize != consumed))

  8097.             av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);

  8098. 

  8099.         buf_index += consumed;

  8100. 

  8101.         if(  (s->hurry_up == 1 && h->nal_ref_idc  == 0) //FIXME dont discard SEI id

  8102.            ||(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc  == 0))

  8103.             continue;

  8104. 

  8105.         switch(h->nal_unit_type){

  8106.         case NAL_IDR_SLICE:

  8107.             idr(h); //FIXME ensure we don't loose some frames if there is reordering

  8108.         case NAL_SLICE:

  8109.             init_get_bits(&s->gb, ptr, bit_length);

  8110.             h->intra_gb_ptr=

  8111.             h->inter_gb_ptr= &s->gb;

  8112.             s->data_partitioning = 0;

  8113. 

  8114.             if(decode_slice_header(h) < 0){

  8115.                 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");

  8116.                 break;

  8117.             }

  8118.             s->current_picture_ptr->key_frame= (h->nal_unit_type == NAL_IDR_SLICE);

  8119.             if(h->redundant_pic_count==0 && s->hurry_up < 5

  8120.                && (avctx->skip_frame < AVDISCARD_NONREF || h->nal_ref_idc)

  8121.                && (avctx->skip_frame < AVDISCARD_BIDIR  || h->slice_type!=B_TYPE)

  8122.                && (avctx->skip_frame < AVDISCARD_NONKEY || h->slice_type==I_TYPE)

  8123.                && avctx->skip_frame < AVDISCARD_ALL)

  8124.                 decode_slice(h);

  8125.             break;

  8126.         case NAL_DPA:

  8127.             init_get_bits(&s->gb, ptr, bit_length);

  8128.             h->intra_gb_ptr=

  8129.             h->inter_gb_ptr= NULL;

  8130.             s->data_partitioning = 1;

  8131. 

  8132.             if(decode_slice_header(h) < 0){

  8133.                 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");

  8134.             }

  8135.             break;

  8136.         case NAL_DPB:

  8137.             init_get_bits(&h->intra_gb, ptr, bit_length);

  8138.             h->intra_gb_ptr= &h->intra_gb;

  8139.             break;

  8140.         case NAL_DPC:

  8141.             init_get_bits(&h->inter_gb, ptr, bit_length);

  8142.             h->inter_gb_ptr= &h->inter_gb;

  8143. 

  8144.             if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning

  8145.                && s->hurry_up < 5

  8146.                && (avctx->skip_frame < AVDISCARD_NONREF || h->nal_ref_idc)

  8147.                && (avctx->skip_frame < AVDISCARD_BIDIR  || h->slice_type!=B_TYPE)

  8148.                && (avctx->skip_frame < AVDISCARD_NONKEY || h->slice_type==I_TYPE)

  8149.                && avctx->skip_frame < AVDISCARD_ALL)

  8150.                 decode_slice(h);

  8151.             break;

  8152.         case NAL_SEI:

  8153.             init_get_bits(&s->gb, ptr, bit_length);

  8154.             decode_sei(h);

  8155.             break;

  8156.         case NAL_SPS:

  8157.             init_get_bits(&s->gb, ptr, bit_length);

  8158.             decode_seq_parameter_set(h);

  8159. 

  8160.             if(s->flags& CODEC_FLAG_LOW_DELAY)

  8161.                 s->low_delay=1;

  8162. 

  8163.             if(avctx->has_b_frames < 2)

  8164.                 avctx->has_b_frames= !s->low_delay;

  8165.             break;

  8166.         case NAL_PPS:

  8167.             init_get_bits(&s->gb, ptr, bit_length);

  8168. 

  8169.             decode_picture_parameter_set(h, bit_length);

  8170. 

  8171.             break;

  8172.         case NAL_AUD:

  8173.         case NAL_END_SEQUENCE:

  8174.         case NAL_END_STREAM:

  8175.         case NAL_FILLER_DATA:

  8176.         case NAL_SPS_EXT:

  8177.         case NAL_AUXILIARY_SLICE:

  8178.             break;

  8179.         default:

  8180.             av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);

  8181.         }

  8182.     }

  8183. 

  8184.     if(!s->current_picture_ptr) return buf_index; //no frame

  8185. 

  8186.     s->current_picture_ptr->qscale_type= FF_QSCALE_TYPE_H264;

  8187.     s->current_picture_ptr->pict_type= s->pict_type;

  8188. 

  8189.     h->prev_frame_num_offset= h->frame_num_offset;

  8190.     h->prev_frame_num= h->frame_num;

  8191.     if(s->current_picture_ptr->reference){

  8192.         h->prev_poc_msb= h->poc_msb;

  8193.         h->prev_poc_lsb= h->poc_lsb;

  8194.     }

  8195.     if(s->current_picture_ptr->reference)

  8196.         execute_ref_pic_marking(h, h->mmco, h->mmco_index);

  8197. 

  8198.     ff_er_frame_end(s);

  8199. 

  8200.     MPV_frame_end(s);

  8201. 

  8202.     return buf_index;

  8203. }

  8204. 

  8205. /**

  8206.  * returns the number of bytes consumed for building the current frame

  8207.  */

  8208. static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){

  8209.     if(s->flags&CODEC_FLAG_TRUNCATED){

  8210.         pos -= s->parse_context.last_index;

  8211.         if(pos<0) pos=0; // FIXME remove (unneeded?)

  8212. 

  8213.         return pos;

  8214.     }else{

  8215.         if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)

  8216.         if(pos+10>buf_size) pos=buf_size; // oops ;)

  8217. 

  8218.         return pos;

  8219.     }

  8220. }

  8221. 

  8222. static int decode_frame(AVCodecContext *avctx,

  8223.                              void *data, int *data_size,

  8224.                              uint8_t *buf, int buf_size)

  8225. {

  8226.     H264Context *h = avctx->priv_data;

  8227.     MpegEncContext *s = &h->s;

  8228.     AVFrame *pict = data;

  8229.     int buf_index;

  8230. 

  8231.     s->flags= avctx->flags;

  8232.     s->flags2= avctx->flags2;

  8233. 

  8234.    /* no supplementary picture */

  8235.     if (buf_size == 0) {

  8236.         return 0;

  8237.     }

  8238. 

  8239.     if(s->flags&CODEC_FLAG_TRUNCATED){

  8240.         int next= find_frame_end(h, buf, buf_size);

  8241. 

  8242.         if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )

  8243.             return buf_size;

  8244. //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);

  8245.     }

  8246. 

  8247.     if(h->is_avc && !h->got_avcC) {

  8248.         int i, cnt, nalsize;

  8249.         unsigned char *p = avctx->extradata;

  8250.         if(avctx->extradata_size < 7) {

  8251.             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");

  8252.             return -1;

  8253.         }

  8254.         if(*p != 1) {

  8255.             av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);

  8256.             return -1;

  8257.         }

  8258.         /* sps and pps in the avcC always have length coded with 2 bytes,

  8259.            so put a fake nal_length_size = 2 while parsing them */

  8260.         h->nal_length_size = 2;

  8261.         // Decode sps from avcC

  8262.         cnt = *(p+5) & 0x1f; // Number of sps

  8263.         p += 6;

  8264.         for (i = 0; i < cnt; i++) {

  8265.             nalsize = BE_16(p) + 2;

  8266.             if(decode_nal_units(h, p, nalsize) < 0) {

  8267.                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);

  8268.                 return -1;

  8269.             }

  8270.             p += nalsize;

  8271.         }

  8272.         // Decode pps from avcC

  8273.         cnt = *(p++); // Number of pps

  8274.         for (i = 0; i < cnt; i++) {

  8275.             nalsize = BE_16(p) + 2;

  8276.             if(decode_nal_units(h, p, nalsize)  != nalsize) {

  8277.                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);

  8278.                 return -1;

  8279.             }

  8280.             p += nalsize;

  8281.         }

  8282.         // Now store right nal length size, that will be use to parse all other nals

  8283.         h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;

  8284.         // Do not reparse avcC

  8285.         h->got_avcC = 1;

  8286.     }

  8287. 

  8288.     if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){

  8289.         if(decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) < 0)

  8290.             return -1;

  8291.     }

  8292. 

  8293.     buf_index=decode_nal_units(h, buf, buf_size);

  8294.     if(buf_index < 0)

  8295.         return -1;

  8296. 

  8297.     //FIXME do something with unavailable reference frames

  8298. 

  8299. //    if(ret==FRAME_SKIPPED) return get_consumed_bytes(s, buf_index, buf_size);

  8300.     if(!s->current_picture_ptr){

  8301.         av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");

  8302.         return -1;

  8303.     }

  8304. 

  8305.     {

  8306.         Picture *out = s->current_picture_ptr;

  8307. #if 0 //decode order

  8308.         *data_size = sizeof(AVFrame);

  8309. #else

  8310.         /* Sort B-frames into display order */

  8311.         Picture *cur = s->current_picture_ptr;

  8312.         Picture *prev = h->delayed_output_pic;

  8313.         int i, pics, cross_idr, out_of_order, out_idx;

  8314. 

  8315.         if(h->sps.bitstream_restriction_flag

  8316.            && s->avctx->has_b_frames < h->sps.num_reorder_frames){

  8317.             s->avctx->has_b_frames = h->sps.num_reorder_frames;

  8318.             s->low_delay = 0;

  8319.         }

  8320. 

  8321.         pics = 0;

  8322.         while(h->delayed_pic[pics]) pics++;

  8323.         h->delayed_pic[pics++] = cur;

  8324.         if(cur->reference == 0)

  8325.             cur->reference = 1;

  8326. 

  8327.         cross_idr = 0;

  8328.         for(i=0; h->delayed_pic[i]; i++)

  8329.             if(h->delayed_pic[i]->key_frame || h->delayed_pic[i]->poc==0)

  8330.                 cross_idr = 1;

  8331. 

  8332.         out = h->delayed_pic[0];

  8333.         out_idx = 0;

  8334.         for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++)

  8335.             if(h->delayed_pic[i]->poc < out->poc){

  8336.                 out = h->delayed_pic[i];

  8337.                 out_idx = i;

  8338.             }

  8339. 

  8340.         out_of_order = !cross_idr && prev && out->poc < prev->poc;

  8341.         if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)

  8342.             { }

  8343.         else if(prev && pics <= s->avctx->has_b_frames)

  8344.             out = prev;

  8345.         else if((out_of_order && pics-1 == s->avctx->has_b_frames && pics < 15)

  8346.            || (s->low_delay &&

  8347.             ((!cross_idr && prev && out->poc > prev->poc + 2)

  8348.              || cur->pict_type == B_TYPE)))

  8349.         {

  8350.             s->low_delay = 0;

  8351.             s->avctx->has_b_frames++;

  8352.             out = prev;

  8353.         }

  8354.         else if(out_of_order)

  8355.             out = prev;

  8356. 

  8357.         if(out_of_order || pics > s->avctx->has_b_frames){

  8358.             for(i=out_idx; h->delayed_pic[i]; i++)

  8359.                 h->delayed_pic[i] = h->delayed_pic[i+1];

  8360.         }

  8361. 

  8362.         if(prev == out)

  8363.             *data_size = 0;

  8364.         else

  8365.             *data_size = sizeof(AVFrame);

  8366.         if(prev && prev != out && prev->reference == 1)

  8367.             prev->reference = 0;

  8368.         h->delayed_output_pic = out;

  8369. #endif

  8370. 

  8371.         if(out)

  8372.             *pict= *(AVFrame*)out;

  8373.         else

  8374.             av_log(avctx, AV_LOG_DEBUG, "no picture\n");

  8375.     }

  8376. 

  8377.     assert(pict->data[0] || !*data_size);

  8378.     ff_print_debug_info(s, pict);

  8379. //printf("out %d\n", (int)pict->data[0]);

  8380. #if 0 //?

  8381. 

  8382.     /* Return the Picture timestamp as the frame number */

  8383.     /* we substract 1 because it is added on utils.c    */

  8384.     avctx->frame_number = s->picture_number - 1;

  8385. #endif

  8386.     return get_consumed_bytes(s, buf_index, buf_size);

  8387. }

  8388. #if 0

  8389. static inline void fill_mb_avail(H264Context *h){

  8390.     MpegEncContext * const s = &h->s;

  8391.     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

  8392. 

  8393.     if(s->mb_y){

  8394.         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;

  8395.         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;

  8396.         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;

  8397.     }else{

  8398.         h->mb_avail[0]=

  8399.         h->mb_avail[1]=

  8400.         h->mb_avail[2]= 0;

  8401.     }

  8402.     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;

  8403.     h->mb_avail[4]= 1; //FIXME move out

  8404.     h->mb_avail[5]= 0; //FIXME move out

  8405. }

  8406. #endif

  8407. 

  8408. #if 0 //selftest

  8409. #define COUNT 8000

  8410. #define SIZE (COUNT*40)

  8411. int main(){

  8412.     int i;

  8413.     uint8_t temp[SIZE];

  8414.     PutBitContext pb;

  8415.     GetBitContext gb;

  8416. //    int int_temp[10000];

  8417.     DSPContext dsp;

  8418.     AVCodecContext avctx;

  8419. 

  8420.     dsputil_init(&dsp, &avctx);

  8421. 

  8422.     init_put_bits(&pb, temp, SIZE);

  8423.     printf("testing unsigned exp golomb\n");

  8424.     for(i=0; i<COUNT; i++){

  8425.         START_TIMER

  8426.         set_ue_golomb(&pb, i);

  8427.         STOP_TIMER("set_ue_golomb");

  8428.     }

  8429.     flush_put_bits(&pb);

  8430. 

  8431.     init_get_bits(&gb, temp, 8*SIZE);

  8432.     for(i=0; i<COUNT; i++){

  8433.         int j, s;

  8434. 

  8435.         s= show_bits(&gb, 24);

  8436. 

  8437.         START_TIMER

  8438.         j= get_ue_golomb(&gb);

  8439.         if(j != i){

  8440.             printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);

  8441. //            return -1;

  8442.         }

  8443.         STOP_TIMER("get_ue_golomb");

  8444.     }

  8445. 

  8446. 

  8447.     init_put_bits(&pb, temp, SIZE);

  8448.     printf("testing signed exp golomb\n");

  8449.     for(i=0; i<COUNT; i++){

  8450.         START_TIMER

  8451.         set_se_golomb(&pb, i - COUNT/2);

  8452.         STOP_TIMER("set_se_golomb");

  8453.     }

  8454.     flush_put_bits(&pb);

  8455. 

  8456.     init_get_bits(&gb, temp, 8*SIZE);

  8457.     for(i=0; i<COUNT; i++){

  8458.         int j, s;

  8459. 

  8460.         s= show_bits(&gb, 24);

  8461. 

  8462.         START_TIMER

  8463.         j= get_se_golomb(&gb);

  8464.         if(j != i - COUNT/2){

  8465.             printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);

  8466. //            return -1;

  8467.         }

  8468.         STOP_TIMER("get_se_golomb");

  8469.     }

  8470. 

  8471.     printf("testing 4x4 (I)DCT\n");

  8472. 

  8473.     DCTELEM block[16];

  8474.     uint8_t src[16], ref[16];

  8475.     uint64_t error= 0, max_error=0;

  8476. 

  8477.     for(i=0; i<COUNT; i++){

  8478.         int j;

  8479. //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);

  8480.         for(j=0; j<16; j++){

  8481.             ref[j]= random()%255;

  8482.             src[j]= random()%255;

  8483.         }

  8484. 

  8485.         h264_diff_dct_c(block, src, ref, 4);

  8486. 

  8487.         //normalize

  8488.         for(j=0; j<16; j++){

  8489. //            printf("%d ", block[j]);

  8490.             block[j]= block[j]*4;

  8491.             if(j&1) block[j]= (block[j]*4 + 2)/5;

  8492.             if(j&4) block[j]= (block[j]*4 + 2)/5;

  8493.         }

  8494. //        printf("\n");

  8495. 

  8496.         s->dsp.h264_idct_add(ref, block, 4);

  8497. /*        for(j=0; j<16; j++){

  8498.             printf("%d ", ref[j]);

  8499.         }

  8500.         printf("\n");*/

  8501. 

  8502.         for(j=0; j<16; j++){

  8503.             int diff= ABS(src[j] - ref[j]);

  8504. 

  8505.             error+= diff*diff;

  8506.             max_error= FFMAX(max_error, diff);

  8507.         }

  8508.     }

  8509.     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );

  8510. #if 0

  8511.     printf("testing quantizer\n");

  8512.     for(qp=0; qp<52; qp++){

  8513.         for(i=0; i<16; i++)

  8514.             src1_block[i]= src2_block[i]= random()%255;

  8515. 

  8516.     }

  8517. #endif

  8518.     printf("Testing NAL layer\n");

  8519. 

  8520.     uint8_t bitstream[COUNT];

  8521.     uint8_t nal[COUNT*2];

  8522.     H264Context h;

  8523.     memset(&h, 0, sizeof(H264Context));

  8524. 

  8525.     for(i=0; i<COUNT; i++){

  8526.         int zeros= i;

  8527.         int nal_length;

  8528.         int consumed;

  8529.         int out_length;

  8530.         uint8_t *out;

  8531.         int j;

  8532. 

  8533.         for(j=0; j<COUNT; j++){

  8534.             bitstream[j]= (random() % 255) + 1;

  8535.         }

  8536. 

  8537.         for(j=0; j<zeros; j++){

  8538.             int pos= random() % COUNT;

  8539.             while(bitstream[pos] == 0){

  8540.                 pos++;

  8541.                 pos %= COUNT;

  8542.             }

  8543.             bitstream[pos]=0;

  8544.         }

  8545. 

  8546.         START_TIMER

  8547. 

  8548.         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);

  8549.         if(nal_length<0){

  8550.             printf("encoding failed\n");

  8551.             return -1;

  8552.         }

  8553. 

  8554.         out= decode_nal(&h, nal, &out_length, &consumed, nal_length);

  8555. 

  8556.         STOP_TIMER("NAL")

  8557. 

  8558.         if(out_length != COUNT){

  8559.             printf("incorrect length %d %d\n", out_length, COUNT);

  8560.             return -1;

  8561.         }

  8562. 

  8563.         if(consumed != nal_length){

  8564.             printf("incorrect consumed length %d %d\n", nal_length, consumed);

  8565.             return -1;

  8566.         }

  8567. 

  8568.         if(memcmp(bitstream, out, COUNT)){

  8569.             printf("missmatch\n");

  8570.             return -1;

  8571.         }

  8572.     }

  8573. 

  8574.     printf("Testing RBSP\n");

  8575. 

  8576. 

  8577.     return 0;

  8578. }

  8579. #endif

  8580. 

  8581. 

  8582. static int decode_end(AVCodecContext *avctx)

  8583. {

  8584.     H264Context *h = avctx->priv_data;

  8585.     MpegEncContext *s = &h->s;

  8586. 

  8587.     av_freep(&h->rbsp_buffer);

  8588.     free_tables(h); //FIXME cleanup init stuff perhaps

  8589.     MPV_common_end(s);

  8590. 

  8591. //    memset(h, 0, sizeof(H264Context));

  8592. 

  8593.     return 0;

  8594. }

  8595. 

  8596. 

  8597. AVCodec h264_decoder = {

  8598.     "h264",

  8599.     CODEC_TYPE_VIDEO,

  8600.     CODEC_ID_H264,

  8601.     sizeof(H264Context),

  8602.     decode_init,

  8603.     NULL,

  8604.     decode_end,

  8605.     decode_frame,

  8606.     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY,

  8607.     .flush= flush_dpb,

  8608. };

  8609. 

  8610. #ifdef CONFIG_H264_PARSER

  8611. AVCodecParser h264_parser = {

  8612.     { CODEC_ID_H264 },

  8613.     sizeof(H264Context),

  8614.     NULL,

  8615.     h264_parse,

  8616.     ff_parse_close,

  8617.     h264_split,

  8618. };

  8619. #endif

  8620. 

  8621. #include "svq3.c"