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.
70890 Commits
32 Branches
981MB
Tree: c27f53b9fb
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c27f53b9fb'
${ noResults }
FFmpeg/libavcodec/h264_loopfilter.c

847 lines
42KB
Raw Blame History 

  1. /*

  2.  * H.26L/H.264/AVC/JVT/14496-10/... loop filter

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

  4.  *

  5.  * This file is part of FFmpeg.

  6.  *

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

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

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

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

  11.  *

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

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

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

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

  20.  */

  21. 

  22. /**

  23.  * @file

  24.  * H.264 / AVC / MPEG4 part10 loop filter.

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

  26.  */

  27. 

  28. #include "libavutil/internal.h"

  29. #include "libavutil/intreadwrite.h"

  30. #include "internal.h"

  31. #include "avcodec.h"

  32. #include "h264.h"

  33. #include "mathops.h"

  34. #include "mpegutils.h"

  35. #include "rectangle.h"

  36. 

  37. /* Deblocking filter (p153) */

  38. static const uint8_t alpha_table[52*3] = {

  39.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  40.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  41.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  42.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  43.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  44.      0,  0,  0,  0,  0,  0,  4,  4,  5,  6,

  45.      7,  8,  9, 10, 12, 13, 15, 17, 20, 22,

  46.     25, 28, 32, 36, 40, 45, 50, 56, 63, 71,

  47.     80, 90,101,113,127,144,162,182,203,226,

  48.    255,255,

  49.    255,255,255,255,255,255,255,255,255,255,255,255,255,

  50.    255,255,255,255,255,255,255,255,255,255,255,255,255,

  51.    255,255,255,255,255,255,255,255,255,255,255,255,255,

  52.    255,255,255,255,255,255,255,255,255,255,255,255,255,

  53. };

  54. static const uint8_t beta_table[52*3] = {

  55.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  56.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  57.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  58.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  59.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  60.      0,  0,  0,  0,  0,  0,  2,  2,  2,  3,

  61.      3,  3,  3,  4,  4,  4,  6,  6,  7,  7,

  62.      8,  8,  9,  9, 10, 10, 11, 11, 12, 12,

  63.     13, 13, 14, 14, 15, 15, 16, 16, 17, 17,

  64.     18, 18,

  65.     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,

  66.     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,

  67.     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,

  68.     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,

  69. };

  70. static const uint8_t tc0_table[52*3][4] = {

  71.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  72.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  73.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  74.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  75.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  76.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  77.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  78.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  79.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  80.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  81.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },

  82.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 1 },

  83.     {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },

  84.     {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },

  85.     {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },

  86.     {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },

  87.     {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },

  88.     {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-1,13,17,25 },

  89.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  90.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  91.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  92.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  93.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  94.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  95.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  96.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  97.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  98. };

  99. 

  100. /* intra: 0 if this loopfilter call is guaranteed to be inter (bS < 4), 1 if it might be intra (bS == 4) */

  101. static av_always_inline void filter_mb_edgev(uint8_t *pix, int stride,

  102.                                              const int16_t bS[4],

  103.                                              unsigned int qp, int a, int b,

  104.                                              H264Context *h, int intra)

  105. {

  106.     const unsigned int index_a = qp + a;

  107.     const int alpha = alpha_table[index_a];

  108.     const int beta  = beta_table[qp + b];

  109.     if (alpha ==0 || beta == 0) return;

  110. 

  111.     if( bS[0] < 4 || !intra ) {

  112.         int8_t tc[4];

  113.         tc[0] = tc0_table[index_a][bS[0]];

  114.         tc[1] = tc0_table[index_a][bS[1]];

  115.         tc[2] = tc0_table[index_a][bS[2]];

  116.         tc[3] = tc0_table[index_a][bS[3]];

  117.         h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);

  118.     } else {

  119.         h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);

  120.     }

  121. }

  122. 

  123. static av_always_inline void filter_mb_edgecv(uint8_t *pix, int stride,

  124.                                               const int16_t bS[4],

  125.                                               unsigned int qp, int a, int b,

  126.                                               H264Context *h, int intra)

  127. {

  128.     const unsigned int index_a = qp + a;

  129.     const int alpha = alpha_table[index_a];

  130.     const int beta  = beta_table[qp + b];

  131.     if (alpha ==0 || beta == 0) return;

  132. 

  133.     if( bS[0] < 4 || !intra ) {

  134.         int8_t tc[4];

  135.         tc[0] = tc0_table[index_a][bS[0]]+1;

  136.         tc[1] = tc0_table[index_a][bS[1]]+1;

  137.         tc[2] = tc0_table[index_a][bS[2]]+1;

  138.         tc[3] = tc0_table[index_a][bS[3]]+1;

  139.         h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);

  140.     } else {

  141.         h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);

  142.     }

  143. }

  144. 

  145. static av_always_inline void filter_mb_mbaff_edgev(H264Context *h, uint8_t *pix,

  146.                                                    int stride,

  147.                                                    const int16_t bS[7], int bsi,

  148.                                                    int qp, int a, int b,

  149.                                                    int intra)

  150. {

  151.     const unsigned int index_a = qp + a;

  152.     const int alpha = alpha_table[index_a];

  153.     const int beta  = beta_table[qp + b];

  154.     if (alpha ==0 || beta == 0) return;

  155. 

  156.     if( bS[0] < 4 || !intra ) {

  157.         int8_t tc[4];

  158.         tc[0] = tc0_table[index_a][bS[0*bsi]];

  159.         tc[1] = tc0_table[index_a][bS[1*bsi]];

  160.         tc[2] = tc0_table[index_a][bS[2*bsi]];

  161.         tc[3] = tc0_table[index_a][bS[3*bsi]];

  162.         h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);

  163.     } else {

  164.         h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);

  165.     }

  166. }

  167. 

  168. static av_always_inline void filter_mb_mbaff_edgecv(H264Context *h,

  169.                                                     uint8_t *pix, int stride,

  170.                                                     const int16_t bS[7],

  171.                                                     int bsi, int qp, int a,

  172.                                                     int b, int intra)

  173. {

  174.     const unsigned int index_a = qp + a;

  175.     const int alpha = alpha_table[index_a];

  176.     const int beta  = beta_table[qp + b];

  177.     if (alpha ==0 || beta == 0) return;

  178. 

  179.     if( bS[0] < 4 || !intra ) {

  180.         int8_t tc[4];

  181.         tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;

  182.         tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;

  183.         tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;

  184.         tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;

  185.         h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);

  186.     } else {

  187.         h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);

  188.     }

  189. }

  190. 

  191. static av_always_inline void filter_mb_edgeh(uint8_t *pix, int stride,

  192.                                              const int16_t bS[4],

  193.                                              unsigned int qp, int a, int b,

  194.                                              H264Context *h, int intra)

  195. {

  196.     const unsigned int index_a = qp + a;

  197.     const int alpha = alpha_table[index_a];

  198.     const int beta  = beta_table[qp + b];

  199.     if (alpha ==0 || beta == 0) return;

  200. 

  201.     if( bS[0] < 4 || !intra ) {

  202.         int8_t tc[4];

  203.         tc[0] = tc0_table[index_a][bS[0]];

  204.         tc[1] = tc0_table[index_a][bS[1]];

  205.         tc[2] = tc0_table[index_a][bS[2]];

  206.         tc[3] = tc0_table[index_a][bS[3]];

  207.         h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);

  208.     } else {

  209.         h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);

  210.     }

  211. }

  212. 

  213. static av_always_inline void filter_mb_edgech(uint8_t *pix, int stride,

  214.                                               const int16_t bS[4],

  215.                                               unsigned int qp, int a, int b,

  216.                                               H264Context *h, int intra)

  217. {

  218.     const unsigned int index_a = qp + a;

  219.     const int alpha = alpha_table[index_a];

  220.     const int beta  = beta_table[qp + b];

  221.     if (alpha ==0 || beta == 0) return;

  222. 

  223.     if( bS[0] < 4 || !intra ) {

  224.         int8_t tc[4];

  225.         tc[0] = tc0_table[index_a][bS[0]]+1;

  226.         tc[1] = tc0_table[index_a][bS[1]]+1;

  227.         tc[2] = tc0_table[index_a][bS[2]]+1;

  228.         tc[3] = tc0_table[index_a][bS[3]]+1;

  229.         h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);

  230.     } else {

  231.         h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);

  232.     }

  233. }

  234. 

  235. static av_always_inline void h264_filter_mb_fast_internal(H264Context *h,

  236.                                                           int mb_x, int mb_y,

  237.                                                           uint8_t *img_y,

  238.                                                           uint8_t *img_cb,

  239.                                                           uint8_t *img_cr,

  240.                                                           unsigned int linesize,

  241.                                                           unsigned int uvlinesize,

  242.                                                           int pixel_shift)

  243. {

  244.     int chroma = CHROMA(h) && !(CONFIG_GRAY && (h->flags&CODEC_FLAG_GRAY));

  245.     int chroma444 = CHROMA444(h);

  246.     int chroma422 = CHROMA422(h);

  247. 

  248.     int mb_xy = h->mb_xy;

  249.     int left_type= h->left_type[LTOP];

  250.     int top_type= h->top_type;

  251. 

  252.     int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);

  253.     int a = 52 + h->slice_alpha_c0_offset - qp_bd_offset;

  254.     int b = 52 + h->slice_beta_offset - qp_bd_offset;

  255. 

  256.     int mb_type = h->cur_pic.mb_type[mb_xy];

  257.     int qp      = h->cur_pic.qscale_table[mb_xy];

  258.     int qp0     = h->cur_pic.qscale_table[mb_xy - 1];

  259.     int qp1     = h->cur_pic.qscale_table[h->top_mb_xy];

  260.     int qpc = get_chroma_qp( h, 0, qp );

  261.     int qpc0 = get_chroma_qp( h, 0, qp0 );

  262.     int qpc1 = get_chroma_qp( h, 0, qp1 );

  263.     qp0 = (qp + qp0 + 1) >> 1;

  264.     qp1 = (qp + qp1 + 1) >> 1;

  265.     qpc0 = (qpc + qpc0 + 1) >> 1;

  266.     qpc1 = (qpc + qpc1 + 1) >> 1;

  267. 

  268.     if( IS_INTRA(mb_type) ) {

  269.         static const int16_t bS4[4] = {4,4,4,4};

  270.         static const int16_t bS3[4] = {3,3,3,3};

  271.         const int16_t *bSH = FIELD_PICTURE(h) ? bS3 : bS4;

  272.         if(left_type)

  273.             filter_mb_edgev( &img_y[4*0<<pixel_shift], linesize, bS4, qp0, a, b, h, 1);

  274.         if( IS_8x8DCT(mb_type) ) {

  275.             filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);

  276.             if(top_type){

  277.                 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);

  278.             }

  279.             filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);

  280.         } else {

  281.             filter_mb_edgev( &img_y[4*1<<pixel_shift], linesize, bS3, qp, a, b, h, 0);

  282.             filter_mb_edgev( &img_y[4*2<<pixel_shift], linesize, bS3, qp, a, b, h, 0);

  283.             filter_mb_edgev( &img_y[4*3<<pixel_shift], linesize, bS3, qp, a, b, h, 0);

  284.             if(top_type){

  285.                 filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, a, b, h, 1);

  286.             }

  287.             filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, a, b, h, 0);

  288.             filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, a, b, h, 0);

  289.             filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, a, b, h, 0);

  290.         }

  291.         if(chroma){

  292.             if(chroma444){

  293.                 if(left_type){

  294.                     filter_mb_edgev( &img_cb[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);

  295.                     filter_mb_edgev( &img_cr[4*0<<pixel_shift], linesize, bS4, qpc0, a, b, h, 1);

  296.                 }

  297.                 if( IS_8x8DCT(mb_type) ) {

  298.                     filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);

  299.                     filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);

  300.                     if(top_type){

  301.                         filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );

  302.                         filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1 );

  303.                     }

  304.                     filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);

  305.                     filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);

  306.                 } else {

  307.                     filter_mb_edgev( &img_cb[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);

  308.                     filter_mb_edgev( &img_cr[4*1<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);

  309.                     filter_mb_edgev( &img_cb[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);

  310.                     filter_mb_edgev( &img_cr[4*2<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);

  311.                     filter_mb_edgev( &img_cb[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);

  312.                     filter_mb_edgev( &img_cr[4*3<<pixel_shift], linesize, bS3, qpc, a, b, h, 0);

  313.                     if(top_type){

  314.                         filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);

  315.                         filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, a, b, h, 1);

  316.                     }

  317.                     filter_mb_edgeh( &img_cb[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);

  318.                     filter_mb_edgeh( &img_cr[4*1*linesize], linesize, bS3, qpc, a, b, h, 0);

  319.                     filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);

  320.                     filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, a, b, h, 0);

  321.                     filter_mb_edgeh( &img_cb[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);

  322.                     filter_mb_edgeh( &img_cr[4*3*linesize], linesize, bS3, qpc, a, b, h, 0);

  323.                 }

  324.             }else if(chroma422){

  325.                 if(left_type){

  326.                     filter_mb_edgecv(&img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);

  327.                     filter_mb_edgecv(&img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);

  328.                 }

  329.                 filter_mb_edgecv(&img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);

  330.                 filter_mb_edgecv(&img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);

  331.                 if(top_type){

  332.                     filter_mb_edgech(&img_cb[4*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);

  333.                     filter_mb_edgech(&img_cr[4*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);

  334.                 }

  335.                 filter_mb_edgech(&img_cb[4*1*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);

  336.                 filter_mb_edgech(&img_cr[4*1*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);

  337.                 filter_mb_edgech(&img_cb[4*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);

  338.                 filter_mb_edgech(&img_cr[4*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);

  339.                 filter_mb_edgech(&img_cb[4*3*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);

  340.                 filter_mb_edgech(&img_cr[4*3*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);

  341.             }else{

  342.                 if(left_type){

  343.                     filter_mb_edgecv( &img_cb[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);

  344.                     filter_mb_edgecv( &img_cr[2*0<<pixel_shift], uvlinesize, bS4, qpc0, a, b, h, 1);

  345.                 }

  346.                 filter_mb_edgecv( &img_cb[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);

  347.                 filter_mb_edgecv( &img_cr[2*2<<pixel_shift], uvlinesize, bS3, qpc, a, b, h, 0);

  348.                 if(top_type){

  349.                     filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);

  350.                     filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, a, b, h, 1);

  351.                 }

  352.                 filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);

  353.                 filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, a, b, h, 0);

  354.             }

  355.         }

  356.         return;

  357.     } else {

  358.         LOCAL_ALIGNED(8, int16_t, bS, [2], [4][4]);

  359.         int edges;

  360.         if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 && !chroma444 ) {

  361.             edges = 4;

  362.             AV_WN64A(bS[0][0], 0x0002000200020002ULL);

  363.             AV_WN64A(bS[0][2], 0x0002000200020002ULL);

  364.             AV_WN64A(bS[1][0], 0x0002000200020002ULL);

  365.             AV_WN64A(bS[1][2], 0x0002000200020002ULL);

  366.         } else {

  367.             int mask_edge1 = (3*(((5*mb_type)>>5)&1)) | (mb_type>>4); //(mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : (mb_type & MB_TYPE_16x8) ? 1 : 0;

  368.             int mask_edge0 = 3*((mask_edge1>>1) & ((5*left_type)>>5)&1); // (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) && (h->left_type[LTOP] & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 : 0;

  369.             int step =  1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;

  370.             edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;

  371.             h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,

  372.                                               h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE(h));

  373.         }

  374.         if( IS_INTRA(left_type) )

  375.             AV_WN64A(bS[0][0], 0x0004000400040004ULL);

  376.         if( IS_INTRA(top_type) )

  377.             AV_WN64A(bS[1][0], FIELD_PICTURE(h) ? 0x0003000300030003ULL : 0x0004000400040004ULL);

  378. 

  379. #define FILTER(hv,dir,edge,intra)\

  380.         if(AV_RN64A(bS[dir][edge])) {                                   \

  381.             filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qp : qp##dir, a, b, h, intra );\

  382.             if(chroma){\

  383.                 if(chroma444){\

  384.                     filter_mb_edge##hv( &img_cb[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\

  385.                     filter_mb_edge##hv( &img_cr[4*edge*(dir?linesize:1<<pixel_shift)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\

  386.                 } else if(!(edge&1)) {\

  387.                     filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1<<pixel_shift)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\

  388.                     filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1<<pixel_shift)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, a, b, h, intra );\

  389.                 }\

  390.             }\

  391.         }

  392.         if(left_type)

  393.             FILTER(v,0,0,1);

  394.         if( edges == 1 ) {

  395.             if(top_type)

  396.                 FILTER(h,1,0,1);

  397.         } else if( IS_8x8DCT(mb_type) ) {

  398.             FILTER(v,0,2,0);

  399.             if(top_type)

  400.                 FILTER(h,1,0,1);

  401.             FILTER(h,1,2,0);

  402.         } else {

  403.             FILTER(v,0,1,0);

  404.             FILTER(v,0,2,0);

  405.             FILTER(v,0,3,0);

  406.             if(top_type)

  407.                 FILTER(h,1,0,1);

  408.             FILTER(h,1,1,0);

  409.             FILTER(h,1,2,0);

  410.             FILTER(h,1,3,0);

  411.         }

  412. #undef FILTER

  413.     }

  414. }

  415. 

  416. void ff_h264_filter_mb_fast(H264Context *h, H264SliceContext *sl,

  417.                             int mb_x, int mb_y, uint8_t *img_y,

  418.                             uint8_t *img_cb, uint8_t *img_cr,

  419.                             unsigned int linesize, unsigned int uvlinesize)

  420. {

  421.     av_assert2(!FRAME_MBAFF(h));

  422.     if(!h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {

  423.         ff_h264_filter_mb(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);

  424.         return;

  425.     }

  426. 

  427. #if CONFIG_SMALL

  428.     h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, h->pixel_shift);

  429. #else

  430.     if(h->pixel_shift){

  431.         h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 1);

  432.     }else{

  433.         h264_filter_mb_fast_internal(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, 0);

  434.     }

  435. #endif

  436. }

  437. 

  438. static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){

  439.     int v;

  440. 

  441.     v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];

  442.     if(!v && h->ref_cache[0][b_idx]!=-1)

  443.         v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |

  444.            FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;

  445. 

  446.     if(h->list_count==2){

  447.         if(!v)

  448.             v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |

  449.                 h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |

  450.                 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;

  451. 

  452.         if(v){

  453.             if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |

  454.                h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])

  455.                 return 1;

  456.             return

  457.                 h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |

  458.                 FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |

  459.                 h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |

  460.                 FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;

  461.         }

  462.     }

  463. 

  464.     return v;

  465. }

  466. 

  467. static av_always_inline void filter_mb_dir(H264Context *h, H264SliceContext *sl,

  468.                                            int mb_x, int mb_y,

  469.                                            uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,

  470.                                            unsigned int linesize, unsigned int uvlinesize,

  471.                                            int mb_xy, int mb_type, int mvy_limit,

  472.                                            int first_vertical_edge_done, int a, int b,

  473.                                            int chroma, int dir)

  474. {

  475.     int edge;

  476.     int chroma_qp_avg[2];

  477.     int chroma444 = CHROMA444(h);

  478.     int chroma422 = CHROMA422(h);

  479.     const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;

  480.     const int mbm_type = dir == 0 ? h->left_type[LTOP] : h->top_type;

  481. 

  482.     // how often to recheck mv-based bS when iterating between edges

  483.     static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},

  484.                                               {0,3,1,1,3,3,3,3}};

  485.     const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];

  486.     const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;

  487. 

  488.     // how often to recheck mv-based bS when iterating along each edge

  489.     const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));

  490. 

  491.     if(mbm_type && !first_vertical_edge_done){

  492. 

  493.         if (FRAME_MBAFF(h) && (dir == 1) && ((mb_y&1) == 0)

  494.             && IS_INTERLACED(mbm_type&~mb_type)

  495.             ) {

  496.             // This is a special case in the norm where the filtering must

  497.             // be done twice (one each of the field) even if we are in a

  498.             // frame macroblock.

  499.             //

  500.             unsigned int tmp_linesize   = 2 *   linesize;

  501.             unsigned int tmp_uvlinesize = 2 * uvlinesize;

  502.             int mbn_xy = mb_xy - 2 * h->mb_stride;

  503.             int j;

  504. 

  505.             for(j=0; j<2; j++, mbn_xy += h->mb_stride){

  506.                 LOCAL_ALIGNED(8, int16_t, bS, [4]);

  507.                 int qp;

  508.                 if (IS_INTRA(mb_type | h->cur_pic.mb_type[mbn_xy])) {

  509.                     AV_WN64A(bS, 0x0003000300030003ULL);

  510.                 } else {

  511.                     if (!CABAC(h) && IS_8x8DCT(h->cur_pic.mb_type[mbn_xy])) {

  512.                         bS[0]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+0]);

  513.                         bS[1]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+1]);

  514.                         bS[2]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+2]);

  515.                         bS[3]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+3]);

  516.                     }else{

  517.                     const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 3*4;

  518.                     int i;

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

  520.                         bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);

  521.                     }

  522.                     }

  523.                 }

  524.                 // Do not use s->qscale as luma quantizer because it has not the same

  525.                 // value in IPCM macroblocks.

  526.                 qp = (h->cur_pic.qscale_table[mb_xy] + h->cur_pic.qscale_table[mbn_xy] + 1) >> 1;

  527.                 tprintf(h->avctx, "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);

  528.                 { int i; for (i = 0; i < 4; i++) tprintf(h->avctx, " bS[%d]:%d", i, bS[i]); tprintf(h->avctx, "\n"); }

  529.                 filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, a, b, h, 0 );

  530.                 chroma_qp_avg[0] = (sl->chroma_qp[0] + get_chroma_qp(h, 0, h->cur_pic.qscale_table[mbn_xy]) + 1) >> 1;

  531.                 chroma_qp_avg[1] = (sl->chroma_qp[1] + get_chroma_qp(h, 1, h->cur_pic.qscale_table[mbn_xy]) + 1) >> 1;

  532.                 if (chroma) {

  533.                     if (chroma444) {

  534.                         filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);

  535.                         filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);

  536.                     } else {

  537.                         filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], a, b, h, 0);

  538.                         filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], a, b, h, 0);

  539.                     }

  540.                 }

  541.             }

  542.         }else{

  543.             LOCAL_ALIGNED(8, int16_t, bS, [4]);

  544.             int qp;

  545. 

  546.             if( IS_INTRA(mb_type|mbm_type)) {

  547.                 AV_WN64A(bS, 0x0003000300030003ULL);

  548.                 if (   (!IS_INTERLACED(mb_type|mbm_type))

  549.                     || ((FRAME_MBAFF(h) || (h->picture_structure != PICT_FRAME)) && (dir == 0))

  550.                 )

  551.                     AV_WN64A(bS, 0x0004000400040004ULL);

  552.             } else {

  553.                 int i;

  554.                 int mv_done;

  555. 

  556.                 if( dir && FRAME_MBAFF(h) && IS_INTERLACED(mb_type ^ mbm_type)) {

  557.                     AV_WN64A(bS, 0x0001000100010001ULL);

  558.                     mv_done = 1;

  559.                 }

  560.                 else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {

  561.                     int b_idx= 8 + 4;

  562.                     int bn_idx= b_idx - (dir ? 8:1);

  563. 

  564.                     bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);

  565.                     mv_done = 1;

  566.                 }

  567.                 else

  568.                     mv_done = 0;

  569. 

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

  571.                     int x = dir == 0 ? 0 : i;

  572.                     int y = dir == 0 ? i    : 0;

  573.                     int b_idx= 8 + 4 + x + 8*y;

  574.                     int bn_idx= b_idx - (dir ? 8:1);

  575. 

  576.                     if( h->non_zero_count_cache[b_idx] |

  577.                         h->non_zero_count_cache[bn_idx] ) {

  578.                         bS[i] = 2;

  579.                     }

  580.                     else if(!mv_done)

  581.                     {

  582.                         bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);

  583.                     }

  584.                 }

  585.             }

  586. 

  587.             /* Filter edge */

  588.             // Do not use s->qscale as luma quantizer because it has not the same

  589.             // value in IPCM macroblocks.

  590.             if(bS[0]+bS[1]+bS[2]+bS[3]){

  591.                 qp = (h->cur_pic.qscale_table[mb_xy] + h->cur_pic.qscale_table[mbm_xy] + 1) >> 1;

  592.                 //tprintf(h->avctx, "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[0], h->cur_pic.qscale_table[mbn_xy]);

  593.                 tprintf(h->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);

  594.                 //{ int i; for (i = 0; i < 4; i++) tprintf(h->avctx, " bS[%d]:%d", i, bS[i]); tprintf(h->avctx, "\n"); }

  595.                 chroma_qp_avg[0] = (sl->chroma_qp[0] + get_chroma_qp(h, 0, h->cur_pic.qscale_table[mbm_xy]) + 1) >> 1;

  596.                 chroma_qp_avg[1] = (sl->chroma_qp[1] + get_chroma_qp(h, 1, h->cur_pic.qscale_table[mbm_xy]) + 1) >> 1;

  597.                 if( dir == 0 ) {

  598.                     filter_mb_edgev( &img_y[0], linesize, bS, qp, a, b, h, 1 );

  599.                     if (chroma) {

  600.                         if (chroma444) {

  601.                             filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);

  602.                             filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);

  603.                         } else {

  604.                             filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);

  605.                             filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);

  606.                         }

  607.                     }

  608.                 } else {

  609.                     filter_mb_edgeh( &img_y[0], linesize, bS, qp, a, b, h, 1 );

  610.                     if (chroma) {

  611.                         if (chroma444) {

  612.                             filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);

  613.                             filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);

  614.                         } else {

  615.                             filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], a, b, h, 1);

  616.                             filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], a, b, h, 1);

  617.                         }

  618.                     }

  619.                 }

  620.             }

  621.         }

  622.     }

  623. 

  624.     /* Calculate bS */

  625.     for( edge = 1; edge < edges; edge++ ) {

  626.         LOCAL_ALIGNED(8, int16_t, bS, [4]);

  627.         int qp;

  628.         const int deblock_edge = !IS_8x8DCT(mb_type & (edge<<24)); // (edge&1) && IS_8x8DCT(mb_type)

  629. 

  630.         if (!deblock_edge && (!chroma422 || dir == 0))

  631.             continue;

  632. 

  633.         if( IS_INTRA(mb_type)) {

  634.             AV_WN64A(bS, 0x0003000300030003ULL);

  635.         } else {

  636.             int i;

  637.             int mv_done;

  638. 

  639.             if( edge & mask_edge ) {

  640.                 AV_ZERO64(bS);

  641.                 mv_done = 1;

  642.             }

  643.             else if( mask_par0 ) {

  644.                 int b_idx= 8 + 4 + edge * (dir ? 8:1);

  645.                 int bn_idx= b_idx - (dir ? 8:1);

  646. 

  647.                 bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);

  648.                 mv_done = 1;

  649.             }

  650.             else

  651.                 mv_done = 0;

  652. 

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

  654.                 int x = dir == 0 ? edge : i;

  655.                 int y = dir == 0 ? i    : edge;

  656.                 int b_idx= 8 + 4 + x + 8*y;

  657.                 int bn_idx= b_idx - (dir ? 8:1);

  658. 

  659.                 if( h->non_zero_count_cache[b_idx] |

  660.                     h->non_zero_count_cache[bn_idx] ) {

  661.                     bS[i] = 2;

  662.                 }

  663.                 else if(!mv_done)

  664.                 {

  665.                     bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);

  666.                 }

  667.             }

  668. 

  669.             if(bS[0]+bS[1]+bS[2]+bS[3] == 0)

  670.                 continue;

  671.         }

  672. 

  673.         /* Filter edge */

  674.         // Do not use s->qscale as luma quantizer because it has not the same

  675.         // value in IPCM macroblocks.

  676.         qp = h->cur_pic.qscale_table[mb_xy];

  677.         //tprintf(h->avctx, "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[0], h->cur_pic.qscale_table[mbn_xy]);

  678.         tprintf(h->avctx, "filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);

  679.         //{ int i; for (i = 0; i < 4; i++) tprintf(h->avctx, " bS[%d]:%d", i, bS[i]); tprintf(h->avctx, "\n"); }

  680.         if( dir == 0 ) {

  681.             filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, a, b, h, 0 );

  682.             if (chroma) {

  683.                 if (chroma444) {

  684.                     filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);

  685.                     filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);

  686.                 } else if( (edge&1) == 0 ) {

  687.                     filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);

  688.                     filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);

  689.                 }

  690.             }

  691.         } else {

  692.             if (chroma422) {

  693.                 if (deblock_edge)

  694.                     filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0);

  695.                 if (chroma) {

  696.                     filter_mb_edgech(&img_cb[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);

  697.                     filter_mb_edgech(&img_cr[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);

  698.                 }

  699.             } else {

  700.                 filter_mb_edgeh(&img_y[4*edge*linesize], linesize, bS, qp, a, b, h, 0);

  701.                 if (chroma) {

  702.                     if (chroma444) {

  703.                         filter_mb_edgeh (&img_cb[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);

  704.                         filter_mb_edgeh (&img_cr[4*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);

  705.                     } else if ((edge&1) == 0) {

  706.                         filter_mb_edgech(&img_cb[2*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[0], a, b, h, 0);

  707.                         filter_mb_edgech(&img_cr[2*edge*uvlinesize], uvlinesize, bS, sl->chroma_qp[1], a, b, h, 0);

  708.                     }

  709.                 }

  710.             }

  711.         }

  712.     }

  713. }

  714. 

  715. void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl,

  716.                        int mb_x, int mb_y,

  717.                        uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,

  718.                        unsigned int linesize, unsigned int uvlinesize)

  719. {

  720.     const int mb_xy= mb_x + mb_y*h->mb_stride;

  721.     const int mb_type = h->cur_pic.mb_type[mb_xy];

  722.     const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;

  723.     int first_vertical_edge_done = 0;

  724.     int chroma = CHROMA(h) && !(CONFIG_GRAY && (h->flags&CODEC_FLAG_GRAY));

  725.     int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);

  726.     int a = 52 + h->slice_alpha_c0_offset - qp_bd_offset;

  727.     int b = 52 + h->slice_beta_offset - qp_bd_offset;

  728. 

  729.     if (FRAME_MBAFF(h)

  730.             // and current and left pair do not have the same interlaced type

  731.             && IS_INTERLACED(mb_type^h->left_type[LTOP])

  732.             // and left mb is in available to us

  733.             && h->left_type[LTOP]) {

  734.         /* First vertical edge is different in MBAFF frames

  735.          * There are 8 different bS to compute and 2 different Qp

  736.          */

  737.         LOCAL_ALIGNED(8, int16_t, bS, [8]);

  738.         int qp[2];

  739.         int bqp[2];

  740.         int rqp[2];

  741.         int mb_qp, mbn0_qp, mbn1_qp;

  742.         int i;

  743.         first_vertical_edge_done = 1;

  744. 

  745.         if( IS_INTRA(mb_type) ) {

  746.             AV_WN64A(&bS[0], 0x0004000400040004ULL);

  747.             AV_WN64A(&bS[4], 0x0004000400040004ULL);

  748.         } else {

  749.             static const uint8_t offset[2][2][8]={

  750.                 {

  751.                     {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},

  752.                     {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},

  753.                 },{

  754.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  755.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  756.                 }

  757.             };

  758.             const uint8_t *off= offset[MB_FIELD(h)][mb_y&1];

  759.             for( i = 0; i < 8; i++ ) {

  760.                 int j= MB_FIELD(h) ? i>>2 : i&1;

  761.                 int mbn_xy = h->left_mb_xy[LEFT(j)];

  762.                 int mbn_type= h->left_type[LEFT(j)];

  763. 

  764.                 if( IS_INTRA( mbn_type ) )

  765.                     bS[i] = 4;

  766.                 else{

  767.                     bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |

  768.                          ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?

  769.                             (h->cbp_table[mbn_xy] & (((MB_FIELD(h) ? (i&2) : (mb_y&1)) ? 8 : 2) << 12))

  770.                                                                        :

  771.                             h->non_zero_count[mbn_xy][ off[i] ]));

  772.                 }

  773.             }

  774.         }

  775. 

  776.         mb_qp   = h->cur_pic.qscale_table[mb_xy];

  777.         mbn0_qp = h->cur_pic.qscale_table[h->left_mb_xy[0]];

  778.         mbn1_qp = h->cur_pic.qscale_table[h->left_mb_xy[1]];

  779.         qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;

  780.         bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +

  781.                    get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;

  782.         rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +

  783.                    get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;

  784.         qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;

  785.         bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +

  786.                    get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;

  787.         rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +

  788.                    get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;

  789. 

  790.         /* Filter edge */

  791.         tprintf(h->avctx, "filter mb:%d/%d MBAFF, QPy:%d/%d, QPb:%d/%d QPr:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], bqp[0], bqp[1], rqp[0], rqp[1], linesize, uvlinesize);

  792.         { int i; for (i = 0; i < 8; i++) tprintf(h->avctx, " bS[%d]:%d", i, bS[i]); tprintf(h->avctx, "\n"); }

  793.         if (MB_FIELD(h)) {

  794.             filter_mb_mbaff_edgev ( h, img_y                ,   linesize, bS  , 1, qp [0], a, b, 1 );

  795.             filter_mb_mbaff_edgev ( h, img_y  + 8*  linesize,   linesize, bS+4, 1, qp [1], a, b, 1 );

  796.             if (chroma){

  797.                 if (CHROMA444(h)) {

  798.                     filter_mb_mbaff_edgev ( h, img_cb,                uvlinesize, bS  , 1, bqp[0], a, b, 1 );

  799.                     filter_mb_mbaff_edgev ( h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );

  800.                     filter_mb_mbaff_edgev ( h, img_cr,                uvlinesize, bS  , 1, rqp[0], a, b, 1 );

  801.                     filter_mb_mbaff_edgev ( h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );

  802.                 } else if (CHROMA422(h)) {

  803.                     filter_mb_mbaff_edgecv(h, img_cb,                uvlinesize, bS  , 1, bqp[0], a, b, 1);

  804.                     filter_mb_mbaff_edgecv(h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1);

  805.                     filter_mb_mbaff_edgecv(h, img_cr,                uvlinesize, bS  , 1, rqp[0], a, b, 1);

  806.                     filter_mb_mbaff_edgecv(h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1);

  807.                 }else{

  808.                     filter_mb_mbaff_edgecv( h, img_cb,                uvlinesize, bS  , 1, bqp[0], a, b, 1 );

  809.                     filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1], a, b, 1 );

  810.                     filter_mb_mbaff_edgecv( h, img_cr,                uvlinesize, bS  , 1, rqp[0], a, b, 1 );

  811.                     filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1], a, b, 1 );

  812.                 }

  813.             }

  814.         }else{

  815.             filter_mb_mbaff_edgev ( h, img_y              , 2*  linesize, bS  , 2, qp [0], a, b, 1 );

  816.             filter_mb_mbaff_edgev ( h, img_y  +   linesize, 2*  linesize, bS+1, 2, qp [1], a, b, 1 );

  817.             if (chroma){

  818.                 if (CHROMA444(h)) {

  819.                     filter_mb_mbaff_edgev ( h, img_cb,              2*uvlinesize, bS  , 2, bqp[0], a, b, 1 );

  820.                     filter_mb_mbaff_edgev ( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );

  821.                     filter_mb_mbaff_edgev ( h, img_cr,              2*uvlinesize, bS  , 2, rqp[0], a, b, 1 );

  822.                     filter_mb_mbaff_edgev ( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );

  823.                 }else{

  824.                     filter_mb_mbaff_edgecv( h, img_cb,              2*uvlinesize, bS  , 2, bqp[0], a, b, 1 );

  825.                     filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1], a, b, 1 );

  826.                     filter_mb_mbaff_edgecv( h, img_cr,              2*uvlinesize, bS  , 2, rqp[0], a, b, 1 );

  827.                     filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1], a, b, 1 );

  828.                 }

  829.             }

  830.         }

  831.     }

  832. 

  833. #if CONFIG_SMALL

  834.     {

  835.         int dir;

  836.         for (dir = 0; dir < 2; dir++)

  837.             filter_mb_dir(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize,

  838.                           uvlinesize, mb_xy, mb_type, mvy_limit,

  839.                           dir ? 0 : first_vertical_edge_done, a, b,

  840.                           chroma, dir);

  841.     }

  842. #else

  843.     filter_mb_dir(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, first_vertical_edge_done, a, b, chroma, 0);

  844.     filter_mb_dir(h, sl, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0,                        a, b, chroma, 1);

  845. #endif

  846. }