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.
43166 Commits
32 Branches
824MB
Tree: 41ed7ab45f
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '41ed7ab45f'
${ 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 Libav.

  6.  *

  7.  * Libav 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.  * Libav 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 Libav; 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 / MPEG-4 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. #include <assert.h>

  38. 

  39. /* Deblocking filter (p153) */

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

  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,  0,  0,  0,

  44.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  45.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  46.      0,  0,  0,  0,  0,  0,  4,  4,  5,  6,

  47.      7,  8,  9, 10, 12, 13, 15, 17, 20, 22,

  48.     25, 28, 32, 36, 40, 45, 50, 56, 63, 71,

  49.     80, 90,101,113,127,144,162,182,203,226,

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

  54.    255,255,255,255,255,255,255,255,255,255,255,255,255,

  55. };

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

  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,  0,  0,  0,

  60.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  61.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  62.      0,  0,  0,  0,  0,  0,  2,  2,  2,  3,

  63.      3,  3,  3,  4,  4,  4,  6,  6,  7,  7,

  64.      8,  8,  9,  9, 10, 10, 11, 11, 12, 12,

  65.     13, 13, 14, 14, 15, 15, 16, 16, 17, 17,

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

  70.     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,

  71. };

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

  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 }, {-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 },

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

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

  84.     {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 1 },

  85.     {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },

  86.     {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },

  87.     {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },

  88.     {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },

  89.     {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },

  90.     {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-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 }, {-1,13,17,25 }, {-1,13,17,25 },

  98.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  99.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  100. };

  101. 

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

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

  104.                                              const int16_t bS[4],

  105.                                              unsigned int qp, int a, int b,

  106.                                              const H264Context *h, int intra)

  107. {

  108.     const unsigned int index_a = qp + a;

  109.     const int alpha = alpha_table[index_a];

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

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

  112. 

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

  114.         int8_t tc[4];

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

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

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

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

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

  120.     } else {

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

  122.     }

  123. }

  124. 

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

  126.                                               const int16_t bS[4],

  127.                                               unsigned int qp, int a, int b,

  128.                                               const H264Context *h, int intra)

  129. {

  130.     const unsigned int index_a = qp + a;

  131.     const int alpha = alpha_table[index_a];

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

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

  134. 

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

  136.         int8_t tc[4];

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

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

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

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

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

  142.     } else {

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

  144.     }

  145. }

  146. 

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

  148.                                                    int stride,

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

  150.                                                    int qp, int a, int b,

  151.                                                    int intra)

  152. {

  153.     const unsigned int index_a = qp + a;

  154.     const int alpha = alpha_table[index_a];

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

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

  157. 

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

  159.         int8_t tc[4];

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

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

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

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

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

  165.     } else {

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

  167.     }

  168. }

  169. 

  170. static av_always_inline void filter_mb_mbaff_edgecv(const H264Context *h,

  171.                                                     uint8_t *pix, int stride,

  172.                                                     const int16_t bS[7],

  173.                                                     int bsi, int qp, int a,

  174.                                                     int b, int intra)

  175. {

  176.     const unsigned int index_a = qp + a;

  177.     const int alpha = alpha_table[index_a];

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

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

  180. 

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

  182.         int8_t tc[4];

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

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

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

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

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

  188.     } else {

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

  190.     }

  191. }

  192. 

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

  194.                                              const int16_t bS[4],

  195.                                              unsigned int qp, int a, int b,

  196.                                              const H264Context *h, int intra)

  197. {

  198.     const unsigned int index_a = qp + a;

  199.     const int alpha = alpha_table[index_a];

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

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

  202. 

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

  204.         int8_t tc[4];

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

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

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

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

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

  210.     } else {

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

  212.     }

  213. }

  214. 

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

  216.                                               const int16_t bS[4],

  217.                                               unsigned int qp, int a, int b,

  218.                                               const H264Context *h, int intra)

  219. {

  220.     const unsigned int index_a = qp + a;

  221.     const int alpha = alpha_table[index_a];

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

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

  224. 

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

  226.         int8_t tc[4];

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

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

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

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

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

  232.     } else {

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

  234.     }

  235. }

  236. 

  237. static av_always_inline void h264_filter_mb_fast_internal(const H264Context *h,

  238.                                                           H264SliceContext *sl,

  239.                                                           int mb_x, int mb_y,

  240.                                                           uint8_t *img_y,

  241.                                                           uint8_t *img_cb,

  242.                                                           uint8_t *img_cr,

  243.                                                           unsigned int linesize,

  244.                                                           unsigned int uvlinesize,

  245.                                                           int pixel_shift)

  246. {

  247.     int chroma = !(CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));

  248.     int chroma444 = CHROMA444(h);

  249.     int chroma422 = CHROMA422(h);

  250. 

  251.     int mb_xy = sl->mb_xy;

  252.     int left_type = sl->left_type[LTOP];

  253.     int top_type  = sl->top_type;

  254. 

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

  256.     int a = 52 + sl->slice_alpha_c0_offset - qp_bd_offset;

  257.     int b = 52 + sl->slice_beta_offset - qp_bd_offset;

  258. 

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

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

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

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

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

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

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

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

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

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

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

  270. 

  271.     if( IS_INTRA(mb_type) ) {

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

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

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

  275.         if(left_type)

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

  277.         if( IS_8x8DCT(mb_type) ) {

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

  279.             if(top_type){

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

  281.             }

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

  283.         } else {

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

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

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

  287.             if(top_type){

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

  289.             }

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

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

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

  293.         }

  294.         if(chroma){

  295.             if(chroma444){

  296.                 if(left_type){

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

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

  299.                 }

  300.                 if( IS_8x8DCT(mb_type) ) {

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

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

  303.                     if(top_type){

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

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

  306.                     }

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

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

  309.                 } else {

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

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

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

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

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

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

  316.                     if(top_type){

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

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

  319.                     }

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

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

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

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

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

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

  326.                 }

  327.             }else if(chroma422){

  328.                 if(left_type){

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

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

  331.                 }

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

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

  334.                 if(top_type){

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

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

  337.                 }

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

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

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

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

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

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

  344.             }else{

  345.                 if(left_type){

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

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

  348.                 }

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

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

  351.                 if(top_type){

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

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

  354.                 }

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

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

  357.             }

  358.         }

  359.         return;

  360.     } else {

  361.         LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);

  362.         int edges;

  363.         if( IS_8x8DCT(mb_type) && (sl->cbp&7) == 7 && !chroma444 ) {

  364.             edges = 4;

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

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

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

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

  369.         } else {

  370.             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;

  371.             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;

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

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

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

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

  376.         }

  377.         if( IS_INTRA(left_type) )

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

  379.         if( IS_INTRA(top_type) )

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

  381. 

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

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

  384.             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 );\

  385.             if(chroma){\

  386.                 if(chroma444){\

  387.                     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 );\

  388.                     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 );\

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

  390.                     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 );\

  391.                     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 );\

  392.                 }\

  393.             }\

  394.         }

  395.         if(left_type)

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

  397.         if( edges == 1 ) {

  398.             if(top_type)

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

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

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

  402.             if(top_type)

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

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

  405.         } else {

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

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

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

  409.             if(top_type)

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

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

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

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

  414.         }

  415. #undef FILTER

  416.     }

  417. }

  418. 

  419. void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl,

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

  421.                             uint8_t *img_cb, uint8_t *img_cr,

  422.                             unsigned int linesize, unsigned int uvlinesize)

  423. {

  424.     assert(!FRAME_MBAFF(h));

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

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

  427.         return;

  428.     }

  429. 

  430. #if CONFIG_SMALL

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

  432. #else

  433.     if(h->pixel_shift){

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

  435.     }else{

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

  437.     }

  438. #endif

  439. }

  440. 

  441. static int check_mv(H264SliceContext *sl, long b_idx, long bn_idx, int mvy_limit)

  442. {

  443.     int v;

  444. 

  445.     v = sl->ref_cache[0][b_idx] != sl->ref_cache[0][bn_idx];

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

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

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

  449. 

  450.     if (sl->list_count == 2) {

  451.         if(!v)

  452.             v = sl->ref_cache[1][b_idx] != sl->ref_cache[1][bn_idx] |

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

  454.                 FFABS(sl->mv_cache[1][b_idx][1] - sl->mv_cache[1][bn_idx][1]) >= mvy_limit;

  455. 

  456.         if(v){

  457.             if (sl->ref_cache[0][b_idx] != sl->ref_cache[1][bn_idx] |

  458.                 sl->ref_cache[1][b_idx] != sl->ref_cache[0][bn_idx])

  459.                 return 1;

  460.             return

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

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

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

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

  465.         }

  466.     }

  467. 

  468.     return v;

  469. }

  470. 

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

  472.                                            int mb_x, int mb_y,

  473.                                            uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,

  474.                                            unsigned int linesize, unsigned int uvlinesize,

  475.                                            int mb_xy, int mb_type, int mvy_limit,

  476.                                            int first_vertical_edge_done, int a, int b,

  477.                                            int chroma, int dir)

  478. {

  479.     int edge;

  480.     int chroma_qp_avg[2];

  481.     int chroma444 = CHROMA444(h);

  482.     int chroma422 = CHROMA422(h);

  483.     const int mbm_xy = dir == 0 ? mb_xy -1 : sl->top_mb_xy;

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

  485. 

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

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

  488.                                               {0,3,1,1,3,3,3,3}};

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

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

  491. 

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

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

  494. 

  495.     if(mbm_type && !first_vertical_edge_done){

  496. 

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

  498.             && IS_INTERLACED(mbm_type&~mb_type)

  499.             ) {

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

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

  502.             // frame macroblock.

  503.             //

  504.             unsigned int tmp_linesize   = 2 *   linesize;

  505.             unsigned int tmp_uvlinesize = 2 * uvlinesize;

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

  507.             int j;

  508. 

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

  510.                 DECLARE_ALIGNED(8, int16_t, bS)[4];

  511.                 int qp;

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

  513.                     AV_WN64A(bS, 0x0003000300030003ULL);

  514.                 } else {

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

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

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

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

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

  520.                     }else{

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

  522.                     int i;

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

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

  525.                     }

  526.                     }

  527.                 }

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

  529.                 // value in IPCM macroblocks.

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

  531.                 ff_tlog(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);

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

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

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

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

  536.                 if (chroma) {

  537.                     if (chroma444) {

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

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

  540.                     } else {

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

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

  543.                     }

  544.                 }

  545.             }

  546.         }else{

  547.             DECLARE_ALIGNED(8, int16_t, bS)[4];

  548.             int qp;

  549. 

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

  551.                 AV_WN64A(bS, 0x0003000300030003ULL);

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

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

  554.                 )

  555.                     AV_WN64A(bS, 0x0004000400040004ULL);

  556.             } else {

  557.                 int i;

  558.                 int mv_done;

  559. 

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

  561.                     AV_WN64A(bS, 0x0001000100010001ULL);

  562.                     mv_done = 1;

  563.                 }

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

  565.                     int b_idx= 8 + 4;

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

  567. 

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

  569.                     mv_done = 1;

  570.                 }

  571.                 else

  572.                     mv_done = 0;

  573. 

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

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

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

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

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

  579. 

  580.                     if (sl->non_zero_count_cache[b_idx] |

  581.                         sl->non_zero_count_cache[bn_idx]) {

  582.                         bS[i] = 2;

  583.                     }

  584.                     else if(!mv_done)

  585.                     {

  586.                         bS[i] = check_mv(sl, b_idx, bn_idx, mvy_limit);

  587.                     }

  588.                 }

  589.             }

  590. 

  591.             /* Filter edge */

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

  593.             // value in IPCM macroblocks.

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

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

  596.                 ff_tlog(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);

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

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

  599.                 if( dir == 0 ) {

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

  601.                     if (chroma) {

  602.                         if (chroma444) {

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

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

  605.                         } else {

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

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

  608.                         }

  609.                     }

  610.                 } else {

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

  612.                     if (chroma) {

  613.                         if (chroma444) {

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

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

  616.                         } else {

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

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

  619.                         }

  620.                     }

  621.                 }

  622.             }

  623.         }

  624.     }

  625. 

  626.     /* Calculate bS */

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

  628.         DECLARE_ALIGNED(8, int16_t, bS)[4];

  629.         int qp;

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

  631. 

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

  633.             continue;

  634. 

  635.         if( IS_INTRA(mb_type)) {

  636.             AV_WN64A(bS, 0x0003000300030003ULL);

  637.         } else {

  638.             int i;

  639.             int mv_done;

  640. 

  641.             if( edge & mask_edge ) {

  642.                 AV_ZERO64(bS);

  643.                 mv_done = 1;

  644.             }

  645.             else if( mask_par0 ) {

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

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

  648. 

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

  650.                 mv_done = 1;

  651.             }

  652.             else

  653.                 mv_done = 0;

  654. 

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

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

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

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

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

  660. 

  661.                 if (sl->non_zero_count_cache[b_idx] |

  662.                     sl->non_zero_count_cache[bn_idx]) {

  663.                     bS[i] = 2;

  664.                 }

  665.                 else if(!mv_done)

  666.                 {

  667.                     bS[i] = check_mv(sl, b_idx, bn_idx, mvy_limit);

  668.                 }

  669.             }

  670. 

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

  672.                 continue;

  673.         }

  674. 

  675.         /* Filter edge */

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

  677.         // value in IPCM macroblocks.

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

  679.         ff_tlog(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);

  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(const 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 = !(CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));

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

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

  727.     int b = 52 + sl->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 ^ sl->left_type[LTOP])

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

  733.             && sl->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.         DECLARE_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(sl)][mb_y&1];

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

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

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

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

  763. 

  764.                 if( IS_INTRA( mbn_type ) )

  765.                     bS[i] = 4;

  766.                 else{

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

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

  769.                             (h->cbp_table[mbn_xy] & (((MB_FIELD(sl) ? (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[sl->left_mb_xy[0]];

  778.         mbn1_qp = h->cur_pic.qscale_table[sl->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.         ff_tlog(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++) ff_tlog(h->avctx, " bS[%d]:%d", i, bS[i]); ff_tlog(h->avctx, "\n"); }

  793.         if (MB_FIELD(sl)) {

  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. }