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.
43714 Commits
32 Branches
824MB
Tree: 6c180b35c4
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '6c180b35c4'
${ noResults }
FFmpeg/libavcodec/h264_loopfilter.c

830 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/intreadwrite.h"

  29. #include "internal.h"

  30. #include "dsputil.h"

  31. #include "avcodec.h"

  32. #include "mpegvideo.h"

  33. #include "h264.h"

  34. #include "mathops.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.     MpegEncContext * const s = &h->s;

  245.     int chroma = CHROMA && !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));

  246.     int chroma444 = CHROMA444;

  247.     int chroma422 = CHROMA422;

  248. 

  249.     int mb_xy = h->mb_xy;

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

  251.     int top_type= h->top_type;

  252. 

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

  254.     int a = h->slice_alpha_c0_offset - qp_bd_offset;

  255.     int b = h->slice_beta_offset - qp_bd_offset;

  256. 

  257.     int mb_type = s->current_picture.f.mb_type[mb_xy];

  258.     int qp      = s->current_picture.f.qscale_table[mb_xy];

  259.     int qp0     = s->current_picture.f.qscale_table[mb_xy - 1];

  260.     int qp1     = s->current_picture.f.qscale_table[h->top_mb_xy];

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

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

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

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

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

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

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

  268. 

  269.     if( IS_INTRA(mb_type) ) {

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

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

  272.         const int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;

  273.         if(left_type)

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

  275.         if( IS_8x8DCT(mb_type) ) {

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

  277.             if(top_type){

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

  279.             }

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

  281.         } else {

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

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

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

  285.             if(top_type){

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

  287.             }

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

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

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

  291.         }

  292.         if(chroma){

  293.             if(chroma444){

  294.                 if(left_type){

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

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

  297.                 }

  298.                 if( IS_8x8DCT(mb_type) ) {

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

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

  301.                     if(top_type){

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

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

  304.                     }

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

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

  307.                 } else {

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

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

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

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

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

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

  314.                     if(top_type){

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

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

  317.                     }

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

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

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

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

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

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

  324.                 }

  325.             }else if(chroma422){

  326.                 if(left_type){

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

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

  329.                 }

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

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

  332.                 if(top_type){

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

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

  335.                 }

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

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

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

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

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

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

  342.             }else{

  343.                 if(left_type){

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

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

  346.                 }

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

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

  349.                 if(top_type){

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

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

  352.                 }

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

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

  355.             }

  356.         }

  357.         return;

  358.     } else {

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

  360.         int edges;

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

  362.             edges = 4;

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

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

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

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

  367.         } else {

  368.             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;

  369.             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;

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

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

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

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

  374.         }

  375.         if( IS_INTRA(left_type) )

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

  377.         if( IS_INTRA(top_type) )

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

  379. 

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

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

  382.             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 );\

  383.             if(chroma){\

  384.                 if(chroma444){\

  385.                     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 );\

  386.                     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 );\

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

  388.                     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 );\

  389.                     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 );\

  390.                 }\

  391.             }\

  392.         }

  393.         if(left_type)

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

  395.         if( edges == 1 ) {

  396.             if(top_type)

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

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

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

  400.             if(top_type)

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

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

  403.         } else {

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

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

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

  407.             if(top_type)

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

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

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

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

  412.         }

  413. #undef FILTER

  414.     }

  415. }

  416. 

  417. void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {

  418.     av_assert2(!FRAME_MBAFF);

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

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

  421.         return;

  422.     }

  423. 

  424. #if CONFIG_SMALL

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

  426. #else

  427.     if(h->pixel_shift){

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

  429.     }else{

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

  431.     }

  432. #endif

  433. }

  434. 

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

  436.     int v;

  437. 

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

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

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

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

  442. 

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

  444.         if(!v)

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

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

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

  448. 

  449.         if(v){

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

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

  452.                 return 1;

  453.             return

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

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

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

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

  458.         }

  459.     }

  460. 

  461.     return v;

  462. }

  463. 

  464. static av_always_inline void filter_mb_dir(H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize, int mb_xy, int mb_type, int mvy_limit, int first_vertical_edge_done, int a, int b, int chroma, int dir) {

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

  466.     int edge;

  467.     int chroma_qp_avg[2];

  468.     int chroma444 = CHROMA444;

  469.     int chroma422 = CHROMA422;

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

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

  472. 

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

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

  475.                                               {0,3,1,1,3,3,3,3}};

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

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

  478. 

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

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

  481. 

  482.     if(mbm_type && !first_vertical_edge_done){

  483. 

  484.         if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)

  485.             && IS_INTERLACED(mbm_type&~mb_type)

  486.             ) {

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

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

  489.             // frame macroblock.

  490.             //

  491.             unsigned int tmp_linesize   = 2 *   linesize;

  492.             unsigned int tmp_uvlinesize = 2 * uvlinesize;

  493.             int mbn_xy = mb_xy - 2 * s->mb_stride;

  494.             int j;

  495. 

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

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

  498.                 int qp;

  499.                 if (IS_INTRA(mb_type | s->current_picture.f.mb_type[mbn_xy])) {

  500.                     AV_WN64A(bS, 0x0003000300030003ULL);

  501.                 } else {

  502.                     if (!CABAC && IS_8x8DCT(s->current_picture.f.mb_type[mbn_xy])) {

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

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

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

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

  507.                     }else{

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

  509.                     int i;

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

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

  512.                     }

  513.                     }

  514.                 }

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

  516.                 // value in IPCM macroblocks.

  517.                 qp = (s->current_picture.f.qscale_table[mb_xy] + s->current_picture.f.qscale_table[mbn_xy] + 1) >> 1;

  518.                 tprintf(s->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);

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

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

  521.                 chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mbn_xy]) + 1) >> 1;

  522.                 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mbn_xy]) + 1) >> 1;

  523.                 if (chroma) {

  524.                     if (chroma444) {

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

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

  527.                     } else {

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

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

  530.                     }

  531.                 }

  532.             }

  533.         }else{

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

  535.             int qp;

  536. 

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

  538.                 AV_WN64A(bS, 0x0003000300030003ULL);

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

  540.                     || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))

  541.                 )

  542.                     AV_WN64A(bS, 0x0004000400040004ULL);

  543.             } else {

  544.                 int i;

  545.                 int mv_done;

  546. 

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

  548.                     AV_WN64A(bS, 0x0001000100010001ULL);

  549.                     mv_done = 1;

  550.                 }

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

  552.                     int b_idx= 8 + 4;

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

  554. 

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

  556.                     mv_done = 1;

  557.                 }

  558.                 else

  559.                     mv_done = 0;

  560. 

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

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

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

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

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

  566. 

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

  568.                         h->non_zero_count_cache[bn_idx] ) {

  569.                         bS[i] = 2;

  570.                     }

  571.                     else if(!mv_done)

  572.                     {

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

  574.                     }

  575.                 }

  576.             }

  577. 

  578.             /* Filter edge */

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

  580.             // value in IPCM macroblocks.

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

  582.                 qp = (s->current_picture.f.qscale_table[mb_xy] + s->current_picture.f.qscale_table[mbm_xy] + 1) >> 1;

  583.                 //tprintf(s->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], s->current_picture.qscale_table[mbn_xy]);

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

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

  586.                 chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mbm_xy]) + 1) >> 1;

  587.                 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mbm_xy]) + 1) >> 1;

  588.                 if( dir == 0 ) {

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

  590.                     if (chroma) {

  591.                         if (chroma444) {

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

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

  594.                         } else {

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

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

  597.                         }

  598.                     }

  599.                 } else {

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

  601.                     if (chroma) {

  602.                         if (chroma444) {

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

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

  605.                         } else {

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

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

  608.                         }

  609.                     }

  610.                 }

  611.             }

  612.         }

  613.     }

  614. 

  615.     /* Calculate bS */

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

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

  618.         int qp;

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

  620. 

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

  622.             continue;

  623. 

  624.         if( IS_INTRA(mb_type)) {

  625.             AV_WN64A(bS, 0x0003000300030003ULL);

  626.         } else {

  627.             int i;

  628.             int mv_done;

  629. 

  630.             if( edge & mask_edge ) {

  631.                 AV_ZERO64(bS);

  632.                 mv_done = 1;

  633.             }

  634.             else if( mask_par0 ) {

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

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

  637. 

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

  639.                 mv_done = 1;

  640.             }

  641.             else

  642.                 mv_done = 0;

  643. 

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

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

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

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

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

  649. 

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

  651.                     h->non_zero_count_cache[bn_idx] ) {

  652.                     bS[i] = 2;

  653.                 }

  654.                 else if(!mv_done)

  655.                 {

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

  657.                 }

  658.             }

  659. 

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

  661.                 continue;

  662.         }

  663. 

  664.         /* Filter edge */

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

  666.         // value in IPCM macroblocks.

  667.         qp = s->current_picture.f.qscale_table[mb_xy];

  668.         //tprintf(s->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], s->current_picture.qscale_table[mbn_xy]);

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

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

  671.         if( dir == 0 ) {

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

  673.             if (chroma) {

  674.                 if (chroma444) {

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

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

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

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

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

  680.                 }

  681.             }

  682.         } else {

  683.             if (chroma422) {

  684.                 if (deblock_edge)

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

  686.                 if (chroma) {

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

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

  689.                 }

  690.             } else {

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

  692.                 if (chroma) {

  693.                     if (chroma444) {

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

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

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

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

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

  699.                     }

  700.                 }

  701.             }

  702.         }

  703.     }

  704. }

  705. 

  706. void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {

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

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

  709.     const int mb_type = s->current_picture.f.mb_type[mb_xy];

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

  711.     int first_vertical_edge_done = 0;

  712.     av_unused int dir;

  713.     int chroma = CHROMA && !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));

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

  715.     int a = h->slice_alpha_c0_offset - qp_bd_offset;

  716.     int b = h->slice_beta_offset - qp_bd_offset;

  717. 

  718.     if (FRAME_MBAFF

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

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

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

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

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

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

  725.          */

  726.         DECLARE_ALIGNED(8, int16_t, bS)[8];

  727.         int qp[2];

  728.         int bqp[2];

  729.         int rqp[2];

  730.         int mb_qp, mbn0_qp, mbn1_qp;

  731.         int i;

  732.         first_vertical_edge_done = 1;

  733. 

  734.         if( IS_INTRA(mb_type) ) {

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

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

  737.         } else {

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

  739.                 {

  740.                     {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},

  741.                     {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},

  742.                 },{

  743.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  744.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  745.                 }

  746.             };

  747.             const uint8_t *off= offset[MB_FIELD][mb_y&1];

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

  749.                 int j= MB_FIELD ? i>>2 : i&1;

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

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

  752. 

  753.                 if( IS_INTRA( mbn_type ) )

  754.                     bS[i] = 4;

  755.                 else{

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

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

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

  759.                                                                        :

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

  761.                 }

  762.             }

  763.         }

  764. 

  765.         mb_qp   = s->current_picture.f.qscale_table[mb_xy];

  766.         mbn0_qp = s->current_picture.f.qscale_table[h->left_mb_xy[0]];

  767.         mbn1_qp = s->current_picture.f.qscale_table[h->left_mb_xy[1]];

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

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

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

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

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

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

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

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

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

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

  778. 

  779.         /* Filter edge */

  780.         tprintf(s->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);

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

  782.         if(MB_FIELD){

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

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

  785.             if (chroma){

  786.                 if (CHROMA444) {

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

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

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

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

  791.                 } else if (CHROMA422) {

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

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

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

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

  796.                 }else{

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

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

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

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

  801.                 }

  802.             }

  803.         }else{

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

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

  806.             if (chroma){

  807.                 if (CHROMA444) {

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

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

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

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

  812.                 }else{

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

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

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

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

  817.                 }

  818.             }

  819.         }

  820.     }

  821. 

  822. #if CONFIG_SMALL

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

  824.         filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, dir ? 0 : first_vertical_edge_done, a, b, chroma, dir);

  825. #else

  826.     filter_mb_dir(h, 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);

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

  828. #endif

  829. }