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.
34063 Commits
32 Branches
283MB
Tree: 25b51b2c44
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '25b51b2c44'
${ noResults }
FFmpeg/libavcodec/h264_loopfilter.c

833 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 / 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. //#undef NDEBUG

  38. #include <assert.h>

  39. 

  40. /* Deblocking filter (p153) */

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

  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,  0,  0,  0,

  46.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  47.      0,  0,  0,  0,  0,  0,  4,  4,  5,  6,

  48.      7,  8,  9, 10, 12, 13, 15, 17, 20, 22,

  49.     25, 28, 32, 36, 40, 45, 50, 56, 63, 71,

  50.     80, 90,101,113,127,144,162,182,203,226,

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

  56. };

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

  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,  0,  0,  0,

  62.      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,

  63.      0,  0,  0,  0,  0,  0,  2,  2,  2,  3,

  64.      3,  3,  3,  4,  4,  4,  6,  6,  7,  7,

  65.      8,  8,  9,  9, 10, 10, 11, 11, 12, 12,

  66.     13, 13, 14, 14, 15, 15, 16, 16, 17, 17,

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

  72. };

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

  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 }, {-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 },

  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, 0 },

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

  86.     {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },

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

  88.     {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },

  89.     {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },

  90.     {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },

  91.     {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-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 }, {-1,13,17,25 }, {-1,13,17,25 },

  100.     {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },

  101. };

  102. 

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

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

  105.                                              const int16_t bS[4],

  106.                                              unsigned int qp, int a, int b,

  107.                                              H264Context *h, int intra)

  108. {

  109.     const unsigned int index_a = qp + a;

  110.     const int alpha = alpha_table[index_a];

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

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

  113. 

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

  115.         int8_t tc[4];

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

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

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

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

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

  121.     } else {

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

  123.     }

  124. }

  125. 

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

  127.                                               const int16_t bS[4],

  128.                                               unsigned int qp, int a, int b,

  129.                                               H264Context *h, int intra)

  130. {

  131.     const unsigned int index_a = qp + a;

  132.     const int alpha = alpha_table[index_a];

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

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

  135. 

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

  137.         int8_t tc[4];

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

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

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

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

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

  143.     } else {

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

  145.     }

  146. }

  147. 

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

  149.                                                    int stride,

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

  151.                                                    int qp, int a, int b,

  152.                                                    int intra)

  153. {

  154.     const unsigned int index_a = qp + a;

  155.     const int alpha = alpha_table[index_a];

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

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

  158. 

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

  160.         int8_t tc[4];

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

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

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

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

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

  166.     } else {

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

  168.     }

  169. }

  170. 

  171. static av_always_inline void filter_mb_mbaff_edgecv(H264Context *h,

  172.                                                     uint8_t *pix, int stride,

  173.                                                     const int16_t bS[7],

  174.                                                     int bsi, int qp, int a,

  175.                                                     int b, int intra)

  176. {

  177.     const unsigned int index_a = qp + a;

  178.     const int alpha = alpha_table[index_a];

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

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

  181. 

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

  183.         int8_t tc[4];

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

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

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

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

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

  189.     } else {

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

  191.     }

  192. }

  193. 

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

  195.                                              const int16_t bS[4],

  196.                                              unsigned int qp, int a, int b,

  197.                                              H264Context *h, int intra)

  198. {

  199.     const unsigned int index_a = qp + a;

  200.     const int alpha = alpha_table[index_a];

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

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

  203. 

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

  205.         int8_t tc[4];

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

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

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

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

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

  211.     } else {

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

  213.     }

  214. }

  215. 

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

  217.                                               const int16_t bS[4],

  218.                                               unsigned int qp, int a, int b,

  219.                                               H264Context *h, int intra)

  220. {

  221.     const unsigned int index_a = qp + a;

  222.     const int alpha = alpha_table[index_a];

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

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

  225. 

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

  227.         int8_t tc[4];

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

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

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

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

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

  233.     } else {

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

  235.     }

  236. }

  237. 

  238. static av_always_inline void h264_filter_mb_fast_internal(H264Context *h,

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

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

  249.     int chroma444 = CHROMA444;

  250.     int chroma422 = CHROMA422;

  251. 

  252.     int mb_xy = h->mb_xy;

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

  254.     int top_type= h->top_type;

  255. 

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

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

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

  259. 

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

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

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

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

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

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

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

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

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

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

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

  271. 

  272.     if( IS_INTRA(mb_type) ) {

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

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

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

  276.         if(left_type)

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

  278.         if( IS_8x8DCT(mb_type) ) {

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

  280.             if(top_type){

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

  282.             }

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

  284.         } else {

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

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

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

  288.             if(top_type){

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

  290.             }

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

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

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

  294.         }

  295.         if(chroma){

  296.             if(chroma444){

  297.                 if(left_type){

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

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

  300.                 }

  301.                 if( IS_8x8DCT(mb_type) ) {

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

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

  304.                     if(top_type){

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

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

  307.                     }

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

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

  310.                 } else {

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

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

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

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

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

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

  317.                     if(top_type){

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

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

  320.                     }

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

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

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

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

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

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

  327.                 }

  328.             }else if(chroma422){

  329.                 if(left_type){

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

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

  332.                 }

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

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

  335.                 if(top_type){

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

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

  338.                 }

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

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

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

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

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

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

  345.             }else{

  346.                 if(left_type){

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

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

  349.                 }

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

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

  352.                 if(top_type){

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

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

  355.                 }

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

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

  358.             }

  359.         }

  360.         return;

  361.     } else {

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

  363.         int edges;

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

  365.             edges = 4;

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

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

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

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

  370.         } else {

  371.             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;

  372.             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;

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

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

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

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

  377.         }

  378.         if( IS_INTRA(left_type) )

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

  380.         if( IS_INTRA(top_type) )

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

  382. 

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

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

  385.             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 );\

  386.             if(chroma){\

  387.                 if(chroma444){\

  388.                     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 );\

  389.                     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 );\

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

  391.                     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 );\

  392.                     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 );\

  393.                 }\

  394.             }\

  395.         }

  396.         if(left_type)

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

  398.         if( edges == 1 ) {

  399.             if(top_type)

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

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

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

  403.             if(top_type)

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

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

  406.         } else {

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

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

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

  410.             if(top_type)

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

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

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

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

  415.         }

  416. #undef FILTER

  417.     }

  418. }

  419. 

  420. 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) {

  421.     assert(!FRAME_MBAFF);

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

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

  424.         return;

  425.     }

  426. 

  427. #if CONFIG_SMALL

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

  429. #else

  430.     if(h->pixel_shift){

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

  432.     }else{

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

  434.     }

  435. #endif

  436. }

  437. 

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

  439.     int v;

  440. 

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

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

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

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

  445. 

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

  447.         if(!v)

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

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

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

  451. 

  452.         if(v){

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

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

  455.                 return 1;

  456.             return

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

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

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

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

  461.         }

  462.     }

  463. 

  464.     return v;

  465. }

  466. 

  467. static av_always_inline void filter_mb_dir(H264Context *h, 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) {

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

  469.     int edge;

  470.     int chroma_qp_avg[2];

  471.     int chroma444 = CHROMA444;

  472.     int chroma422 = CHROMA422;

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

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

  475. 

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

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

  478.                                               {0,3,1,1,3,3,3,3}};

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

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

  481. 

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

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

  484. 

  485.     if(mbm_type && !first_vertical_edge_done){

  486. 

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

  488.             && IS_INTERLACED(mbm_type&~mb_type)

  489.             ) {

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

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

  492.             // frame macroblock.

  493.             //

  494.             unsigned int tmp_linesize   = 2 *   linesize;

  495.             unsigned int tmp_uvlinesize = 2 * uvlinesize;

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

  497.             int j;

  498. 

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

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

  501.                 int qp;

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

  503.                     AV_WN64A(bS, 0x0003000300030003ULL);

  504.                 } else {

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

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

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

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

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

  510.                     }else{

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

  512.                     int i;

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

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

  515.                     }

  516.                     }

  517.                 }

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

  519.                 // value in IPCM macroblocks.

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

  521.                 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);

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

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

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

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

  526.                 if (chroma) {

  527.                     if (chroma444) {

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

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

  530.                     } else {

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

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

  533.                     }

  534.                 }

  535.             }

  536.         }else{

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

  538.             int qp;

  539. 

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

  541.                 AV_WN64A(bS, 0x0003000300030003ULL);

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

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

  544.                 )

  545.                     AV_WN64A(bS, 0x0004000400040004ULL);

  546.             } else {

  547.                 int i;

  548.                 int mv_done;

  549. 

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

  551.                     AV_WN64A(bS, 0x0001000100010001ULL);

  552.                     mv_done = 1;

  553.                 }

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

  555.                     int b_idx= 8 + 4;

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

  557. 

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

  559.                     mv_done = 1;

  560.                 }

  561.                 else

  562.                     mv_done = 0;

  563. 

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

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

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

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

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

  569. 

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

  571.                         h->non_zero_count_cache[bn_idx] ) {

  572.                         bS[i] = 2;

  573.                     }

  574.                     else if(!mv_done)

  575.                     {

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

  577.                     }

  578.                 }

  579.             }

  580. 

  581.             /* Filter edge */

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

  583.             // value in IPCM macroblocks.

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

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

  586.                 //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]);

  587.                 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);

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

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

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

  591.                 if( dir == 0 ) {

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

  593.                     if (chroma) {

  594.                         if (chroma444) {

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

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

  597.                         } else {

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

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

  600.                         }

  601.                     }

  602.                 } else {

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

  604.                     if (chroma) {

  605.                         if (chroma444) {

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

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

  608.                         } else {

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

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

  611.                         }

  612.                     }

  613.                 }

  614.             }

  615.         }

  616.     }

  617. 

  618.     /* Calculate bS */

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

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

  621.         int qp;

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

  623. 

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

  625.             continue;

  626. 

  627.         if( IS_INTRA(mb_type)) {

  628.             AV_WN64A(bS, 0x0003000300030003ULL);

  629.         } else {

  630.             int i;

  631.             int mv_done;

  632. 

  633.             if( edge & mask_edge ) {

  634.                 AV_ZERO64(bS);

  635.                 mv_done = 1;

  636.             }

  637.             else if( mask_par0 ) {

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

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

  640. 

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

  642.                 mv_done = 1;

  643.             }

  644.             else

  645.                 mv_done = 0;

  646. 

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

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

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

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

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

  652. 

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

  654.                     h->non_zero_count_cache[bn_idx] ) {

  655.                     bS[i] = 2;

  656.                 }

  657.                 else if(!mv_done)

  658.                 {

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

  660.                 }

  661.             }

  662. 

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

  664.                 continue;

  665.         }

  666. 

  667.         /* Filter edge */

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

  669.         // value in IPCM macroblocks.

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

  671.         //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]);

  672.         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);

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

  674.         if( dir == 0 ) {

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

  676.             if (chroma) {

  677.                 if (chroma444) {

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

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

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

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

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

  683.                 }

  684.             }

  685.         } else {

  686.             if (chroma422) {

  687.                 if (deblock_edge)

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

  689.                 if (chroma) {

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

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

  692.                 }

  693.             } else {

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

  695.                 if (chroma) {

  696.                     if (chroma444) {

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

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

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

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

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

  702.                     }

  703.                 }

  704.             }

  705.         }

  706.     }

  707. }

  708. 

  709. 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) {

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

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

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

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

  714.     int first_vertical_edge_done = 0;

  715.     av_unused int dir;

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

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

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

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

  720. 

  721.     if (FRAME_MBAFF

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

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

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

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

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

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

  728.          */

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

  730.         int qp[2];

  731.         int bqp[2];

  732.         int rqp[2];

  733.         int mb_qp, mbn0_qp, mbn1_qp;

  734.         int i;

  735.         first_vertical_edge_done = 1;

  736. 

  737.         if( IS_INTRA(mb_type) ) {

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

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

  740.         } else {

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

  742.                 {

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

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

  745.                 },{

  746.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  747.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  748.                 }

  749.             };

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

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

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

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

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

  755. 

  756.                 if( IS_INTRA( mbn_type ) )

  757.                     bS[i] = 4;

  758.                 else{

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

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

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

  762.                                                                        :

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

  764.                 }

  765.             }

  766.         }

  767. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  781. 

  782.         /* Filter edge */

  783.         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);

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

  785.         if(MB_FIELD){

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

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

  788.             if (chroma){

  789.                 if (CHROMA444) {

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

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

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

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

  794.                 } else if (CHROMA422) {

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

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

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

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

  799.                 }else{

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

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

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

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

  804.                 }

  805.             }

  806.         }else{

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

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

  809.             if (chroma){

  810.                 if (CHROMA444) {

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

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

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

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

  815.                 }else{

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

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

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

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

  820.                 }

  821.             }

  822.         }

  823.     }

  824. 

  825. #if CONFIG_SMALL

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

  827.         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);

  828. #else

  829.     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);

  830.     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);

  831. #endif

  832. }