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.
39979 Commits
32 Branches
283MB
Tree: 593886b5dd
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '593886b5dd'
${ noResults }
FFmpeg/libavcodec/h264_loopfilter.c

825 lines
41KB
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/internal.h"

  29. #include "libavutil/intreadwrite.h"

  30. #include "internal.h"

  31. #include "avcodec.h"

  32. #include "h264.h"

  33. #include "mathops.h"

  34. #include "mpegutils.h"

  35. #include "rectangle.h"

  36. 

  37. #include <assert.h>

  38. 

  39. /* Deblocking filter (p153) */

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

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

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

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

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

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

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

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

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

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

  50.    255,255,

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

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

  53.    255,255,255,255,255,255,255,255,255,255,255,255,255,

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

  55. };

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

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

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

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

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

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

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

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

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

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

  66.     18, 18,

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

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

  69.     18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,

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

  71. };

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  100. };

  101. 

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

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

  104.                                              const int16_t bS[4],

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

  106.                                              H264Context *h, int intra)

  107. {

  108.     const unsigned int index_a = qp + a;

  109.     const int alpha = alpha_table[index_a];

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

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

  112. 

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

  114.         int8_t tc[4];

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

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

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

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

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

  120.     } else {

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

  122.     }

  123. }

  124. 

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

  126.                                               const int16_t bS[4],

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

  128.                                               H264Context *h, int intra)

  129. {

  130.     const unsigned int index_a = qp + a;

  131.     const int alpha = alpha_table[index_a];

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

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

  134. 

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

  136.         int8_t tc[4];

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

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

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

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

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

  142.     } else {

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

  144.     }

  145. }

  146. 

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

  148.                                                    int stride,

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

  150.                                                    int qp, int a, int b,

  151.                                                    int intra)

  152. {

  153.     const unsigned int index_a = qp + a;

  154.     const int alpha = alpha_table[index_a];

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

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

  157. 

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

  159.         int8_t tc[4];

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

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

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

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

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

  165.     } else {

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

  167.     }

  168. }

  169. 

  170. static av_always_inline void filter_mb_mbaff_edgecv(H264Context *h,

  171.                                                     uint8_t *pix, int stride,

  172.                                                     const int16_t bS[7],

  173.                                                     int bsi, int qp, int a,

  174.                                                     int b, int intra)

  175. {

  176.     const unsigned int index_a = qp + a;

  177.     const int alpha = alpha_table[index_a];

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

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

  180. 

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

  182.         int8_t tc[4];

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

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

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

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

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

  188.     } else {

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

  190.     }

  191. }

  192. 

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

  194.                                              const int16_t bS[4],

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

  196.                                              H264Context *h, int intra)

  197. {

  198.     const unsigned int index_a = qp + a;

  199.     const int alpha = alpha_table[index_a];

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

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

  202. 

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

  204.         int8_t tc[4];

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

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

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

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

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

  210.     } else {

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

  212.     }

  213. }

  214. 

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

  216.                                               const int16_t bS[4],

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

  218.                                               H264Context *h, int intra)

  219. {

  220.     const unsigned int index_a = qp + a;

  221.     const int alpha = alpha_table[index_a];

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

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

  224. 

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

  226.         int8_t tc[4];

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

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

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

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

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

  232.     } else {

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

  234.     }

  235. }

  236. 

  237. static av_always_inline void h264_filter_mb_fast_internal(H264Context *h,

  238.                                                           int mb_x, int mb_y,

  239.                                                           uint8_t *img_y,

  240.                                                           uint8_t *img_cb,

  241.                                                           uint8_t *img_cr,

  242.                                                           unsigned int linesize,

  243.                                                           unsigned int uvlinesize,

  244.                                                           int pixel_shift)

  245. {

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

  247.     int chroma444 = CHROMA444(h);

  248.     int chroma422 = CHROMA422(h);

  249. 

  250.     int mb_xy = h->mb_xy;

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

  252.     int top_type= h->top_type;

  253. 

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

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

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

  257. 

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

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

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

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

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

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

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

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

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

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

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

  269. 

  270.     if( IS_INTRA(mb_type) ) {

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

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

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

  274.         if(left_type)

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

  276.         if( IS_8x8DCT(mb_type) ) {

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

  278.             if(top_type){

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

  280.             }

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

  282.         } else {

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

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

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

  286.             if(top_type){

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

  288.             }

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

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

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

  292.         }

  293.         if(chroma){

  294.             if(chroma444){

  295.                 if(left_type){

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

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

  298.                 }

  299.                 if( IS_8x8DCT(mb_type) ) {

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

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

  302.                     if(top_type){

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

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

  305.                     }

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

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

  308.                 } else {

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

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

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

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

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

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

  315.                     if(top_type){

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

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

  318.                     }

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

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

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

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

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

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

  325.                 }

  326.             }else if(chroma422){

  327.                 if(left_type){

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

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

  330.                 }

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

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

  333.                 if(top_type){

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

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

  336.                 }

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

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

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

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

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

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

  343.             }else{

  344.                 if(left_type){

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

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

  347.                 }

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

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

  350.                 if(top_type){

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

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

  353.                 }

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

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

  356.             }

  357.         }

  358.         return;

  359.     } else {

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

  361.         int edges;

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

  363.             edges = 4;

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

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

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

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

  368.         } else {

  369.             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;

  370.             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;

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

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

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

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

  375.         }

  376.         if( IS_INTRA(left_type) )

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

  378.         if( IS_INTRA(top_type) )

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

  380. 

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

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

  383.             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 );\

  384.             if(chroma){\

  385.                 if(chroma444){\

  386.                     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 );\

  387.                     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 );\

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

  389.                     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 );\

  390.                     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 );\

  391.                 }\

  392.             }\

  393.         }

  394.         if(left_type)

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

  396.         if( edges == 1 ) {

  397.             if(top_type)

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

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

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

  401.             if(top_type)

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

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

  404.         } else {

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

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

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

  408.             if(top_type)

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

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

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

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

  413.         }

  414. #undef FILTER

  415.     }

  416. }

  417. 

  418. 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) {

  419.     assert(!FRAME_MBAFF(h));

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

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

  422.         return;

  423.     }

  424. 

  425. #if CONFIG_SMALL

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

  427. #else

  428.     if(h->pixel_shift){

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

  430.     }else{

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

  432.     }

  433. #endif

  434. }

  435. 

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

  437.     int v;

  438. 

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

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

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

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

  443. 

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

  445.         if(!v)

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

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

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

  449. 

  450.         if(v){

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

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

  453.                 return 1;

  454.             return

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

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

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

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

  459.         }

  460.     }

  461. 

  462.     return v;

  463. }

  464. 

  465. 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) {

  466.     int edge;

  467.     int chroma_qp_avg[2];

  468.     int chroma444 = CHROMA444(h);

  469.     int chroma422 = CHROMA422(h);

  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(h) && (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 * h->mb_stride;

  494.             int j;

  495. 

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

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

  498.                 int qp;

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

  500.                     AV_WN64A(bS, 0x0003000300030003ULL);

  501.                 } else {

  502.                     if (!CABAC(h) && IS_8x8DCT(h->cur_pic.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 = (h->cur_pic.qscale_table[mb_xy] + h->cur_pic.qscale_table[mbn_xy] + 1) >> 1;

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

  519.                 { int i; for (i = 0; i < 4; i++) tprintf(h->avctx, " bS[%d]:%d", i, bS[i]); tprintf(h->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, h->cur_pic.qscale_table[mbn_xy]) + 1) >> 1;

  522.                 chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp(h, 1, h->cur_pic.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(h) || (h->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(h) && 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 = (h->cur_pic.qscale_table[mb_xy] + h->cur_pic.qscale_table[mbm_xy] + 1) >> 1;

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

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

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

  586.                 if( dir == 0 ) {

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

  588.                     if (chroma) {

  589.                         if (chroma444) {

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

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

  592.                         } else {

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

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

  595.                         }

  596.                     }

  597.                 } else {

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

  599.                     if (chroma) {

  600.                         if (chroma444) {

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

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

  603.                         } else {

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

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

  606.                         }

  607.                     }

  608.                 }

  609.             }

  610.         }

  611.     }

  612. 

  613.     /* Calculate bS */

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

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

  616.         int qp;

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

  618. 

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

  620.             continue;

  621. 

  622.         if( IS_INTRA(mb_type)) {

  623.             AV_WN64A(bS, 0x0003000300030003ULL);

  624.         } else {

  625.             int i;

  626.             int mv_done;

  627. 

  628.             if( edge & mask_edge ) {

  629.                 AV_ZERO64(bS);

  630.                 mv_done = 1;

  631.             }

  632.             else if( mask_par0 ) {

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

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

  635. 

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

  637.                 mv_done = 1;

  638.             }

  639.             else

  640.                 mv_done = 0;

  641. 

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

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

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

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

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

  647. 

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

  649.                     h->non_zero_count_cache[bn_idx] ) {

  650.                     bS[i] = 2;

  651.                 }

  652.                 else if(!mv_done)

  653.                 {

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

  655.                 }

  656.             }

  657. 

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

  659.                 continue;

  660.         }

  661. 

  662.         /* Filter edge */

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

  664.         // value in IPCM macroblocks.

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

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

  667.         if( dir == 0 ) {

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

  669.             if (chroma) {

  670.                 if (chroma444) {

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

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

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

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

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

  676.                 }

  677.             }

  678.         } else {

  679.             if (chroma422) {

  680.                 if (deblock_edge)

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

  682.                 if (chroma) {

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

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

  685.                 }

  686.             } else {

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

  688.                 if (chroma) {

  689.                     if (chroma444) {

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

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

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

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

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

  695.                     }

  696.                 }

  697.             }

  698.         }

  699.     }

  700. }

  701. 

  702. 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) {

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

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

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

  706.     int first_vertical_edge_done = 0;

  707.     av_unused int dir;

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

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

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

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

  712. 

  713.     if (FRAME_MBAFF(h)

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

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

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

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

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

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

  720.          */

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

  722.         int qp[2];

  723.         int bqp[2];

  724.         int rqp[2];

  725.         int mb_qp, mbn0_qp, mbn1_qp;

  726.         int i;

  727.         first_vertical_edge_done = 1;

  728. 

  729.         if( IS_INTRA(mb_type) ) {

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

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

  732.         } else {

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

  734.                 {

  735.                     {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},

  736.                     {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},

  737.                 },{

  738.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  739.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  740.                 }

  741.             };

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

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

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

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

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

  747. 

  748.                 if( IS_INTRA( mbn_type ) )

  749.                     bS[i] = 4;

  750.                 else{

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

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

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

  754.                                                                        :

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

  756.                 }

  757.             }

  758.         }

  759. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  773. 

  774.         /* Filter edge */

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

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

  777.         if (MB_FIELD(h)) {

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

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

  780.             if (chroma){

  781.                 if (CHROMA444(h)) {

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

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

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

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

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

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

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

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

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

  791.                 }else{

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

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

  796.                 }

  797.             }

  798.         }else{

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

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

  801.             if (chroma){

  802.                 if (CHROMA444(h)) {

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

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

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

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

  807.                 }else{

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

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

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

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

  812.                 }

  813.             }

  814.         }

  815.     }

  816. 

  817. #if CONFIG_SMALL

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

  819.         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);

  820. #else

  821.     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);

  822.     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);

  823. #endif

  824. }