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.
33469 Commits
32 Branches
283MB
Tree: 4b4a6ebd68
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '4b4a6ebd68'
${ noResults }
FFmpeg/libavcodec/h264_loopfilter.c

798 lines
40KB
Raw Blame History 

  1. /*

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

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

  4.  *

  5.  * This file is part of FFmpeg.

  6.  *

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

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

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

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

  11.  *

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

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

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

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

  20.  */

  21. 

  22. /**

  23.  * @file

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

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

  26.  */

  27. 

  28. #include "libavutil/intreadwrite.h"

  29. #include "internal.h"

  30. #include "dsputil.h"

  31. #include "avcodec.h"

  32. #include "mpegvideo.h"

  33. #include "h264.h"

  34. #include "mathops.h"

  35. #include "rectangle.h"

  36. 

  37. //#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 void av_always_inline filter_mb_edgev( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, H264Context *h, int intra ) {

  105.     const unsigned int index_a = qp + a;

  106.     const int alpha = alpha_table[index_a];

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

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

  109. 

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

  111.         int8_t tc[4];

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

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

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

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

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

  117.     } else {

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

  119.     }

  120. }

  121. static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, H264Context *h, int intra ) {

  122.     const unsigned int index_a = qp + a;

  123.     const int alpha = alpha_table[index_a];

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

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

  126. 

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

  128.         int8_t tc[4];

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

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

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

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

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

  134.     } else {

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

  136.     }

  137. }

  138. 

  139. static void av_always_inline filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, const int16_t bS[7], int bsi, int qp, int a, int b, int intra ) {

  140.     const unsigned int index_a = qp + a;

  141.     const int alpha = alpha_table[index_a];

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

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

  144. 

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

  146.         int8_t tc[4];

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

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

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

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

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

  152.     } else {

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

  154.     }

  155. }

  156. static void av_always_inline filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, const int16_t bS[7], int bsi, int qp, int a, int b, int intra ) {

  157.     const unsigned int index_a = qp + a;

  158.     const int alpha = alpha_table[index_a];

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

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

  161. 

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

  163.         int8_t tc[4];

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

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

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

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

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

  169.     } else {

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

  171.     }

  172. }

  173. 

  174. static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, H264Context *h, int intra ) {

  175.     const unsigned int index_a = qp + a;

  176.     const int alpha = alpha_table[index_a];

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

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

  179. 

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

  181.         int8_t tc[4];

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

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

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

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

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

  187.     } else {

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

  189.     }

  190. }

  191. 

  192. static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, int a, int b, H264Context *h, int intra ) {

  193.     const unsigned int index_a = qp + a;

  194.     const int alpha = alpha_table[index_a];

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

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

  197. 

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

  199.         int8_t tc[4];

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

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

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

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

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

  205.     } else {

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

  207.     }

  208. }

  209. 

  210. static void av_always_inline h264_filter_mb_fast_internal( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,

  211.                                                            unsigned int linesize, unsigned int uvlinesize, int pixel_shift) {

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

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

  214.     int chroma444 = CHROMA444;

  215.     int chroma422 = CHROMA422;

  216. 

  217.     int mb_xy = h->mb_xy;

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

  219.     int top_type= h->top_type;

  220. 

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

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

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

  224. 

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

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

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

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

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

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

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

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

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

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

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

  236. 

  237.     if( IS_INTRA(mb_type) ) {

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

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

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

  241.         if(left_type)

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

  243.         if( IS_8x8DCT(mb_type) ) {

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

  245.             if(top_type){

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

  247.             }

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

  249.         } else {

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

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

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

  253.             if(top_type){

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

  255.             }

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

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

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

  259.         }

  260.         if(chroma){

  261.             if(chroma444){

  262.                 if(left_type){

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

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

  265.                 }

  266.                 if( IS_8x8DCT(mb_type) ) {

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

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

  269.                     if(top_type){

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

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

  272.                     }

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

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

  275.                 } else {

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

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

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

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

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

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

  282.                     if(top_type){

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

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

  285.                     }

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

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

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

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

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

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

  292.                 }

  293.             }else if(chroma422){

  294.                 if(left_type){

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

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

  297.                 }

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

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

  300.                 if(top_type){

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

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

  303.                 }

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

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

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

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

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

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

  310.             }else{

  311.                 if(left_type){

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

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

  314.                 }

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

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

  317.                 if(top_type){

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

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

  320.                 }

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

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

  323.             }

  324.         }

  325.         return;

  326.     } else {

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

  328.         int edges;

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

  330.             edges = 4;

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

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

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

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

  335.         } else {

  336.             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;

  337.             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;

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

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

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

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

  342.         }

  343.         if( IS_INTRA(left_type) )

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

  345.         if( IS_INTRA(top_type) )

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

  347. 

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

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

  350.             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 );\

  351.             if(chroma){\

  352.                 if(chroma444){\

  353.                     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 );\

  354.                     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 );\

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

  356.                     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 );\

  357.                     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 );\

  358.                 }\

  359.             }\

  360.         }

  361.         if(left_type)

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

  363.         if( edges == 1 ) {

  364.             if(top_type)

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

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

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

  368.             if(top_type)

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

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

  371.         } else {

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

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

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

  375.             if(top_type)

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

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

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

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

  380.         }

  381. #undef FILTER

  382.     }

  383. }

  384. 

  385. 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) {

  386.     assert(!FRAME_MBAFF);

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

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

  389.         return;

  390.     }

  391. 

  392. #if CONFIG_SMALL

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

  394. #else

  395.     if(h->pixel_shift){

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

  397.     }else{

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

  399.     }

  400. #endif

  401. }

  402. 

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

  404.     int v;

  405. 

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

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

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

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

  410. 

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

  412.         if(!v)

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

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

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

  416. 

  417.         if(v){

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

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

  420.                 return 1;

  421.             return

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

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

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

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

  426.         }

  427.     }

  428. 

  429.     return v;

  430. }

  431. 

  432. 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) {

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

  434.     int edge;

  435.     int chroma_qp_avg[2];

  436.     int chroma444 = CHROMA444;

  437.     int chroma422 = CHROMA422;

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

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

  440. 

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

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

  443.                                               {0,3,1,1,3,3,3,3}};

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

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

  446. 

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

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

  449. 

  450.     if(mbm_type && !first_vertical_edge_done){

  451. 

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

  453.             && IS_INTERLACED(mbm_type&~mb_type)

  454.             ) {

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

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

  457.             // frame macroblock.

  458.             //

  459.             unsigned int tmp_linesize   = 2 *   linesize;

  460.             unsigned int tmp_uvlinesize = 2 * uvlinesize;

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

  462.             int j;

  463. 

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

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

  466.                 int qp;

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

  468.                     AV_WN64A(bS, 0x0003000300030003ULL);

  469.                 } else {

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

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

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

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

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

  475.                     }else{

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

  477.                     int i;

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

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

  480.                     }

  481.                     }

  482.                 }

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

  484.                 // value in IPCM macroblocks.

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

  486.                 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);

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

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

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

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

  491.                 if (chroma) {

  492.                     if (chroma444) {

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

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

  495.                     } else {

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

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

  498.                     }

  499.                 }

  500.             }

  501.         }else{

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

  503.             int qp;

  504. 

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

  506.                 AV_WN64A(bS, 0x0003000300030003ULL);

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

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

  509.                 )

  510.                     AV_WN64A(bS, 0x0004000400040004ULL);

  511.             } else {

  512.                 int i;

  513.                 int mv_done;

  514. 

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

  516.                     AV_WN64A(bS, 0x0001000100010001ULL);

  517.                     mv_done = 1;

  518.                 }

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

  520.                     int b_idx= 8 + 4;

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

  522. 

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

  524.                     mv_done = 1;

  525.                 }

  526.                 else

  527.                     mv_done = 0;

  528. 

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

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

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

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

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

  534. 

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

  536.                         h->non_zero_count_cache[bn_idx] ) {

  537.                         bS[i] = 2;

  538.                     }

  539.                     else if(!mv_done)

  540.                     {

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

  542.                     }

  543.                 }

  544.             }

  545. 

  546.             /* Filter edge */

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

  548.             // value in IPCM macroblocks.

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

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

  551.                 //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]);

  552.                 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);

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

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

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

  556.                 if( dir == 0 ) {

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

  558.                     if (chroma) {

  559.                         if (chroma444) {

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

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

  562.                         } else {

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

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

  565.                         }

  566.                     }

  567.                 } else {

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

  569.                     if (chroma) {

  570.                         if (chroma444) {

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

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

  573.                         } else {

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

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

  576.                         }

  577.                     }

  578.                 }

  579.             }

  580.         }

  581.     }

  582. 

  583.     /* Calculate bS */

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

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

  586.         int qp;

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

  588. 

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

  590.             continue;

  591. 

  592.         if( IS_INTRA(mb_type)) {

  593.             AV_WN64A(bS, 0x0003000300030003ULL);

  594.         } else {

  595.             int i;

  596.             int mv_done;

  597. 

  598.             if( edge & mask_edge ) {

  599.                 AV_ZERO64(bS);

  600.                 mv_done = 1;

  601.             }

  602.             else if( mask_par0 ) {

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

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

  605. 

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

  607.                 mv_done = 1;

  608.             }

  609.             else

  610.                 mv_done = 0;

  611. 

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

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

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

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

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

  617. 

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

  619.                     h->non_zero_count_cache[bn_idx] ) {

  620.                     bS[i] = 2;

  621.                 }

  622.                 else if(!mv_done)

  623.                 {

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

  625.                 }

  626.             }

  627. 

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

  629.                 continue;

  630.         }

  631. 

  632.         /* Filter edge */

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

  634.         // value in IPCM macroblocks.

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

  636.         //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]);

  637.         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);

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

  639.         if( dir == 0 ) {

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

  641.             if (chroma) {

  642.                 if (chroma444) {

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

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

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

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

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

  648.                 }

  649.             }

  650.         } else {

  651.             if (chroma422) {

  652.                 if (deblock_edge)

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

  654.                 if (chroma) {

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

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

  657.                 }

  658.             } else {

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

  660.                 if (chroma) {

  661.                     if (chroma444) {

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

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

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

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

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

  667.                     }

  668.                 }

  669.             }

  670.         }

  671.     }

  672. }

  673. 

  674. 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) {

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

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

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

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

  679.     int first_vertical_edge_done = 0;

  680.     av_unused int dir;

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

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

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

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

  685. 

  686.     if (FRAME_MBAFF

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

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

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

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

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

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

  693.          */

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

  695.         int qp[2];

  696.         int bqp[2];

  697.         int rqp[2];

  698.         int mb_qp, mbn0_qp, mbn1_qp;

  699.         int i;

  700.         first_vertical_edge_done = 1;

  701. 

  702.         if( IS_INTRA(mb_type) ) {

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

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

  705.         } else {

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

  707.                 {

  708.                     {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},

  709.                     {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},

  710.                 },{

  711.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  712.                     {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},

  713.                 }

  714.             };

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

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

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

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

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

  720. 

  721.                 if( IS_INTRA( mbn_type ) )

  722.                     bS[i] = 4;

  723.                 else{

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

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

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

  727.                                                                        :

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

  729.                 }

  730.             }

  731.         }

  732. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  746. 

  747.         /* Filter edge */

  748.         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);

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

  750.         if(MB_FIELD){

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

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

  753.             if (chroma){

  754.                 if (CHROMA444) {

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

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

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

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

  759.                 } else if (CHROMA422) {

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

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

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

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

  764.                 }else{

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

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

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

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

  769.                 }

  770.             }

  771.         }else{

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

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

  774.             if (chroma){

  775.                 if (CHROMA444) {

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

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

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

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

  780.                 }else{

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

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

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

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

  785.                 }

  786.             }

  787.         }

  788.     }

  789. 

  790. #if CONFIG_SMALL

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

  792.         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);

  793. #else

  794.     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);

  795.     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);

  796. #endif

  797. }