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.

751 lines
36KB

  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. * @file
  23. * H.264 / AVC / MPEG4 part10 loop filter.
  24. * @author Michael Niedermayer <michaelni@gmx.at>
  25. */
  26. #include "libavutil/intreadwrite.h"
  27. #include "internal.h"
  28. #include "dsputil.h"
  29. #include "avcodec.h"
  30. #include "mpegvideo.h"
  31. #include "h264.h"
  32. #include "mathops.h"
  33. #include "rectangle.h"
  34. //#undef NDEBUG
  35. #include <assert.h>
  36. /* Deblocking filter (p153) */
  37. static const uint8_t alpha_table[52*3] = {
  38. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  39. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  40. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  41. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  42. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  43. 0, 0, 0, 0, 0, 0, 4, 4, 5, 6,
  44. 7, 8, 9, 10, 12, 13, 15, 17, 20, 22,
  45. 25, 28, 32, 36, 40, 45, 50, 56, 63, 71,
  46. 80, 90,101,113,127,144,162,182,203,226,
  47. 255,255,
  48. 255,255,255,255,255,255,255,255,255,255,255,255,255,
  49. 255,255,255,255,255,255,255,255,255,255,255,255,255,
  50. 255,255,255,255,255,255,255,255,255,255,255,255,255,
  51. 255,255,255,255,255,255,255,255,255,255,255,255,255,
  52. };
  53. static const uint8_t beta_table[52*3] = {
  54. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  55. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  56. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  57. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  58. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  59. 0, 0, 0, 0, 0, 0, 2, 2, 2, 3,
  60. 3, 3, 3, 4, 4, 4, 6, 6, 7, 7,
  61. 8, 8, 9, 9, 10, 10, 11, 11, 12, 12,
  62. 13, 13, 14, 14, 15, 15, 16, 16, 17, 17,
  63. 18, 18,
  64. 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
  65. 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
  66. 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
  67. 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18,
  68. };
  69. static const uint8_t tc0_table[52*3][4] = {
  70. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  71. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  72. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  73. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  74. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  75. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  76. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  77. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  78. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  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, 1 },
  82. {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },
  83. {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },
  84. {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },
  85. {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },
  86. {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },
  87. {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-1,13,17,25 },
  88. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  89. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  90. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  91. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  92. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  93. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  94. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  95. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  96. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 },
  97. };
  98. static void av_always_inline filter_mb_edgev( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, H264Context *h) {
  99. const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
  100. const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
  101. const int alpha = alpha_table[index_a];
  102. const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
  103. if (alpha ==0 || beta == 0) return;
  104. if( bS[0] < 4 ) {
  105. int8_t tc[4];
  106. tc[0] = tc0_table[index_a][bS[0]];
  107. tc[1] = tc0_table[index_a][bS[1]];
  108. tc[2] = tc0_table[index_a][bS[2]];
  109. tc[3] = tc0_table[index_a][bS[3]];
  110. h->h264dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
  111. } else {
  112. h->h264dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
  113. }
  114. }
  115. static void av_always_inline filter_mb_edgecv( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, H264Context *h ) {
  116. const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
  117. const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
  118. const int alpha = alpha_table[index_a];
  119. const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
  120. if (alpha ==0 || beta == 0) return;
  121. if( bS[0] < 4 ) {
  122. int8_t tc[4];
  123. tc[0] = tc0_table[index_a][bS[0]]+1;
  124. tc[1] = tc0_table[index_a][bS[1]]+1;
  125. tc[2] = tc0_table[index_a][bS[2]]+1;
  126. tc[3] = tc0_table[index_a][bS[3]]+1;
  127. h->h264dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
  128. } else {
  129. h->h264dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
  130. }
  131. }
  132. static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, const int16_t bS[7], int bsi, int qp ) {
  133. const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
  134. int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
  135. int alpha = alpha_table[index_a];
  136. int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
  137. if (alpha ==0 || beta == 0) return;
  138. if( bS[0] < 4 ) {
  139. int8_t tc[4];
  140. tc[0] = tc0_table[index_a][bS[0*bsi]];
  141. tc[1] = tc0_table[index_a][bS[1*bsi]];
  142. tc[2] = tc0_table[index_a][bS[2*bsi]];
  143. tc[3] = tc0_table[index_a][bS[3*bsi]];
  144. h->h264dsp.h264_h_loop_filter_luma_mbaff(pix, stride, alpha, beta, tc);
  145. } else {
  146. h->h264dsp.h264_h_loop_filter_luma_mbaff_intra(pix, stride, alpha, beta);
  147. }
  148. }
  149. static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, const int16_t bS[7], int bsi, int qp ) {
  150. const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
  151. int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
  152. int alpha = alpha_table[index_a];
  153. int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
  154. if (alpha ==0 || beta == 0) return;
  155. if( bS[0] < 4 ) {
  156. int8_t tc[4];
  157. tc[0] = tc0_table[index_a][bS[0*bsi]] + 1;
  158. tc[1] = tc0_table[index_a][bS[1*bsi]] + 1;
  159. tc[2] = tc0_table[index_a][bS[2*bsi]] + 1;
  160. tc[3] = tc0_table[index_a][bS[3*bsi]] + 1;
  161. h->h264dsp.h264_h_loop_filter_chroma_mbaff(pix, stride, alpha, beta, tc);
  162. } else {
  163. h->h264dsp.h264_h_loop_filter_chroma_mbaff_intra(pix, stride, alpha, beta);
  164. }
  165. }
  166. static void av_always_inline filter_mb_edgeh( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, H264Context *h ) {
  167. const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
  168. const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
  169. const int alpha = alpha_table[index_a];
  170. const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
  171. if (alpha ==0 || beta == 0) return;
  172. if( bS[0] < 4 ) {
  173. int8_t tc[4];
  174. tc[0] = tc0_table[index_a][bS[0]];
  175. tc[1] = tc0_table[index_a][bS[1]];
  176. tc[2] = tc0_table[index_a][bS[2]];
  177. tc[3] = tc0_table[index_a][bS[3]];
  178. h->h264dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
  179. } else {
  180. h->h264dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
  181. }
  182. }
  183. static void av_always_inline filter_mb_edgech( uint8_t *pix, int stride, const int16_t bS[4], unsigned int qp, H264Context *h ) {
  184. const int qp_bd_offset = 6 * (h->sps.bit_depth_luma - 8);
  185. const unsigned int index_a = qp - qp_bd_offset + h->slice_alpha_c0_offset;
  186. const int alpha = alpha_table[index_a];
  187. const int beta = beta_table[qp - qp_bd_offset + h->slice_beta_offset];
  188. if (alpha ==0 || beta == 0) return;
  189. if( bS[0] < 4 ) {
  190. int8_t tc[4];
  191. tc[0] = tc0_table[index_a][bS[0]]+1;
  192. tc[1] = tc0_table[index_a][bS[1]]+1;
  193. tc[2] = tc0_table[index_a][bS[2]]+1;
  194. tc[3] = tc0_table[index_a][bS[3]]+1;
  195. h->h264dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
  196. } else {
  197. h->h264dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
  198. }
  199. }
  200. 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) {
  201. MpegEncContext * const s = &h->s;
  202. int mb_xy;
  203. int mb_type, left_type, top_type;
  204. int qp, qp0, qp1, qpc, qpc0, qpc1;
  205. int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
  206. int chroma444 = CHROMA444;
  207. mb_xy = h->mb_xy;
  208. if(!h->h264dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff) {
  209. ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
  210. return;
  211. }
  212. assert(!FRAME_MBAFF);
  213. left_type= h->left_type[LTOP];
  214. top_type= h->top_type;
  215. mb_type = s->current_picture.mb_type[mb_xy];
  216. qp = s->current_picture.qscale_table[mb_xy];
  217. qp0 = s->current_picture.qscale_table[mb_xy-1];
  218. qp1 = s->current_picture.qscale_table[h->top_mb_xy];
  219. qpc = get_chroma_qp( h, 0, qp );
  220. qpc0 = get_chroma_qp( h, 0, qp0 );
  221. qpc1 = get_chroma_qp( h, 0, qp1 );
  222. qp0 = (qp + qp0 + 1) >> 1;
  223. qp1 = (qp + qp1 + 1) >> 1;
  224. qpc0 = (qpc + qpc0 + 1) >> 1;
  225. qpc1 = (qpc + qpc1 + 1) >> 1;
  226. if( IS_INTRA(mb_type) ) {
  227. static const int16_t bS4[4] = {4,4,4,4};
  228. static const int16_t bS3[4] = {3,3,3,3};
  229. const int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
  230. if(left_type)
  231. filter_mb_edgev( &img_y[4*0], linesize, bS4, qp0, h);
  232. if( IS_8x8DCT(mb_type) ) {
  233. filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
  234. if(top_type){
  235. filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
  236. }
  237. filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
  238. } else {
  239. filter_mb_edgev( &img_y[4*1], linesize, bS3, qp, h);
  240. filter_mb_edgev( &img_y[4*2], linesize, bS3, qp, h);
  241. filter_mb_edgev( &img_y[4*3], linesize, bS3, qp, h);
  242. if(top_type){
  243. filter_mb_edgeh( &img_y[4*0*linesize], linesize, bSH, qp1, h);
  244. }
  245. filter_mb_edgeh( &img_y[4*1*linesize], linesize, bS3, qp, h);
  246. filter_mb_edgeh( &img_y[4*2*linesize], linesize, bS3, qp, h);
  247. filter_mb_edgeh( &img_y[4*3*linesize], linesize, bS3, qp, h);
  248. }
  249. if(chroma){
  250. if(chroma444){
  251. if(left_type){
  252. filter_mb_edgev( &img_cb[4*0], linesize, bS4, qpc0, h);
  253. filter_mb_edgev( &img_cr[4*0], linesize, bS4, qpc0, h);
  254. }
  255. if( IS_8x8DCT(mb_type) ) {
  256. filter_mb_edgev( &img_cb[4*2], linesize, bS3, qpc, h);
  257. filter_mb_edgev( &img_cr[4*2], linesize, bS3, qpc, h);
  258. if(top_type){
  259. filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, h);
  260. filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, h);
  261. }
  262. filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, h);
  263. filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, h);
  264. } else {
  265. filter_mb_edgev( &img_cb[4*1], linesize, bS3, qpc, h);
  266. filter_mb_edgev( &img_cr[4*1], linesize, bS3, qpc, h);
  267. filter_mb_edgev( &img_cb[4*2], linesize, bS3, qpc, h);
  268. filter_mb_edgev( &img_cr[4*2], linesize, bS3, qpc, h);
  269. filter_mb_edgev( &img_cb[4*3], linesize, bS3, qpc, h);
  270. filter_mb_edgev( &img_cr[4*3], linesize, bS3, qpc, h);
  271. if(top_type){
  272. filter_mb_edgeh( &img_cb[4*0*linesize], linesize, bSH, qpc1, h);
  273. filter_mb_edgeh( &img_cr[4*0*linesize], linesize, bSH, qpc1, h);
  274. }
  275. filter_mb_edgeh( &img_cb[4*1*linesize], linesize, bS3, qpc, h);
  276. filter_mb_edgeh( &img_cr[4*1*linesize], linesize, bS3, qpc, h);
  277. filter_mb_edgeh( &img_cb[4*2*linesize], linesize, bS3, qpc, h);
  278. filter_mb_edgeh( &img_cr[4*2*linesize], linesize, bS3, qpc, h);
  279. filter_mb_edgeh( &img_cb[4*3*linesize], linesize, bS3, qpc, h);
  280. filter_mb_edgeh( &img_cr[4*3*linesize], linesize, bS3, qpc, h);
  281. }
  282. }else{
  283. if(left_type){
  284. filter_mb_edgecv( &img_cb[2*0], uvlinesize, bS4, qpc0, h);
  285. filter_mb_edgecv( &img_cr[2*0], uvlinesize, bS4, qpc0, h);
  286. }
  287. filter_mb_edgecv( &img_cb[2*2], uvlinesize, bS3, qpc, h);
  288. filter_mb_edgecv( &img_cr[2*2], uvlinesize, bS3, qpc, h);
  289. if(top_type){
  290. filter_mb_edgech( &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
  291. filter_mb_edgech( &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1, h);
  292. }
  293. filter_mb_edgech( &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
  294. filter_mb_edgech( &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc, h);
  295. }
  296. }
  297. return;
  298. } else {
  299. LOCAL_ALIGNED_8(int16_t, bS, [2], [4][4]);
  300. int edges;
  301. if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 && !chroma444 ) {
  302. edges = 4;
  303. AV_WN64A(bS[0][0], 0x0002000200020002ULL);
  304. AV_WN64A(bS[0][2], 0x0002000200020002ULL);
  305. AV_WN64A(bS[1][0], 0x0002000200020002ULL);
  306. AV_WN64A(bS[1][2], 0x0002000200020002ULL);
  307. } else {
  308. 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;
  309. 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;
  310. int step = 1+(mb_type>>24); //IS_8x8DCT(mb_type) ? 2 : 1;
  311. edges = 4 - 3*((mb_type>>3) & !(h->cbp & 15)); //(mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
  312. h->h264dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
  313. h->list_count==2, edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
  314. }
  315. if( IS_INTRA(left_type) )
  316. AV_WN64A(bS[0][0], 0x0004000400040004ULL);
  317. if( IS_INTRA(top_type) )
  318. AV_WN64A(bS[1][0], FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL);
  319. #define FILTER(hv,dir,edge)\
  320. if(AV_RN64A(bS[dir][edge])) { \
  321. filter_mb_edge##hv( &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir, h );\
  322. if(chroma){\
  323. if(chroma444){\
  324. filter_mb_edge##hv( &img_cb[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
  325. filter_mb_edge##hv( &img_cr[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
  326. } else if(!(edge&1)) {\
  327. filter_mb_edgec##hv( &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
  328. filter_mb_edgec##hv( &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir, h );\
  329. }\
  330. }\
  331. }
  332. if(left_type)
  333. FILTER(v,0,0);
  334. if( edges == 1 ) {
  335. if(top_type)
  336. FILTER(h,1,0);
  337. } else if( IS_8x8DCT(mb_type) ) {
  338. FILTER(v,0,2);
  339. if(top_type)
  340. FILTER(h,1,0);
  341. FILTER(h,1,2);
  342. } else {
  343. FILTER(v,0,1);
  344. FILTER(v,0,2);
  345. FILTER(v,0,3);
  346. if(top_type)
  347. FILTER(h,1,0);
  348. FILTER(h,1,1);
  349. FILTER(h,1,2);
  350. FILTER(h,1,3);
  351. }
  352. #undef FILTER
  353. }
  354. }
  355. static int check_mv(H264Context *h, long b_idx, long bn_idx, int mvy_limit){
  356. int v;
  357. v= h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx];
  358. if(!v && h->ref_cache[0][b_idx]!=-1)
  359. v= h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
  360. FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
  361. if(h->list_count==2){
  362. if(!v)
  363. v = h->ref_cache[1][b_idx] != h->ref_cache[1][bn_idx] |
  364. h->mv_cache[1][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
  365. FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit;
  366. if(v){
  367. if(h->ref_cache[0][b_idx] != h->ref_cache[1][bn_idx] |
  368. h->ref_cache[1][b_idx] != h->ref_cache[0][bn_idx])
  369. return 1;
  370. return
  371. h->mv_cache[0][b_idx][0] - h->mv_cache[1][bn_idx][0] + 3 >= 7U |
  372. FFABS( h->mv_cache[0][b_idx][1] - h->mv_cache[1][bn_idx][1] ) >= mvy_limit |
  373. h->mv_cache[1][b_idx][0] - h->mv_cache[0][bn_idx][0] + 3 >= 7U |
  374. FFABS( h->mv_cache[1][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= mvy_limit;
  375. }
  376. }
  377. return v;
  378. }
  379. 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 chroma, int chroma444, int dir) {
  380. MpegEncContext * const s = &h->s;
  381. int edge;
  382. int chroma_qp_avg[2];
  383. const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
  384. const int mbm_type = dir == 0 ? h->left_type[LTOP] : h->top_type;
  385. // how often to recheck mv-based bS when iterating between edges
  386. static const uint8_t mask_edge_tab[2][8]={{0,3,3,3,1,1,1,1},
  387. {0,3,1,1,3,3,3,3}};
  388. const int mask_edge = mask_edge_tab[dir][(mb_type>>3)&7];
  389. const int edges = mask_edge== 3 && !(h->cbp&15) ? 1 : 4;
  390. // how often to recheck mv-based bS when iterating along each edge
  391. const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
  392. if(mbm_type && !first_vertical_edge_done){
  393. if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0)
  394. && IS_INTERLACED(mbm_type&~mb_type)
  395. ) {
  396. // This is a special case in the norm where the filtering must
  397. // be done twice (one each of the field) even if we are in a
  398. // frame macroblock.
  399. //
  400. unsigned int tmp_linesize = 2 * linesize;
  401. unsigned int tmp_uvlinesize = 2 * uvlinesize;
  402. int mbn_xy = mb_xy - 2 * s->mb_stride;
  403. int j;
  404. for(j=0; j<2; j++, mbn_xy += s->mb_stride){
  405. DECLARE_ALIGNED(8, int16_t, bS)[4];
  406. int qp;
  407. if( IS_INTRA(mb_type|s->current_picture.mb_type[mbn_xy]) ) {
  408. AV_WN64A(bS, 0x0003000300030003ULL);
  409. } else {
  410. if(!CABAC && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])){
  411. bS[0]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+0]);
  412. bS[1]= 1+((h->cbp_table[mbn_xy] & 0x4000)||h->non_zero_count_cache[scan8[0]+1]);
  413. bS[2]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+2]);
  414. bS[3]= 1+((h->cbp_table[mbn_xy] & 0x8000)||h->non_zero_count_cache[scan8[0]+3]);
  415. }else{
  416. const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy] + 3*4;
  417. int i;
  418. for( i = 0; i < 4; i++ ) {
  419. bS[i] = 1 + !!(h->non_zero_count_cache[scan8[0]+i] | mbn_nnz[i]);
  420. }
  421. }
  422. }
  423. // Do not use s->qscale as luma quantizer because it has not the same
  424. // value in IPCM macroblocks.
  425. qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
  426. 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);
  427. { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  428. filter_mb_edgeh( &img_y[j*linesize], tmp_linesize, bS, qp, h );
  429. chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
  430. chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
  431. if (chroma) {
  432. if (chroma444) {
  433. filter_mb_edgeh (&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h);
  434. filter_mb_edgeh (&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h);
  435. } else {
  436. filter_mb_edgech(&img_cb[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[0], h);
  437. filter_mb_edgech(&img_cr[j*uvlinesize], tmp_uvlinesize, bS, chroma_qp_avg[1], h);
  438. }
  439. }
  440. }
  441. }else{
  442. DECLARE_ALIGNED(8, int16_t, bS)[4];
  443. int qp;
  444. if( IS_INTRA(mb_type|mbm_type)) {
  445. AV_WN64A(bS, 0x0003000300030003ULL);
  446. if ( (!IS_INTERLACED(mb_type|mbm_type))
  447. || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
  448. )
  449. AV_WN64A(bS, 0x0004000400040004ULL);
  450. } else {
  451. int i;
  452. int mv_done;
  453. if( dir && FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbm_type)) {
  454. AV_WN64A(bS, 0x0001000100010001ULL);
  455. mv_done = 1;
  456. }
  457. else if( mask_par0 && ((mbm_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
  458. int b_idx= 8 + 4;
  459. int bn_idx= b_idx - (dir ? 8:1);
  460. bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, 8 + 4, bn_idx, mvy_limit);
  461. mv_done = 1;
  462. }
  463. else
  464. mv_done = 0;
  465. for( i = 0; i < 4; i++ ) {
  466. int x = dir == 0 ? 0 : i;
  467. int y = dir == 0 ? i : 0;
  468. int b_idx= 8 + 4 + x + 8*y;
  469. int bn_idx= b_idx - (dir ? 8:1);
  470. if( h->non_zero_count_cache[b_idx] |
  471. h->non_zero_count_cache[bn_idx] ) {
  472. bS[i] = 2;
  473. }
  474. else if(!mv_done)
  475. {
  476. bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
  477. }
  478. }
  479. }
  480. /* Filter edge */
  481. // Do not use s->qscale as luma quantizer because it has not the same
  482. // value in IPCM macroblocks.
  483. if(bS[0]+bS[1]+bS[2]+bS[3]){
  484. qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbm_xy] + 1 ) >> 1;
  485. //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]);
  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, linesize, uvlinesize);
  487. //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  488. chroma_qp_avg[0] = (h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
  489. chroma_qp_avg[1] = (h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbm_xy] ) + 1 ) >> 1;
  490. if( dir == 0 ) {
  491. filter_mb_edgev( &img_y[0], linesize, bS, qp, h );
  492. if (chroma) {
  493. if (chroma444) {
  494. filter_mb_edgev ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
  495. filter_mb_edgev ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
  496. } else {
  497. filter_mb_edgecv( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
  498. filter_mb_edgecv( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
  499. }
  500. }
  501. } else {
  502. filter_mb_edgeh( &img_y[0], linesize, bS, qp, h );
  503. if (chroma) {
  504. if (chroma444) {
  505. filter_mb_edgeh ( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
  506. filter_mb_edgeh ( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
  507. } else {
  508. filter_mb_edgech( &img_cb[0], uvlinesize, bS, chroma_qp_avg[0], h);
  509. filter_mb_edgech( &img_cr[0], uvlinesize, bS, chroma_qp_avg[1], h);
  510. }
  511. }
  512. }
  513. }
  514. }
  515. }
  516. /* Calculate bS */
  517. for( edge = 1; edge < edges; edge++ ) {
  518. DECLARE_ALIGNED(8, int16_t, bS)[4];
  519. int qp;
  520. if( IS_8x8DCT(mb_type & (edge<<24)) ) // (edge&1) && IS_8x8DCT(mb_type)
  521. continue;
  522. if( IS_INTRA(mb_type)) {
  523. AV_WN64A(bS, 0x0003000300030003ULL);
  524. } else {
  525. int i;
  526. int mv_done;
  527. if( edge & mask_edge ) {
  528. AV_ZERO64(bS);
  529. mv_done = 1;
  530. }
  531. else if( mask_par0 ) {
  532. int b_idx= 8 + 4 + edge * (dir ? 8:1);
  533. int bn_idx= b_idx - (dir ? 8:1);
  534. bS[0] = bS[1] = bS[2] = bS[3] = check_mv(h, b_idx, bn_idx, mvy_limit);
  535. mv_done = 1;
  536. }
  537. else
  538. mv_done = 0;
  539. for( i = 0; i < 4; i++ ) {
  540. int x = dir == 0 ? edge : i;
  541. int y = dir == 0 ? i : edge;
  542. int b_idx= 8 + 4 + x + 8*y;
  543. int bn_idx= b_idx - (dir ? 8:1);
  544. if( h->non_zero_count_cache[b_idx] |
  545. h->non_zero_count_cache[bn_idx] ) {
  546. bS[i] = 2;
  547. }
  548. else if(!mv_done)
  549. {
  550. bS[i] = check_mv(h, b_idx, bn_idx, mvy_limit);
  551. }
  552. }
  553. if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
  554. continue;
  555. }
  556. /* Filter edge */
  557. // Do not use s->qscale as luma quantizer because it has not the same
  558. // value in IPCM macroblocks.
  559. qp = s->current_picture.qscale_table[mb_xy];
  560. //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]);
  561. 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);
  562. //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  563. if( dir == 0 ) {
  564. filter_mb_edgev( &img_y[4*edge << h->pixel_shift], linesize, bS, qp, h );
  565. if (chroma) {
  566. if (chroma444) {
  567. filter_mb_edgev ( &img_cb[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
  568. filter_mb_edgev ( &img_cr[4*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
  569. } else if( (edge&1) == 0 ) {
  570. filter_mb_edgecv( &img_cb[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[0], h);
  571. filter_mb_edgecv( &img_cr[2*edge << h->pixel_shift], uvlinesize, bS, h->chroma_qp[1], h);
  572. }
  573. }
  574. } else {
  575. filter_mb_edgeh( &img_y[4*edge*linesize], linesize, bS, qp, h );
  576. if (chroma) {
  577. if (chroma444) {
  578. filter_mb_edgeh ( &img_cb[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
  579. filter_mb_edgeh ( &img_cr[4*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
  580. } else if( (edge&1) == 0 ) {
  581. filter_mb_edgech( &img_cb[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[0], h);
  582. filter_mb_edgech( &img_cr[2*edge*uvlinesize], uvlinesize, bS, h->chroma_qp[1], h);
  583. }
  584. }
  585. }
  586. }
  587. }
  588. 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) {
  589. MpegEncContext * const s = &h->s;
  590. const int mb_xy= mb_x + mb_y*s->mb_stride;
  591. const int mb_type = s->current_picture.mb_type[mb_xy];
  592. const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
  593. int first_vertical_edge_done = 0;
  594. av_unused int dir;
  595. int chroma = !(CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
  596. if (FRAME_MBAFF
  597. // and current and left pair do not have the same interlaced type
  598. && IS_INTERLACED(mb_type^h->left_type[LTOP])
  599. // and left mb is in available to us
  600. && h->left_type[LTOP]) {
  601. /* First vertical edge is different in MBAFF frames
  602. * There are 8 different bS to compute and 2 different Qp
  603. */
  604. DECLARE_ALIGNED(8, int16_t, bS)[8];
  605. int qp[2];
  606. int bqp[2];
  607. int rqp[2];
  608. int mb_qp, mbn0_qp, mbn1_qp;
  609. int i;
  610. first_vertical_edge_done = 1;
  611. if( IS_INTRA(mb_type) ) {
  612. AV_WN64A(&bS[0], 0x0004000400040004ULL);
  613. AV_WN64A(&bS[4], 0x0004000400040004ULL);
  614. } else {
  615. static const uint8_t offset[2][2][8]={
  616. {
  617. {3+4*0, 3+4*0, 3+4*0, 3+4*0, 3+4*1, 3+4*1, 3+4*1, 3+4*1},
  618. {3+4*2, 3+4*2, 3+4*2, 3+4*2, 3+4*3, 3+4*3, 3+4*3, 3+4*3},
  619. },{
  620. {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
  621. {3+4*0, 3+4*1, 3+4*2, 3+4*3, 3+4*0, 3+4*1, 3+4*2, 3+4*3},
  622. }
  623. };
  624. const uint8_t *off= offset[MB_FIELD][mb_y&1];
  625. for( i = 0; i < 8; i++ ) {
  626. int j= MB_FIELD ? i>>2 : i&1;
  627. int mbn_xy = h->left_mb_xy[LEFT(j)];
  628. int mbn_type= h->left_type[LEFT(j)];
  629. if( IS_INTRA( mbn_type ) )
  630. bS[i] = 4;
  631. else{
  632. bS[i] = 1 + !!(h->non_zero_count_cache[12+8*(i>>1)] |
  633. ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
  634. (h->cbp_table[mbn_xy] & (((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2) << 12))
  635. :
  636. h->non_zero_count[mbn_xy][ off[i] ]));
  637. }
  638. }
  639. }
  640. mb_qp = s->current_picture.qscale_table[mb_xy];
  641. mbn0_qp = s->current_picture.qscale_table[h->left_mb_xy[0]];
  642. mbn1_qp = s->current_picture.qscale_table[h->left_mb_xy[1]];
  643. qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
  644. bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
  645. get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
  646. rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
  647. get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
  648. qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
  649. bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
  650. get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
  651. rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
  652. get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
  653. /* Filter edge */
  654. 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);
  655. { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  656. if(MB_FIELD){
  657. filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
  658. filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
  659. if (chroma){
  660. if (CHROMA444) {
  661. filter_mb_mbaff_edgev ( h, img_cb, uvlinesize, bS , 1, bqp[0] );
  662. filter_mb_mbaff_edgev ( h, img_cb + 8*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
  663. filter_mb_mbaff_edgev ( h, img_cr, uvlinesize, bS , 1, rqp[0] );
  664. filter_mb_mbaff_edgev ( h, img_cr + 8*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
  665. }else{
  666. filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
  667. filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
  668. filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
  669. filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
  670. }
  671. }
  672. }else{
  673. filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
  674. filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
  675. if (chroma){
  676. if (CHROMA444) {
  677. filter_mb_mbaff_edgev ( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
  678. filter_mb_mbaff_edgev ( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
  679. filter_mb_mbaff_edgev ( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
  680. filter_mb_mbaff_edgev ( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
  681. }else{
  682. filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
  683. filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
  684. filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
  685. filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
  686. }
  687. }
  688. }
  689. }
  690. #if CONFIG_SMALL
  691. for( dir = 0; dir < 2; dir++ )
  692. 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, chroma, CHROMA444, dir);
  693. #else
  694. 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, chroma, CHROMA444, 0);
  695. filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, chroma, CHROMA444, 1);
  696. #endif
  697. }