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.

720 lines
33KB

  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 libavcodec/h264_loopfilter.c
  23. * H.264 / AVC / MPEG4 part10 loop filter.
  24. * @author Michael Niedermayer <michaelni@gmx.at>
  25. */
  26. #include "internal.h"
  27. #include "dsputil.h"
  28. #include "avcodec.h"
  29. #include "mpegvideo.h"
  30. #include "h264.h"
  31. #include "mathops.h"
  32. #include "rectangle.h"
  33. //#undef NDEBUG
  34. #include <assert.h>
  35. /* Deblocking filter (p153) */
  36. static const uint8_t alpha_table[52*3] = {
  37. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  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,
  42. 0, 0, 0, 0, 0, 0, 4, 4, 5, 6,
  43. 7, 8, 9, 10, 12, 13, 15, 17, 20, 22,
  44. 25, 28, 32, 36, 40, 45, 50, 56, 63, 71,
  45. 80, 90,101,113,127,144,162,182,203,226,
  46. 255,255,
  47. 255,255,255,255,255,255,255,255,255,255,255,255,255,
  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. };
  52. static const uint8_t beta_table[52*3] = {
  53. 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
  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,
  58. 0, 0, 0, 0, 0, 0, 2, 2, 2, 3,
  59. 3, 3, 3, 4, 4, 4, 6, 6, 7, 7,
  60. 8, 8, 9, 9, 10, 10, 11, 11, 12, 12,
  61. 13, 13, 14, 14, 15, 15, 16, 16, 17, 17,
  62. 18, 18,
  63. 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 18, 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. };
  68. static const uint8_t tc0_table[52*3][4] = {
  69. {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 }, {-1, 0, 0, 0 },
  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 },
  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, 1 },
  81. {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },
  82. {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },
  83. {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },
  84. {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },
  85. {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },
  86. {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-1,13,17,25 },
  87. {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-1,13,17,25 }, {-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 },
  96. };
  97. static void av_noinline filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
  98. const int index_a = qp + h->slice_alpha_c0_offset;
  99. const int alpha = (alpha_table+52)[index_a];
  100. const int beta = (beta_table+52)[qp + h->slice_beta_offset];
  101. if (alpha ==0 || beta == 0) return;
  102. if( bS[0] < 4 ) {
  103. int8_t tc[4];
  104. tc[0] = (tc0_table+52)[index_a][bS[0]];
  105. tc[1] = (tc0_table+52)[index_a][bS[1]];
  106. tc[2] = (tc0_table+52)[index_a][bS[2]];
  107. tc[3] = (tc0_table+52)[index_a][bS[3]];
  108. h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
  109. } else {
  110. h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
  111. }
  112. }
  113. static void av_noinline filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
  114. const int index_a = qp + h->slice_alpha_c0_offset;
  115. const int alpha = (alpha_table+52)[index_a];
  116. const int beta = (beta_table+52)[qp + h->slice_beta_offset];
  117. if (alpha ==0 || beta == 0) return;
  118. if( bS[0] < 4 ) {
  119. int8_t tc[4];
  120. tc[0] = (tc0_table+52)[index_a][bS[0]]+1;
  121. tc[1] = (tc0_table+52)[index_a][bS[1]]+1;
  122. tc[2] = (tc0_table+52)[index_a][bS[2]]+1;
  123. tc[3] = (tc0_table+52)[index_a][bS[3]]+1;
  124. h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
  125. } else {
  126. h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
  127. }
  128. }
  129. static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
  130. int i;
  131. for( i = 0; i < 8; i++, pix += stride) {
  132. int index_a;
  133. int alpha;
  134. int beta;
  135. const int bS_index = (i >> 1) * bsi;
  136. if( bS[bS_index] == 0 ) {
  137. continue;
  138. }
  139. index_a = qp + h->slice_alpha_c0_offset;
  140. alpha = (alpha_table+52)[index_a];
  141. beta = (beta_table+52)[qp + h->slice_beta_offset];
  142. if( bS[bS_index] < 4 ) {
  143. const int tc0 = (tc0_table+52)[index_a][bS[bS_index]];
  144. const int p0 = pix[-1];
  145. const int p1 = pix[-2];
  146. const int p2 = pix[-3];
  147. const int q0 = pix[0];
  148. const int q1 = pix[1];
  149. const int q2 = pix[2];
  150. if( FFABS( p0 - q0 ) < alpha &&
  151. FFABS( p1 - p0 ) < beta &&
  152. FFABS( q1 - q0 ) < beta ) {
  153. int tc = tc0;
  154. int i_delta;
  155. if( FFABS( p2 - p0 ) < beta ) {
  156. pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
  157. tc++;
  158. }
  159. if( FFABS( q2 - q0 ) < beta ) {
  160. pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
  161. tc++;
  162. }
  163. i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
  164. pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
  165. pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
  166. tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
  167. }
  168. }else{
  169. const int p0 = pix[-1];
  170. const int p1 = pix[-2];
  171. const int p2 = pix[-3];
  172. const int q0 = pix[0];
  173. const int q1 = pix[1];
  174. const int q2 = pix[2];
  175. if( FFABS( p0 - q0 ) < alpha &&
  176. FFABS( p1 - p0 ) < beta &&
  177. FFABS( q1 - q0 ) < beta ) {
  178. if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
  179. if( FFABS( p2 - p0 ) < beta)
  180. {
  181. const int p3 = pix[-4];
  182. /* p0', p1', p2' */
  183. pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
  184. pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
  185. pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
  186. } else {
  187. /* p0' */
  188. pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
  189. }
  190. if( FFABS( q2 - q0 ) < beta)
  191. {
  192. const int q3 = pix[3];
  193. /* q0', q1', q2' */
  194. pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
  195. pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
  196. pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
  197. } else {
  198. /* q0' */
  199. pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
  200. }
  201. }else{
  202. /* p0', q0' */
  203. pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
  204. pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
  205. }
  206. tprintf(h->s.avctx, "filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
  207. }
  208. }
  209. }
  210. }
  211. static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
  212. int i;
  213. for( i = 0; i < 4; i++, pix += stride) {
  214. int index_a;
  215. int alpha;
  216. int beta;
  217. const int bS_index = i*bsi;
  218. if( bS[bS_index] == 0 ) {
  219. continue;
  220. }
  221. index_a = qp + h->slice_alpha_c0_offset;
  222. alpha = (alpha_table+52)[index_a];
  223. beta = (beta_table+52)[qp + h->slice_beta_offset];
  224. if( bS[bS_index] < 4 ) {
  225. const int tc = (tc0_table+52)[index_a][bS[bS_index]] + 1;
  226. const int p0 = pix[-1];
  227. const int p1 = pix[-2];
  228. const int q0 = pix[0];
  229. const int q1 = pix[1];
  230. if( FFABS( p0 - q0 ) < alpha &&
  231. FFABS( p1 - p0 ) < beta &&
  232. FFABS( q1 - q0 ) < beta ) {
  233. const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
  234. pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
  235. pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
  236. tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
  237. }
  238. }else{
  239. const int p0 = pix[-1];
  240. const int p1 = pix[-2];
  241. const int q0 = pix[0];
  242. const int q1 = pix[1];
  243. if( FFABS( p0 - q0 ) < alpha &&
  244. FFABS( p1 - p0 ) < beta &&
  245. FFABS( q1 - q0 ) < beta ) {
  246. pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
  247. pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
  248. tprintf(h->s.avctx, "filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
  249. }
  250. }
  251. }
  252. }
  253. static void av_noinline filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
  254. const int index_a = qp + h->slice_alpha_c0_offset;
  255. const int alpha = (alpha_table+52)[index_a];
  256. const int beta = (beta_table+52)[qp + h->slice_beta_offset];
  257. if (alpha ==0 || beta == 0) return;
  258. if( bS[0] < 4 ) {
  259. int8_t tc[4];
  260. tc[0] = (tc0_table+52)[index_a][bS[0]];
  261. tc[1] = (tc0_table+52)[index_a][bS[1]];
  262. tc[2] = (tc0_table+52)[index_a][bS[2]];
  263. tc[3] = (tc0_table+52)[index_a][bS[3]];
  264. h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
  265. } else {
  266. h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
  267. }
  268. }
  269. static void av_noinline filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
  270. const int index_a = qp + h->slice_alpha_c0_offset;
  271. const int alpha = (alpha_table+52)[index_a];
  272. const int beta = (beta_table+52)[qp + h->slice_beta_offset];
  273. if (alpha ==0 || beta == 0) return;
  274. if( bS[0] < 4 ) {
  275. int8_t tc[4];
  276. tc[0] = (tc0_table+52)[index_a][bS[0]]+1;
  277. tc[1] = (tc0_table+52)[index_a][bS[1]]+1;
  278. tc[2] = (tc0_table+52)[index_a][bS[2]]+1;
  279. tc[3] = (tc0_table+52)[index_a][bS[3]]+1;
  280. h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
  281. } else {
  282. h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
  283. }
  284. }
  285. 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) {
  286. MpegEncContext * const s = &h->s;
  287. int mb_y_firstrow = s->picture_structure == PICT_BOTTOM_FIELD;
  288. int mb_xy, mb_type;
  289. int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
  290. mb_xy = h->mb_xy;
  291. if(mb_x==0 || mb_y==mb_y_firstrow || !s->dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff ||
  292. !(h->slice_type_nos == FF_I_TYPE ||
  293. h->slice_type_nos == FF_P_TYPE ||
  294. (s->flags2 & CODEC_FLAG2_FAST)) ||
  295. (h->deblocking_filter == 2 && (h->slice_num != h->slice_table[h->top_mb_xy] ||
  296. h->slice_num != h->slice_table[mb_xy - 1]))) {
  297. ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
  298. return;
  299. }
  300. assert(!FRAME_MBAFF);
  301. mb_type = s->current_picture.mb_type[mb_xy];
  302. qp = s->current_picture.qscale_table[mb_xy];
  303. qp0 = s->current_picture.qscale_table[mb_xy-1];
  304. qp1 = s->current_picture.qscale_table[h->top_mb_xy];
  305. qpc = get_chroma_qp( h, 0, qp );
  306. qpc0 = get_chroma_qp( h, 0, qp0 );
  307. qpc1 = get_chroma_qp( h, 0, qp1 );
  308. qp0 = (qp + qp0 + 1) >> 1;
  309. qp1 = (qp + qp1 + 1) >> 1;
  310. qpc0 = (qpc + qpc0 + 1) >> 1;
  311. qpc1 = (qpc + qpc1 + 1) >> 1;
  312. qp_thresh = 15 - h->slice_alpha_c0_offset;
  313. if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
  314. qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
  315. return;
  316. if( IS_INTRA(mb_type) ) {
  317. int16_t bS4[4] = {4,4,4,4};
  318. int16_t bS3[4] = {3,3,3,3};
  319. int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
  320. if( IS_8x8DCT(mb_type) ) {
  321. filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );
  322. filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );
  323. filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 );
  324. filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );
  325. } else {
  326. filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );
  327. filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp );
  328. filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );
  329. filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp );
  330. filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 );
  331. filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp );
  332. filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );
  333. filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp );
  334. }
  335. filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 );
  336. filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc );
  337. filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 );
  338. filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc );
  339. filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1 );
  340. filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc );
  341. filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1 );
  342. filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc );
  343. return;
  344. } else {
  345. DECLARE_ALIGNED_8(int16_t, bS[2][4][4]);
  346. uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
  347. int edges;
  348. if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
  349. edges = 4;
  350. bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;
  351. } else {
  352. int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 :
  353. (mb_type & MB_TYPE_16x8) ? 1 : 0;
  354. int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16))
  355. && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16))
  356. ? 3 : 0;
  357. int step = IS_8x8DCT(mb_type) ? 2 : 1;
  358. edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
  359. s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
  360. (h->slice_type_nos == FF_B_TYPE), edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
  361. }
  362. if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) )
  363. bSv[0][0] = 0x0004000400040004ULL;
  364. if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) )
  365. bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;
  366. #define FILTER(hv,dir,edge)\
  367. if(bSv[dir][edge]) {\
  368. filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\
  369. if(!(edge&1)) {\
  370. filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\
  371. filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\
  372. }\
  373. }
  374. if( edges == 1 ) {
  375. FILTER(v,0,0);
  376. FILTER(h,1,0);
  377. } else if( IS_8x8DCT(mb_type) ) {
  378. FILTER(v,0,0);
  379. FILTER(v,0,2);
  380. FILTER(h,1,0);
  381. FILTER(h,1,2);
  382. } else {
  383. FILTER(v,0,0);
  384. FILTER(v,0,1);
  385. FILTER(v,0,2);
  386. FILTER(v,0,3);
  387. FILTER(h,1,0);
  388. FILTER(h,1,1);
  389. FILTER(h,1,2);
  390. FILTER(h,1,3);
  391. }
  392. #undef FILTER
  393. }
  394. }
  395. 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 dir) {
  396. MpegEncContext * const s = &h->s;
  397. int edge;
  398. const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
  399. const int mbm_type = s->current_picture.mb_type[mbm_xy];
  400. int start = h->slice_table[mbm_xy] == 0xFFFF ? 1 : 0;
  401. const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP))
  402. == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4;
  403. // how often to recheck mv-based bS when iterating between edges
  404. const int mask_edge = (mb_type & (MB_TYPE_16x16 | (MB_TYPE_16x8 << dir))) ? 3 :
  405. (mb_type & (MB_TYPE_8x16 >> dir)) ? 1 : 0;
  406. // how often to recheck mv-based bS when iterating along each edge
  407. const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
  408. if (first_vertical_edge_done) {
  409. start = 1;
  410. }
  411. if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_num)
  412. start = 1;
  413. if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0
  414. && !IS_INTERLACED(mb_type)
  415. && IS_INTERLACED(mbm_type)
  416. ) {
  417. // This is a special case in the norm where the filtering must
  418. // be done twice (one each of the field) even if we are in a
  419. // frame macroblock.
  420. //
  421. unsigned int tmp_linesize = 2 * linesize;
  422. unsigned int tmp_uvlinesize = 2 * uvlinesize;
  423. int mbn_xy = mb_xy - 2 * s->mb_stride;
  424. int qp;
  425. int i, j;
  426. int16_t bS[4];
  427. for(j=0; j<2; j++, mbn_xy += s->mb_stride){
  428. if( IS_INTRA(mb_type) ||
  429. IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) {
  430. bS[0] = bS[1] = bS[2] = bS[3] = 3;
  431. } else {
  432. const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy];
  433. for( i = 0; i < 4; i++ ) {
  434. if( h->non_zero_count_cache[scan8[0]+i] != 0 ||
  435. mbn_nnz[i+4+3*8] != 0 )
  436. bS[i] = 2;
  437. else
  438. bS[i] = 1;
  439. }
  440. }
  441. // Do not use s->qscale as luma quantizer because it has not the same
  442. // value in IPCM macroblocks.
  443. qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
  444. 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);
  445. { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  446. filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp );
  447. filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
  448. ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  449. filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
  450. ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  451. }
  452. start = 1;
  453. }
  454. /* Calculate bS */
  455. for( edge = start; edge < edges; edge++ ) {
  456. /* mbn_xy: neighbor macroblock */
  457. const int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
  458. const int mbn_type = s->current_picture.mb_type[mbn_xy];
  459. int16_t bS[4];
  460. int qp;
  461. if( (edge&1) && IS_8x8DCT(mb_type) )
  462. continue;
  463. if( IS_INTRA(mb_type) ||
  464. IS_INTRA(mbn_type) ) {
  465. int value;
  466. if (edge == 0) {
  467. if ( (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type))
  468. || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
  469. ) {
  470. value = 4;
  471. } else {
  472. value = 3;
  473. }
  474. } else {
  475. value = 3;
  476. }
  477. bS[0] = bS[1] = bS[2] = bS[3] = value;
  478. } else {
  479. int i, l;
  480. int mv_done;
  481. if( edge & mask_edge ) {
  482. bS[0] = bS[1] = bS[2] = bS[3] = 0;
  483. mv_done = 1;
  484. }
  485. else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) {
  486. bS[0] = bS[1] = bS[2] = bS[3] = 1;
  487. mv_done = 1;
  488. }
  489. else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
  490. int b_idx= 8 + 4 + edge * (dir ? 8:1);
  491. int bn_idx= b_idx - (dir ? 8:1);
  492. int v = 0;
  493. for( l = 0; !v && l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
  494. v |= h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] |
  495. h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U |
  496. FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;
  497. }
  498. if(h->slice_type_nos == FF_B_TYPE && v){
  499. v=0;
  500. for( l = 0; !v && l < 2; l++ ) {
  501. int ln= 1-l;
  502. v |= h->ref_cache[l][b_idx] != h->ref_cache[ln][bn_idx] |
  503. h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U |
  504. FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit;
  505. }
  506. }
  507. bS[0] = bS[1] = bS[2] = bS[3] = v;
  508. mv_done = 1;
  509. }
  510. else
  511. mv_done = 0;
  512. for( i = 0; i < 4; i++ ) {
  513. int x = dir == 0 ? edge : i;
  514. int y = dir == 0 ? i : edge;
  515. int b_idx= 8 + 4 + x + 8*y;
  516. int bn_idx= b_idx - (dir ? 8:1);
  517. if( h->non_zero_count_cache[b_idx] |
  518. h->non_zero_count_cache[bn_idx] ) {
  519. bS[i] = 2;
  520. }
  521. else if(!mv_done)
  522. {
  523. bS[i] = 0;
  524. for( l = 0; l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
  525. if( h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] |
  526. h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U |
  527. FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {
  528. bS[i] = 1;
  529. break;
  530. }
  531. }
  532. if(h->slice_type_nos == FF_B_TYPE && bS[i]){
  533. bS[i] = 0;
  534. for( l = 0; l < 2; l++ ) {
  535. int ln= 1-l;
  536. if( h->ref_cache[l][b_idx] != h->ref_cache[ln][bn_idx] |
  537. h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U |
  538. FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit ) {
  539. bS[i] = 1;
  540. break;
  541. }
  542. }
  543. }
  544. }
  545. }
  546. if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
  547. continue;
  548. }
  549. /* Filter edge */
  550. // Do not use s->qscale as luma quantizer because it has not the same
  551. // value in IPCM macroblocks.
  552. qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
  553. //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]);
  554. 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);
  555. //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  556. if( dir == 0 ) {
  557. filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
  558. if( (edge&1) == 0 ) {
  559. filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS,
  560. ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  561. filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS,
  562. ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  563. }
  564. } else {
  565. filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
  566. if( (edge&1) == 0 ) {
  567. filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS,
  568. ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  569. filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS,
  570. ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  571. }
  572. }
  573. }
  574. }
  575. 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) {
  576. MpegEncContext * const s = &h->s;
  577. const int mb_xy= mb_x + mb_y*s->mb_stride;
  578. const int mb_type = s->current_picture.mb_type[mb_xy];
  579. const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
  580. int first_vertical_edge_done = 0;
  581. av_unused int dir;
  582. int list;
  583. if (FRAME_MBAFF
  584. // left mb is in picture
  585. && h->slice_table[mb_xy-1] != 0xFFFF
  586. // and current and left pair do not have the same interlaced type
  587. && (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
  588. // and left mb is in the same slice if deblocking_filter == 2
  589. && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_num)) {
  590. /* First vertical edge is different in MBAFF frames
  591. * There are 8 different bS to compute and 2 different Qp
  592. */
  593. const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;
  594. const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };
  595. int16_t bS[8];
  596. int qp[2];
  597. int bqp[2];
  598. int rqp[2];
  599. int mb_qp, mbn0_qp, mbn1_qp;
  600. int i;
  601. first_vertical_edge_done = 1;
  602. if( IS_INTRA(mb_type) )
  603. bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4;
  604. else {
  605. for( i = 0; i < 8; i++ ) {
  606. int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];
  607. if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )
  608. bS[i] = 4;
  609. else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||
  610. ((!h->pps.cabac && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])) ?
  611. (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
  612. :
  613. h->non_zero_count[mbn_xy][7+(MB_FIELD ? (i&3) : (i>>2)+(mb_y&1)*2)*8]))
  614. bS[i] = 2;
  615. else
  616. bS[i] = 1;
  617. }
  618. }
  619. mb_qp = s->current_picture.qscale_table[mb_xy];
  620. mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];
  621. mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];
  622. qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
  623. bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
  624. get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
  625. rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
  626. get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
  627. qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
  628. bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
  629. get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
  630. rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
  631. get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
  632. /* Filter edge */
  633. 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);
  634. { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  635. if(MB_FIELD){
  636. filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
  637. filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
  638. filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
  639. filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
  640. filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
  641. filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
  642. }else{
  643. filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
  644. filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
  645. filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
  646. filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
  647. filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
  648. filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
  649. }
  650. }
  651. #if CONFIG_SMALL
  652. for( dir = 0; dir < 2; dir++ )
  653. 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, dir);
  654. #else
  655. 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, 0);
  656. filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);
  657. #endif
  658. }