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.

724 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. #if ARCH_X86
  34. #include "x86/h264_i386.h"
  35. #endif
  36. //#undef NDEBUG
  37. #include <assert.h>
  38. /* Deblocking filter (p153) */
  39. static const uint8_t alpha_table[52*3] = {
  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, 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,
  45. 0, 0, 0, 0, 0, 0, 4, 4, 5, 6,
  46. 7, 8, 9, 10, 12, 13, 15, 17, 20, 22,
  47. 25, 28, 32, 36, 40, 45, 50, 56, 63, 71,
  48. 80, 90,101,113,127,144,162,182,203,226,
  49. 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. 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. };
  55. static const uint8_t beta_table[52*3] = {
  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, 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,
  61. 0, 0, 0, 0, 0, 0, 2, 2, 2, 3,
  62. 3, 3, 3, 4, 4, 4, 6, 6, 7, 7,
  63. 8, 8, 9, 9, 10, 10, 11, 11, 12, 12,
  64. 13, 13, 14, 14, 15, 15, 16, 16, 17, 17,
  65. 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. 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. };
  71. static const uint8_t tc0_table[52*3][4] = {
  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 }, {-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 },
  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 }, {-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, 1 },
  84. {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 0, 1 }, {-1, 0, 1, 1 }, {-1, 0, 1, 1 }, {-1, 1, 1, 1 },
  85. {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 1 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 }, {-1, 1, 1, 2 },
  86. {-1, 1, 1, 2 }, {-1, 1, 2, 3 }, {-1, 1, 2, 3 }, {-1, 2, 2, 3 }, {-1, 2, 2, 4 }, {-1, 2, 3, 4 },
  87. {-1, 2, 3, 4 }, {-1, 3, 3, 5 }, {-1, 3, 4, 6 }, {-1, 3, 4, 6 }, {-1, 4, 5, 7 }, {-1, 4, 5, 8 },
  88. {-1, 4, 6, 9 }, {-1, 5, 7,10 }, {-1, 6, 8,11 }, {-1, 6, 8,13 }, {-1, 7,10,14 }, {-1, 8,11,16 },
  89. {-1, 9,12,18 }, {-1,10,13,20 }, {-1,11,15,23 }, {-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 }, {-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 },
  99. };
  100. static void av_noinline filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
  101. const int index_a = qp + h->slice_alpha_c0_offset;
  102. const int alpha = (alpha_table+52)[index_a];
  103. const int beta = (beta_table+52)[qp + h->slice_beta_offset];
  104. if (alpha ==0 || beta == 0) return;
  105. if( bS[0] < 4 ) {
  106. int8_t tc[4];
  107. tc[0] = (tc0_table+52)[index_a][bS[0]];
  108. tc[1] = (tc0_table+52)[index_a][bS[1]];
  109. tc[2] = (tc0_table+52)[index_a][bS[2]];
  110. tc[3] = (tc0_table+52)[index_a][bS[3]];
  111. h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
  112. } else {
  113. h->s.dsp.h264_h_loop_filter_luma_intra(pix, stride, alpha, beta);
  114. }
  115. }
  116. static void av_noinline filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
  117. const int index_a = qp + h->slice_alpha_c0_offset;
  118. const int alpha = (alpha_table+52)[index_a];
  119. const int beta = (beta_table+52)[qp + 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+52)[index_a][bS[0]]+1;
  124. tc[1] = (tc0_table+52)[index_a][bS[1]]+1;
  125. tc[2] = (tc0_table+52)[index_a][bS[2]]+1;
  126. tc[3] = (tc0_table+52)[index_a][bS[3]]+1;
  127. h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
  128. } else {
  129. h->s.dsp.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, int16_t bS[4], int bsi, int qp ) {
  133. int i;
  134. for( i = 0; i < 8; i++, pix += stride) {
  135. int index_a;
  136. int alpha;
  137. int beta;
  138. const int bS_index = (i >> 1) * bsi;
  139. if( bS[bS_index] == 0 ) {
  140. continue;
  141. }
  142. index_a = qp + h->slice_alpha_c0_offset;
  143. alpha = (alpha_table+52)[index_a];
  144. beta = (beta_table+52)[qp + h->slice_beta_offset];
  145. if( bS[bS_index] < 4 ) {
  146. const int tc0 = (tc0_table+52)[index_a][bS[bS_index]];
  147. const int p0 = pix[-1];
  148. const int p1 = pix[-2];
  149. const int p2 = pix[-3];
  150. const int q0 = pix[0];
  151. const int q1 = pix[1];
  152. const int q2 = pix[2];
  153. if( FFABS( p0 - q0 ) < alpha &&
  154. FFABS( p1 - p0 ) < beta &&
  155. FFABS( q1 - q0 ) < beta ) {
  156. int tc = tc0;
  157. int i_delta;
  158. if( FFABS( p2 - p0 ) < beta ) {
  159. pix[-2] = p1 + av_clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
  160. tc++;
  161. }
  162. if( FFABS( q2 - q0 ) < beta ) {
  163. pix[1] = q1 + av_clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
  164. tc++;
  165. }
  166. i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
  167. pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
  168. pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
  169. 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);
  170. }
  171. }else{
  172. const int p0 = pix[-1];
  173. const int p1 = pix[-2];
  174. const int p2 = pix[-3];
  175. const int q0 = pix[0];
  176. const int q1 = pix[1];
  177. const int q2 = pix[2];
  178. if( FFABS( p0 - q0 ) < alpha &&
  179. FFABS( p1 - p0 ) < beta &&
  180. FFABS( q1 - q0 ) < beta ) {
  181. if(FFABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
  182. if( FFABS( p2 - p0 ) < beta)
  183. {
  184. const int p3 = pix[-4];
  185. /* p0', p1', p2' */
  186. pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
  187. pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
  188. pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
  189. } else {
  190. /* p0' */
  191. pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
  192. }
  193. if( FFABS( q2 - q0 ) < beta)
  194. {
  195. const int q3 = pix[3];
  196. /* q0', q1', q2' */
  197. pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
  198. pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
  199. pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
  200. } else {
  201. /* q0' */
  202. pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
  203. }
  204. }else{
  205. /* p0', q0' */
  206. pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
  207. pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
  208. }
  209. 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]);
  210. }
  211. }
  212. }
  213. }
  214. static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int bsi, int qp ) {
  215. int i;
  216. for( i = 0; i < 4; i++, pix += stride) {
  217. int index_a;
  218. int alpha;
  219. int beta;
  220. const int bS_index = i*bsi;
  221. if( bS[bS_index] == 0 ) {
  222. continue;
  223. }
  224. index_a = qp + h->slice_alpha_c0_offset;
  225. alpha = (alpha_table+52)[index_a];
  226. beta = (beta_table+52)[qp + h->slice_beta_offset];
  227. if( bS[bS_index] < 4 ) {
  228. const int tc = (tc0_table+52)[index_a][bS[bS_index]] + 1;
  229. const int p0 = pix[-1];
  230. const int p1 = pix[-2];
  231. const int q0 = pix[0];
  232. const int q1 = pix[1];
  233. if( FFABS( p0 - q0 ) < alpha &&
  234. FFABS( p1 - p0 ) < beta &&
  235. FFABS( q1 - q0 ) < beta ) {
  236. const int i_delta = av_clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
  237. pix[-1] = av_clip_uint8( p0 + i_delta ); /* p0' */
  238. pix[0] = av_clip_uint8( q0 - i_delta ); /* q0' */
  239. 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);
  240. }
  241. }else{
  242. const int p0 = pix[-1];
  243. const int p1 = pix[-2];
  244. const int q0 = pix[0];
  245. const int q1 = pix[1];
  246. if( FFABS( p0 - q0 ) < alpha &&
  247. FFABS( p1 - p0 ) < beta &&
  248. FFABS( q1 - q0 ) < beta ) {
  249. pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
  250. pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
  251. 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]);
  252. }
  253. }
  254. }
  255. }
  256. static void av_noinline filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
  257. const int index_a = qp + h->slice_alpha_c0_offset;
  258. const int alpha = (alpha_table+52)[index_a];
  259. const int beta = (beta_table+52)[qp + h->slice_beta_offset];
  260. if (alpha ==0 || beta == 0) return;
  261. if( bS[0] < 4 ) {
  262. int8_t tc[4];
  263. tc[0] = (tc0_table+52)[index_a][bS[0]];
  264. tc[1] = (tc0_table+52)[index_a][bS[1]];
  265. tc[2] = (tc0_table+52)[index_a][bS[2]];
  266. tc[3] = (tc0_table+52)[index_a][bS[3]];
  267. h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
  268. } else {
  269. h->s.dsp.h264_v_loop_filter_luma_intra(pix, stride, alpha, beta);
  270. }
  271. }
  272. static void av_noinline filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int16_t bS[4], int qp ) {
  273. const int index_a = qp + h->slice_alpha_c0_offset;
  274. const int alpha = (alpha_table+52)[index_a];
  275. const int beta = (beta_table+52)[qp + h->slice_beta_offset];
  276. if (alpha ==0 || beta == 0) return;
  277. if( bS[0] < 4 ) {
  278. int8_t tc[4];
  279. tc[0] = (tc0_table+52)[index_a][bS[0]]+1;
  280. tc[1] = (tc0_table+52)[index_a][bS[1]]+1;
  281. tc[2] = (tc0_table+52)[index_a][bS[2]]+1;
  282. tc[3] = (tc0_table+52)[index_a][bS[3]]+1;
  283. h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
  284. } else {
  285. h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
  286. }
  287. }
  288. 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) {
  289. MpegEncContext * const s = &h->s;
  290. int mb_y_firstrow = s->picture_structure == PICT_BOTTOM_FIELD;
  291. int mb_xy, mb_type;
  292. int qp, qp0, qp1, qpc, qpc0, qpc1, qp_thresh;
  293. mb_xy = h->mb_xy;
  294. if(mb_x==0 || mb_y==mb_y_firstrow || !s->dsp.h264_loop_filter_strength || h->pps.chroma_qp_diff ||
  295. !(h->slice_type_nos == FF_I_TYPE ||
  296. h->slice_type_nos == FF_P_TYPE ||
  297. (s->flags2 & CODEC_FLAG2_FAST)) ||
  298. (h->deblocking_filter == 2 && (h->slice_num != h->slice_table[h->top_mb_xy] ||
  299. h->slice_num != h->slice_table[mb_xy - 1]))) {
  300. ff_h264_filter_mb(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize);
  301. return;
  302. }
  303. assert(!FRAME_MBAFF);
  304. mb_type = s->current_picture.mb_type[mb_xy];
  305. qp = s->current_picture.qscale_table[mb_xy];
  306. qp0 = s->current_picture.qscale_table[mb_xy-1];
  307. qp1 = s->current_picture.qscale_table[h->top_mb_xy];
  308. qpc = get_chroma_qp( h, 0, qp );
  309. qpc0 = get_chroma_qp( h, 0, qp0 );
  310. qpc1 = get_chroma_qp( h, 0, qp1 );
  311. qp0 = (qp + qp0 + 1) >> 1;
  312. qp1 = (qp + qp1 + 1) >> 1;
  313. qpc0 = (qpc + qpc0 + 1) >> 1;
  314. qpc1 = (qpc + qpc1 + 1) >> 1;
  315. qp_thresh = 15 - h->slice_alpha_c0_offset;
  316. if(qp <= qp_thresh && qp0 <= qp_thresh && qp1 <= qp_thresh &&
  317. qpc <= qp_thresh && qpc0 <= qp_thresh && qpc1 <= qp_thresh)
  318. return;
  319. if( IS_INTRA(mb_type) ) {
  320. int16_t bS4[4] = {4,4,4,4};
  321. int16_t bS3[4] = {3,3,3,3};
  322. int16_t *bSH = FIELD_PICTURE ? bS3 : bS4;
  323. if( IS_8x8DCT(mb_type) ) {
  324. filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );
  325. filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );
  326. filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 );
  327. filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );
  328. } else {
  329. filter_mb_edgev( h, &img_y[4*0], linesize, bS4, qp0 );
  330. filter_mb_edgev( h, &img_y[4*1], linesize, bS3, qp );
  331. filter_mb_edgev( h, &img_y[4*2], linesize, bS3, qp );
  332. filter_mb_edgev( h, &img_y[4*3], linesize, bS3, qp );
  333. filter_mb_edgeh( h, &img_y[4*0*linesize], linesize, bSH, qp1 );
  334. filter_mb_edgeh( h, &img_y[4*1*linesize], linesize, bS3, qp );
  335. filter_mb_edgeh( h, &img_y[4*2*linesize], linesize, bS3, qp );
  336. filter_mb_edgeh( h, &img_y[4*3*linesize], linesize, bS3, qp );
  337. }
  338. filter_mb_edgecv( h, &img_cb[2*0], uvlinesize, bS4, qpc0 );
  339. filter_mb_edgecv( h, &img_cb[2*2], uvlinesize, bS3, qpc );
  340. filter_mb_edgecv( h, &img_cr[2*0], uvlinesize, bS4, qpc0 );
  341. filter_mb_edgecv( h, &img_cr[2*2], uvlinesize, bS3, qpc );
  342. filter_mb_edgech( h, &img_cb[2*0*uvlinesize], uvlinesize, bSH, qpc1 );
  343. filter_mb_edgech( h, &img_cb[2*2*uvlinesize], uvlinesize, bS3, qpc );
  344. filter_mb_edgech( h, &img_cr[2*0*uvlinesize], uvlinesize, bSH, qpc1 );
  345. filter_mb_edgech( h, &img_cr[2*2*uvlinesize], uvlinesize, bS3, qpc );
  346. return;
  347. } else {
  348. DECLARE_ALIGNED_8(int16_t, bS[2][4][4]);
  349. uint64_t (*bSv)[4] = (uint64_t(*)[4])bS;
  350. int edges;
  351. if( IS_8x8DCT(mb_type) && (h->cbp&7) == 7 ) {
  352. edges = 4;
  353. bSv[0][0] = bSv[0][2] = bSv[1][0] = bSv[1][2] = 0x0002000200020002ULL;
  354. } else {
  355. int mask_edge1 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16)) ? 3 :
  356. (mb_type & MB_TYPE_16x8) ? 1 : 0;
  357. int mask_edge0 = (mb_type & (MB_TYPE_16x16 | MB_TYPE_8x16))
  358. && (s->current_picture.mb_type[mb_xy-1] & (MB_TYPE_16x16 | MB_TYPE_8x16))
  359. ? 3 : 0;
  360. int step = IS_8x8DCT(mb_type) ? 2 : 1;
  361. edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
  362. s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
  363. (h->slice_type_nos == FF_B_TYPE), edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
  364. }
  365. if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) )
  366. bSv[0][0] = 0x0004000400040004ULL;
  367. if( IS_INTRA(s->current_picture.mb_type[h->top_mb_xy]) )
  368. bSv[1][0] = FIELD_PICTURE ? 0x0003000300030003ULL : 0x0004000400040004ULL;
  369. #define FILTER(hv,dir,edge)\
  370. if(bSv[dir][edge]) {\
  371. filter_mb_edge##hv( h, &img_y[4*edge*(dir?linesize:1)], linesize, bS[dir][edge], edge ? qp : qp##dir );\
  372. if(!(edge&1)) {\
  373. filter_mb_edgec##hv( h, &img_cb[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\
  374. filter_mb_edgec##hv( h, &img_cr[2*edge*(dir?uvlinesize:1)], uvlinesize, bS[dir][edge], edge ? qpc : qpc##dir );\
  375. }\
  376. }
  377. if( edges == 1 ) {
  378. FILTER(v,0,0);
  379. FILTER(h,1,0);
  380. } else if( IS_8x8DCT(mb_type) ) {
  381. FILTER(v,0,0);
  382. FILTER(v,0,2);
  383. FILTER(h,1,0);
  384. FILTER(h,1,2);
  385. } else {
  386. FILTER(v,0,0);
  387. FILTER(v,0,1);
  388. FILTER(v,0,2);
  389. FILTER(v,0,3);
  390. FILTER(h,1,0);
  391. FILTER(h,1,1);
  392. FILTER(h,1,2);
  393. FILTER(h,1,3);
  394. }
  395. #undef FILTER
  396. }
  397. }
  398. 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) {
  399. MpegEncContext * const s = &h->s;
  400. int edge;
  401. const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
  402. const int mbm_type = s->current_picture.mb_type[mbm_xy];
  403. int start = h->slice_table[mbm_xy] == 0xFFFF ? 1 : 0;
  404. const int edges = (mb_type & (MB_TYPE_16x16|MB_TYPE_SKIP))
  405. == (MB_TYPE_16x16|MB_TYPE_SKIP) ? 1 : 4;
  406. // how often to recheck mv-based bS when iterating between edges
  407. const int mask_edge = (mb_type & (MB_TYPE_16x16 | (MB_TYPE_16x8 << dir))) ? 3 :
  408. (mb_type & (MB_TYPE_8x16 >> dir)) ? 1 : 0;
  409. // how often to recheck mv-based bS when iterating along each edge
  410. const int mask_par0 = mb_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir));
  411. if (first_vertical_edge_done) {
  412. start = 1;
  413. }
  414. if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_num)
  415. start = 1;
  416. if (FRAME_MBAFF && (dir == 1) && ((mb_y&1) == 0) && start == 0
  417. && !IS_INTERLACED(mb_type)
  418. && IS_INTERLACED(mbm_type)
  419. ) {
  420. // This is a special case in the norm where the filtering must
  421. // be done twice (one each of the field) even if we are in a
  422. // frame macroblock.
  423. //
  424. unsigned int tmp_linesize = 2 * linesize;
  425. unsigned int tmp_uvlinesize = 2 * uvlinesize;
  426. int mbn_xy = mb_xy - 2 * s->mb_stride;
  427. int qp;
  428. int i, j;
  429. int16_t bS[4];
  430. for(j=0; j<2; j++, mbn_xy += s->mb_stride){
  431. if( IS_INTRA(mb_type) ||
  432. IS_INTRA(s->current_picture.mb_type[mbn_xy]) ) {
  433. bS[0] = bS[1] = bS[2] = bS[3] = 3;
  434. } else {
  435. const uint8_t *mbn_nnz = h->non_zero_count[mbn_xy];
  436. for( i = 0; i < 4; i++ ) {
  437. if( h->non_zero_count_cache[scan8[0]+i] != 0 ||
  438. mbn_nnz[i+4+3*8] != 0 )
  439. bS[i] = 2;
  440. else
  441. bS[i] = 1;
  442. }
  443. }
  444. // Do not use s->qscale as luma quantizer because it has not the same
  445. // value in IPCM macroblocks.
  446. qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
  447. 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);
  448. { int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  449. filter_mb_edgeh( h, &img_y[j*linesize], tmp_linesize, bS, qp );
  450. filter_mb_edgech( h, &img_cb[j*uvlinesize], tmp_uvlinesize, bS,
  451. ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  452. filter_mb_edgech( h, &img_cr[j*uvlinesize], tmp_uvlinesize, bS,
  453. ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  454. }
  455. start = 1;
  456. }
  457. /* Calculate bS */
  458. for( edge = start; edge < edges; edge++ ) {
  459. /* mbn_xy: neighbor macroblock */
  460. const int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
  461. const int mbn_type = s->current_picture.mb_type[mbn_xy];
  462. int16_t bS[4];
  463. int qp;
  464. if( (edge&1) && IS_8x8DCT(mb_type) )
  465. continue;
  466. if( IS_INTRA(mb_type) ||
  467. IS_INTRA(mbn_type) ) {
  468. int value;
  469. if (edge == 0) {
  470. if ( (!IS_INTERLACED(mb_type) && !IS_INTERLACED(mbm_type))
  471. || ((FRAME_MBAFF || (s->picture_structure != PICT_FRAME)) && (dir == 0))
  472. ) {
  473. value = 4;
  474. } else {
  475. value = 3;
  476. }
  477. } else {
  478. value = 3;
  479. }
  480. bS[0] = bS[1] = bS[2] = bS[3] = value;
  481. } else {
  482. int i, l;
  483. int mv_done;
  484. if( edge & mask_edge ) {
  485. bS[0] = bS[1] = bS[2] = bS[3] = 0;
  486. mv_done = 1;
  487. }
  488. else if( FRAME_MBAFF && IS_INTERLACED(mb_type ^ mbn_type)) {
  489. bS[0] = bS[1] = bS[2] = bS[3] = 1;
  490. mv_done = 1;
  491. }
  492. else if( mask_par0 && (edge || (mbn_type & (MB_TYPE_16x16 | (MB_TYPE_8x16 >> dir)))) ) {
  493. int b_idx= 8 + 4 + edge * (dir ? 8:1);
  494. int bn_idx= b_idx - (dir ? 8:1);
  495. int v = 0;
  496. for( l = 0; !v && l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
  497. v |= h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] |
  498. h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U |
  499. FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit;
  500. }
  501. if(h->slice_type_nos == FF_B_TYPE && v){
  502. v=0;
  503. for( l = 0; !v && l < 2; l++ ) {
  504. int ln= 1-l;
  505. v |= h->ref_cache[l][b_idx] != h->ref_cache[ln][bn_idx] |
  506. h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U |
  507. FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit;
  508. }
  509. }
  510. bS[0] = bS[1] = bS[2] = bS[3] = v;
  511. mv_done = 1;
  512. }
  513. else
  514. mv_done = 0;
  515. for( i = 0; i < 4; i++ ) {
  516. int x = dir == 0 ? edge : i;
  517. int y = dir == 0 ? i : edge;
  518. int b_idx= 8 + 4 + x + 8*y;
  519. int bn_idx= b_idx - (dir ? 8:1);
  520. if( h->non_zero_count_cache[b_idx] |
  521. h->non_zero_count_cache[bn_idx] ) {
  522. bS[i] = 2;
  523. }
  524. else if(!mv_done)
  525. {
  526. bS[i] = 0;
  527. for( l = 0; l < 1 + (h->slice_type_nos == FF_B_TYPE); l++ ) {
  528. if( h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] |
  529. h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] + 3 >= 7U |
  530. FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= mvy_limit ) {
  531. bS[i] = 1;
  532. break;
  533. }
  534. }
  535. if(h->slice_type_nos == FF_B_TYPE && bS[i]){
  536. bS[i] = 0;
  537. for( l = 0; l < 2; l++ ) {
  538. int ln= 1-l;
  539. if( h->ref_cache[l][b_idx] != h->ref_cache[ln][bn_idx] |
  540. h->mv_cache[l][b_idx][0] - h->mv_cache[ln][bn_idx][0] + 3 >= 7U |
  541. FFABS( h->mv_cache[l][b_idx][1] - h->mv_cache[ln][bn_idx][1] ) >= mvy_limit ) {
  542. bS[i] = 1;
  543. break;
  544. }
  545. }
  546. }
  547. }
  548. }
  549. if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
  550. continue;
  551. }
  552. /* Filter edge */
  553. // Do not use s->qscale as luma quantizer because it has not the same
  554. // value in IPCM macroblocks.
  555. qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
  556. //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]);
  557. 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);
  558. //{ int i; for (i = 0; i < 4; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  559. if( dir == 0 ) {
  560. filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
  561. if( (edge&1) == 0 ) {
  562. filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS,
  563. ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  564. filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS,
  565. ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  566. }
  567. } else {
  568. filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
  569. if( (edge&1) == 0 ) {
  570. filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS,
  571. ( h->chroma_qp[0] + get_chroma_qp( h, 0, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  572. filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS,
  573. ( h->chroma_qp[1] + get_chroma_qp( h, 1, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1);
  574. }
  575. }
  576. }
  577. }
  578. 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) {
  579. MpegEncContext * const s = &h->s;
  580. const int mb_xy= mb_x + mb_y*s->mb_stride;
  581. const int mb_type = s->current_picture.mb_type[mb_xy];
  582. const int mvy_limit = IS_INTERLACED(mb_type) ? 2 : 4;
  583. int first_vertical_edge_done = 0;
  584. av_unused int dir;
  585. int list;
  586. if (FRAME_MBAFF
  587. // left mb is in picture
  588. && h->slice_table[mb_xy-1] != 0xFFFF
  589. // and current and left pair do not have the same interlaced type
  590. && (IS_INTERLACED(mb_type) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
  591. // and left mb is in the same slice if deblocking_filter == 2
  592. && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_num)) {
  593. /* First vertical edge is different in MBAFF frames
  594. * There are 8 different bS to compute and 2 different Qp
  595. */
  596. const int pair_xy = mb_x + (mb_y&~1)*s->mb_stride;
  597. const int left_mb_xy[2] = { pair_xy-1, pair_xy-1+s->mb_stride };
  598. int16_t bS[8];
  599. int qp[2];
  600. int bqp[2];
  601. int rqp[2];
  602. int mb_qp, mbn0_qp, mbn1_qp;
  603. int i;
  604. first_vertical_edge_done = 1;
  605. if( IS_INTRA(mb_type) )
  606. bS[0] = bS[1] = bS[2] = bS[3] = bS[4] = bS[5] = bS[6] = bS[7] = 4;
  607. else {
  608. for( i = 0; i < 8; i++ ) {
  609. int mbn_xy = MB_FIELD ? left_mb_xy[i>>2] : left_mb_xy[i&1];
  610. if( IS_INTRA( s->current_picture.mb_type[mbn_xy] ) )
  611. bS[i] = 4;
  612. else if( h->non_zero_count_cache[12+8*(i>>1)] != 0 ||
  613. ((!h->pps.cabac && IS_8x8DCT(s->current_picture.mb_type[mbn_xy])) ?
  614. (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
  615. :
  616. h->non_zero_count[mbn_xy][7+(MB_FIELD ? (i&3) : (i>>2)+(mb_y&1)*2)*8]))
  617. bS[i] = 2;
  618. else
  619. bS[i] = 1;
  620. }
  621. }
  622. mb_qp = s->current_picture.qscale_table[mb_xy];
  623. mbn0_qp = s->current_picture.qscale_table[left_mb_xy[0]];
  624. mbn1_qp = s->current_picture.qscale_table[left_mb_xy[1]];
  625. qp[0] = ( mb_qp + mbn0_qp + 1 ) >> 1;
  626. bqp[0] = ( get_chroma_qp( h, 0, mb_qp ) +
  627. get_chroma_qp( h, 0, mbn0_qp ) + 1 ) >> 1;
  628. rqp[0] = ( get_chroma_qp( h, 1, mb_qp ) +
  629. get_chroma_qp( h, 1, mbn0_qp ) + 1 ) >> 1;
  630. qp[1] = ( mb_qp + mbn1_qp + 1 ) >> 1;
  631. bqp[1] = ( get_chroma_qp( h, 0, mb_qp ) +
  632. get_chroma_qp( h, 0, mbn1_qp ) + 1 ) >> 1;
  633. rqp[1] = ( get_chroma_qp( h, 1, mb_qp ) +
  634. get_chroma_qp( h, 1, mbn1_qp ) + 1 ) >> 1;
  635. /* Filter edge */
  636. 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);
  637. { int i; for (i = 0; i < 8; i++) tprintf(s->avctx, " bS[%d]:%d", i, bS[i]); tprintf(s->avctx, "\n"); }
  638. if(MB_FIELD){
  639. filter_mb_mbaff_edgev ( h, img_y , linesize, bS , 1, qp [0] );
  640. filter_mb_mbaff_edgev ( h, img_y + 8* linesize, linesize, bS+4, 1, qp [1] );
  641. filter_mb_mbaff_edgecv( h, img_cb, uvlinesize, bS , 1, bqp[0] );
  642. filter_mb_mbaff_edgecv( h, img_cb + 4*uvlinesize, uvlinesize, bS+4, 1, bqp[1] );
  643. filter_mb_mbaff_edgecv( h, img_cr, uvlinesize, bS , 1, rqp[0] );
  644. filter_mb_mbaff_edgecv( h, img_cr + 4*uvlinesize, uvlinesize, bS+4, 1, rqp[1] );
  645. }else{
  646. filter_mb_mbaff_edgev ( h, img_y , 2* linesize, bS , 2, qp [0] );
  647. filter_mb_mbaff_edgev ( h, img_y + linesize, 2* linesize, bS+1, 2, qp [1] );
  648. filter_mb_mbaff_edgecv( h, img_cb, 2*uvlinesize, bS , 2, bqp[0] );
  649. filter_mb_mbaff_edgecv( h, img_cb + uvlinesize, 2*uvlinesize, bS+1, 2, bqp[1] );
  650. filter_mb_mbaff_edgecv( h, img_cr, 2*uvlinesize, bS , 2, rqp[0] );
  651. filter_mb_mbaff_edgecv( h, img_cr + uvlinesize, 2*uvlinesize, bS+1, 2, rqp[1] );
  652. }
  653. }
  654. #if CONFIG_SMALL
  655. for( dir = 0; dir < 2; dir++ )
  656. 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);
  657. #else
  658. 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);
  659. filter_mb_dir(h, mb_x, mb_y, img_y, img_cb, img_cr, linesize, uvlinesize, mb_xy, mb_type, mvy_limit, 0, 1);
  660. #endif
  661. }