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