API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Aug 4th, 2017
532
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 154.23 KB | None | 0 0
raw download clone embed print report
  1. From 167c2760cbc36911302ace046db5eef6fe1ea54a Mon Sep 17 00:00:00 2001
  2. From: Simon Horlick <simonhorlick@gmail.com>
  3. Date: Tue, 11 Jan 2011 20:05:54 +0000
  4. Subject: [PATCH 01/25] Save interlace decision for all macroblocks
  5.  
  6. ---
  7. common/common.h | 1 +
  8. common/frame.c | 3 +++
  9. common/frame.h | 1 +
  10. common/macroblock.c | 1 +
  11. encoder/encoder.c | 8 ++++++++
  12. 5 files changed, 14 insertions(+), 0 deletions(-)
  13.  
  14. diff --git a/common/common.h b/common/common.h
  15. index 868f526..231254f 100644
  16. --- a/common/common.h
  17. +++ b/common/common.h
  18. @@ -617,6 +617,7 @@ struct x264_t
  19. int8_t *mb_transform_size; /* transform_size_8x8_flag of each mb */
  20. uint16_t *slice_table; /* sh->first_mb of the slice that the indexed mb is part of
  21. * NOTE: this will fail on resolutions above 2^16 MBs... */
  22. + uint8_t *field;
  23.  
  24. /* buffer for weighted versions of the reference frames */
  25. pixel *p_weight_buf[X264_REF_MAX];
  26. diff --git a/common/frame.c b/common/frame.c
  27. index ca90539..eff8ca5 100644
  28. --- a/common/frame.c
  29. +++ b/common/frame.c
  30. @@ -145,6 +145,8 @@ x264_frame_t *x264_frame_new( x264_t *h, int b_fdec )
  31. frame->i_stride[0] * (frame->i_lines[0] + 2*i_padv) * sizeof(uint16_t) << h->frames.b_have_sub8x8_esa );
  32. frame->integral = (uint16_t*)frame->buffer[3] + frame->i_stride[0] * i_padv + PADH;
  33. }
  34. + if( h->param.b_interlaced )
  35. + CHECKED_MALLOC( frame->field, i_mb_count * sizeof(uint8_t) );
  36. }
  37. else /* fenc frame */
  38. {
  39. @@ -219,6 +221,7 @@ void x264_frame_delete( x264_frame_t *frame )
  40. x264_free( frame->i_inv_qscale_factor );
  41. x264_free( frame->i_row_bits );
  42. x264_free( frame->f_row_qp );
  43. + x264_free( frame->field );
  44. x264_free( frame->mb_type );
  45. x264_free( frame->mb_partition );
  46. x264_free( frame->mv[0] );
  47. diff --git a/common/frame.h b/common/frame.h
  48. index 38d0bf2..0e0ab3d 100644
  49. --- a/common/frame.h
  50. +++ b/common/frame.h
  51. @@ -92,6 +92,7 @@ typedef struct x264_frame
  52. int16_t (*mv[2])[2];
  53. int16_t (*mv16x16)[2];
  54. int16_t (*lowres_mvs[2][X264_BFRAME_MAX+1])[2];
  55. + uint8_t *field;
  56.  
  57. /* Stored as (lists_used << LOWRES_COST_SHIFT) + (cost).
  58. * Doesn't need special addressing for intra cost because
  59. diff --git a/common/macroblock.c b/common/macroblock.c
  60. index 24c2af9..569d544 100644
  61. --- a/common/macroblock.c
  62. +++ b/common/macroblock.c
  63. @@ -371,6 +371,7 @@ void x264_macroblock_slice_init( x264_t *h )
  64. h->mb.ref[1] = h->fdec->ref[1];
  65. h->mb.type = h->fdec->mb_type;
  66. h->mb.partition = h->fdec->mb_partition;
  67. + h->mb.field = h->fdec->field;
  68.  
  69. h->fdec->i_ref[0] = h->i_ref[0];
  70. h->fdec->i_ref[1] = h->i_ref[1];
  71. diff --git a/encoder/encoder.c b/encoder/encoder.c
  72. index 67646df..af1342b 100644
  73. --- a/encoder/encoder.c
  74. +++ b/encoder/encoder.c
  75. @@ -1956,6 +1956,14 @@ static int x264_slice_write( x264_t *h )
  76. }
  77. }
  78.  
  79. + if( h->param.b_interlaced )
  80. + {
  81. + if( !(i_mb_y&1) )
  82. + h->mb.b_interlaced = 1;
  83. + x264_zigzag_init( h->param.cpu, &h->zigzagf, h->mb.b_interlaced );
  84. + h->mb.field[mb_xy] = h->mb.b_interlaced;
  85. + }
  86. +
  87. if( i_mb_x == 0 && !h->mb.b_reencode_mb )
  88. x264_fdec_filter_row( h, i_mb_y, 1 );
  89.  
  90. --
  91. 1.7.4
  92.  
  93.  
  94. From 2847b697f1fdbee1b8c3128895f2a50c1cba606e Mon Sep 17 00:00:00 2001
  95. From: Simon Horlick <simonhorlick@gmail.com>
  96. Date: Tue, 11 Jan 2011 20:09:00 +0000
  97. Subject: [PATCH 02/25] Disable adaptive mbaff when subme 0 is used
  98.  
  99. ---
  100. common/common.h | 1 +
  101. encoder/encoder.c | 13 ++++++++++---
  102. x264.h | 1 +
  103. 3 files changed, 12 insertions(+), 3 deletions(-)
  104.  
  105. diff --git a/common/common.h b/common/common.h
  106. index 231254f..75c4d59 100644
  107. --- a/common/common.h
  108. +++ b/common/common.h
  109. @@ -568,6 +568,7 @@ struct x264_t
  110. int i_psy_trellis; /* Psy trellis strength--fixed point value*/
  111.  
  112. int b_interlaced;
  113. + int b_adaptive_mbaff;
  114.  
  115. /* Allowed qpel MV range to stay within the picture + emulated edge pixels */
  116. int mv_min[2];
  117. diff --git a/encoder/encoder.c b/encoder/encoder.c
  118. index af1342b..9f294d1 100644
  119. --- a/encoder/encoder.c
  120. +++ b/encoder/encoder.c
  121. @@ -606,6 +606,10 @@ static int x264_validate_parameters( x264_t *h )
  122. x264_log( h, X264_LOG_WARNING, "subme=0 + direct=temporal is not supported\n" );
  123. h->param.analyse.i_direct_mv_pred = X264_DIRECT_PRED_SPATIAL;
  124. }
  125. + /* Adaptive MBAFF and subme 0 are not supported as motion vectors between
  126. + * field macroblocks and frame macroblocks require halving and hpel pixels.
  127. + * The chosen solution is to make MBAFF non-adaptive in this case. */
  128. + h->mb.b_adaptive_mbaff = !(h->param.b_interlaced && !h->param.analyse.i_subpel_refine);
  129. h->param.i_bframe = x264_clip3( h->param.i_bframe, 0, X264_MIN( X264_BFRAME_MAX, h->param.i_keyint_max-1 ) );
  130. h->param.i_open_gop = x264_clip3( h->param.i_open_gop, X264_OPEN_GOP_NONE, X264_OPEN_GOP_BLURAY );
  131. h->param.i_bframe_bias = x264_clip3( h->param.i_bframe_bias, -90, 100 );
  132. @@ -1958,9 +1962,12 @@ static int x264_slice_write( x264_t *h )
  133.  
  134. if( h->param.b_interlaced )
  135. {
  136. - if( !(i_mb_y&1) )
  137. - h->mb.b_interlaced = 1;
  138. - x264_zigzag_init( h->param.cpu, &h->zigzagf, h->mb.b_interlaced );
  139. + if( h->mb.b_adaptive_mbaff )
  140. + {
  141. + if( !(i_mb_y&1) )
  142. + h->mb.b_interlaced = 1;
  143. + x264_zigzag_init( h->param.cpu, &h->zigzagf, h->mb.b_interlaced );
  144. + }
  145. h->mb.field[mb_xy] = h->mb.b_interlaced;
  146. }
  147.  
  148. diff --git a/x264.h b/x264.h
  149. index 24c3792..da8746b 100644
  150. --- a/x264.h
  151. +++ b/x264.h
  152. @@ -291,6 +291,7 @@ typedef struct x264_param_t
  153. int i_cabac_init_idc;
  154.  
  155. int b_interlaced;
  156. + int b_adaptive_mbaff; /* MBAFF+subme 0 require non-adaptive MBAFF i.e. all field mbs */
  157. int b_constrained_intra;
  158.  
  159. int i_cqm_preset;
  160. --
  161. 1.7.4
  162.  
  163.  
  164. From eb50f5f3757d825b2664e7991b89c7647605dd28 Mon Sep 17 00:00:00 2001
  165. From: Simon Horlick <simonhorlick@gmail.com>
  166. Date: Tue, 11 Jan 2011 20:16:18 +0000
  167. Subject: [PATCH 03/25] Store left references in a table
  168.  
  169. ---
  170. common/common.h | 1 +
  171. common/macroblock.c | 56 ++++++++++++++++++++++++++++++++------------------
  172. 2 files changed, 37 insertions(+), 20 deletions(-)
  173.  
  174. diff --git a/common/common.h b/common/common.h
  175. index 75c4d59..992ba9c 100644
  176. --- a/common/common.h
  177. +++ b/common/common.h
  178. @@ -596,6 +596,7 @@ struct x264_t
  179. int i_mb_top_xy;
  180. int i_mb_topleft_xy;
  181. int i_mb_topright_xy;
  182. + int *left_index_table;
  183.  
  184. /**** thread synchronization ends here ****/
  185. /* subsequent variables are either thread-local or constant,
  186. diff --git a/common/macroblock.c b/common/macroblock.c
  187. index 569d544..d10b19f 100644
  188. --- a/common/macroblock.c
  189. +++ b/common/macroblock.c
  190. @@ -550,6 +550,18 @@ static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int mb_x
  191. }
  192. }
  193.  
  194. +static const int left_indices[5][22] = {
  195. +/* intra modes nnz mv ref real indices */
  196. + /* Current is progressive. */
  197. + { 4, 4, 5, 5, 3, 3, 7, 7, 16+1, 16+1, 16+4+1, 16+4+1, 0, 0, 1, 1, 0, 0, 0, 0, 1, 1 },
  198. + { 6, 6, 3, 3, 11, 11, 15, 15, 16+3, 16+3, 16+4+3, 16+4+3, 2, 2, 3, 3, 1, 1, 2, 2, 3, 3 },
  199. + /* Current is interlaced.*/
  200. + { 4, 6, 4, 6, 3, 11, 3, 11, 16+1, 16+1, 16+4+1, 16+4+1, 0, 2, 0, 2, 0, 0, 0, 2, 0, 2 },
  201. + { 4, 6, 4, 6, 3, 11, 3, 11, 16+1, 16+1, 16+4+1, 16+4+1, 0, 2, 0, 2, 0, 0, 0, 2, 0, 2 },
  202. + /*Both same.*/
  203. + { 4, 5, 6, 3, 3, 7, 11, 15, 16+1, 16+3, 16+4+1, 16+4+3, 0, 1, 2, 3, 0, 1, 0, 1, 2, 3 },
  204. +};
  205. +
  206. static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, int mb_y )
  207. {
  208. int top = (mb_y - (1 << h->mb.b_interlaced)) * h->mb.i_mb_stride + mb_x;
  209. @@ -570,6 +582,7 @@ static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, i
  210. h->mb.i_mb_type_left = -1;
  211. h->mb.i_mb_type_topleft = -1;
  212. h->mb.i_mb_type_topright = -1;
  213. + h->mb.left_index_table = left_indices[4];
  214.  
  215. if( mb_x > 0 )
  216. {
  217. @@ -661,6 +674,8 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  218. uint8_t (*nnz)[24] = h->mb.non_zero_count;
  219. int16_t *cbp = h->mb.cbp;
  220.  
  221. + int *left_index_table = h->mb.left_index_table;
  222. +
  223. /* load cache */
  224. if( h->mb.i_neighbour & MB_TOP )
  225. {
  226. @@ -703,22 +718,22 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  227. h->mb.cache.i_cbp_left = cbp[left];
  228.  
  229. /* load intra4x4 */
  230. - h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] = i4x4[left][4];
  231. - h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] = i4x4[left][5];
  232. - h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] = i4x4[left][6];
  233. - h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = i4x4[left][3];
  234. + h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] = i4x4[left][left_index_table[0]];
  235. + h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] = i4x4[left][left_index_table[1]];
  236. + h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] = i4x4[left][left_index_table[2]];
  237. + h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = i4x4[left][left_index_table[3]];
  238.  
  239. /* load non_zero_count */
  240. - h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][3];
  241. - h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][7];
  242. - h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][11];
  243. - h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left][15];
  244. + h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][left_index_table[4+0]];
  245. + h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][left_index_table[4+1]];
  246. + h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][left_index_table[4+2]];
  247. + h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left][left_index_table[4+3]];
  248.  
  249. - h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] = nnz[left][16+1];
  250. - h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] = nnz[left][16+3];
  251. + h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] = nnz[left][left_index_table[4+4]];
  252. + h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] = nnz[left][left_index_table[4+5]];
  253.  
  254. - h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] = nnz[left][16+4+1];
  255. - h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = nnz[left][16+4+3];
  256. + h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] = nnz[left][left_index_table[4+6]];
  257. + h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = nnz[left][left_index_table[4+7]];
  258. }
  259. else
  260. {
  261. @@ -857,10 +872,10 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  262.  
  263. if( h->mb.i_neighbour & MB_LEFT )
  264. {
  265. - CP16( h->mb.cache.mvd[l][x264_scan8[0 ] - 1], mvd[left][4] );
  266. - CP16( h->mb.cache.mvd[l][x264_scan8[2 ] - 1], mvd[left][5] );
  267. - CP16( h->mb.cache.mvd[l][x264_scan8[8 ] - 1], mvd[left][6] );
  268. - CP16( h->mb.cache.mvd[l][x264_scan8[10] - 1], mvd[left][3] );
  269. + CP16( h->mb.cache.mvd[l][x264_scan8[0 ] - 1], mvd[left][left_index_table[0]] );
  270. + CP16( h->mb.cache.mvd[l][x264_scan8[2 ] - 1], mvd[left][left_index_table[1]] );
  271. + CP16( h->mb.cache.mvd[l][x264_scan8[8 ] - 1], mvd[left][left_index_table[2]] );
  272. + CP16( h->mb.cache.mvd[l][x264_scan8[10] - 1], mvd[left][left_index_table[3]] );
  273. }
  274. else
  275. for( int i = 0; i < 4; i++ )
  276. @@ -949,6 +964,7 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  277. int s4x4 = h->mb.i_b4_stride;
  278.  
  279. uint8_t (*nnz)[24] = h->mb.non_zero_count;
  280. + int *left_index_table = h->mb.left_index_table;
  281.  
  282. if( h->mb.i_neighbour & MB_TOP )
  283. CP32( &h->mb.cache.non_zero_count[x264_scan8[0] - 8], &nnz[h->mb.i_mb_top_xy][12] );
  284. @@ -956,10 +972,10 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  285. if( h->mb.i_neighbour & MB_LEFT )
  286. {
  287. int left = h->mb.i_mb_left_xy;
  288. - h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][3];
  289. - h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][7];
  290. - h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][11];
  291. - h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left][15];
  292. + h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][left_index_table[4+0]];
  293. + h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][left_index_table[4+1]];
  294. + h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][left_index_table[4+2]];
  295. + h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left][left_index_table[4+3]];
  296. }
  297.  
  298. for( int l = 0; l <= (h->sh.i_type == SLICE_TYPE_B); l++ )
  299. --
  300. 1.7.4
  301.  
  302.  
  303. From c83e9ad2f2cb3a7f5d2ae6c4fd4b5a8cc04f894a Mon Sep 17 00:00:00 2001
  304. From: Simon Horlick <simonhorlick@gmail.com>
  305. Date: Tue, 11 Jan 2011 20:21:26 +0000
  306. Subject: [PATCH 04/25] Store references to the two left macroblocks
  307.  
  308. Fix compiler warnings about discarding const qualifiers
  309. ---
  310. common/common.h | 6 +++---
  311. common/deblock.c | 4 ++--
  312. common/macroblock.c | 24 ++++++++++++------------
  313. common/mvpred.c | 2 +-
  314. encoder/analyse.c | 8 ++++----
  315. encoder/cabac.c | 8 ++++----
  316. 6 files changed, 26 insertions(+), 26 deletions(-)
  317.  
  318. diff --git a/common/common.h b/common/common.h
  319. index 992ba9c..f839e7e 100644
  320. --- a/common/common.h
  321. +++ b/common/common.h
  322. @@ -588,15 +588,15 @@ struct x264_t
  323. unsigned int i_neighbour_intra; /* for constrained intra pred */
  324. unsigned int i_neighbour_frame; /* ignoring slice boundaries */
  325. int i_mb_type_top;
  326. - int i_mb_type_left;
  327. + int i_mb_type_left[2];
  328. int i_mb_type_topleft;
  329. int i_mb_type_topright;
  330. int i_mb_prev_xy;
  331. - int i_mb_left_xy;
  332. + int i_mb_left_xy[2];
  333. int i_mb_top_xy;
  334. int i_mb_topleft_xy;
  335. int i_mb_topright_xy;
  336. - int *left_index_table;
  337. + const int *left_index_table;
  338.  
  339. /**** thread synchronization ends here ****/
  340. /* subsequent variables are either thread-local or constant,
  341. diff --git a/common/deblock.c b/common/deblock.c
  342. index 1b6448f..0800461 100644
  343. --- a/common/deblock.c
  344. +++ b/common/deblock.c
  345. @@ -347,10 +347,10 @@ void x264_frame_deblock_row( x264_t *h, int mb_y )
  346.  
  347. if( h->mb.i_neighbour & MB_LEFT )
  348. {
  349. - int qpl = h->mb.qp[h->mb.i_mb_left_xy];
  350. + int qpl = h->mb.qp[h->mb.i_mb_left_xy[0]];
  351. int qp_left = (qp + qpl + 1) >> 1;
  352. int qpc_left = (h->chroma_qp_table[qp] + h->chroma_qp_table[qpl] + 1) >> 1;
  353. - int intra_left = IS_INTRA( h->mb.type[h->mb.i_mb_left_xy] );
  354. + int intra_left = IS_INTRA( h->mb.type[h->mb.i_mb_left_xy[0]] );
  355. if( intra_cur || intra_left )
  356. FILTER( _intra, 0, 0, qp_left, qpc_left );
  357. else
  358. diff --git a/common/macroblock.c b/common/macroblock.c
  359. index d10b19f..aa194a7 100644
  360. --- a/common/macroblock.c
  361. +++ b/common/macroblock.c
  362. @@ -575,11 +575,11 @@ static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, i
  363. h->mb.i_neighbour_intra = 0;
  364. h->mb.i_neighbour_frame = 0;
  365. h->mb.i_mb_top_xy = -1;
  366. - h->mb.i_mb_left_xy = -1;
  367. + h->mb.i_mb_left_xy[0] = h->mb.i_mb_left_xy[1] = -1;
  368. h->mb.i_mb_topleft_xy = -1;
  369. h->mb.i_mb_topright_xy = -1;
  370. h->mb.i_mb_type_top = -1;
  371. - h->mb.i_mb_type_left = -1;
  372. + h->mb.i_mb_type_left[0] = h->mb.i_mb_type_left[1] = -1;
  373. h->mb.i_mb_type_topleft = -1;
  374. h->mb.i_mb_type_topright = -1;
  375. h->mb.left_index_table = left_indices[4];
  376. @@ -587,13 +587,13 @@ static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, i
  377. if( mb_x > 0 )
  378. {
  379. h->mb.i_neighbour_frame |= MB_LEFT;
  380. - h->mb.i_mb_left_xy = h->mb.i_mb_xy - 1;
  381. - h->mb.i_mb_type_left = h->mb.type[h->mb.i_mb_left_xy];
  382. + h->mb.i_mb_left_xy[0] = h->mb.i_mb_xy - 1;
  383. + h->mb.i_mb_type_left[0] = h->mb.type[h->mb.i_mb_left_xy[0]];
  384. if( h->mb.i_mb_xy > h->sh.i_first_mb )
  385. {
  386. h->mb.i_neighbour |= MB_LEFT;
  387.  
  388. - if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_left ) )
  389. + if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_left[0] ) )
  390. h->mb.i_neighbour_intra |= MB_LEFT;
  391. }
  392. }
  393. @@ -659,7 +659,7 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  394. {
  395. x264_macroblock_cache_load_neighbours( h, mb_x, mb_y );
  396.  
  397. - int left = h->mb.i_mb_left_xy;
  398. + int left = h->mb.i_mb_left_xy[0];
  399. int top = h->mb.i_mb_top_xy;
  400. int top_y = mb_y - (1 << h->mb.b_interlaced);
  401. int s8x8 = h->mb.i_b8_stride;
  402. @@ -674,7 +674,7 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  403. uint8_t (*nnz)[24] = h->mb.non_zero_count;
  404. int16_t *cbp = h->mb.cbp;
  405.  
  406. - int *left_index_table = h->mb.left_index_table;
  407. + const int *left_index_table = h->mb.left_index_table;
  408.  
  409. /* load cache */
  410. if( h->mb.i_neighbour & MB_TOP )
  411. @@ -927,8 +927,8 @@ void x264_macroblock_cache_load_neighbours_deblock( x264_t *h, int mb_x, int mb_
  412.  
  413. if( mb_x > 0 )
  414. {
  415. - h->mb.i_mb_left_xy = h->mb.i_mb_xy - 1;
  416. - if( deblock_on_slice_edges || h->mb.slice_table[h->mb.i_mb_left_xy] == h->mb.slice_table[h->mb.i_mb_xy] )
  417. + h->mb.i_mb_left_xy[0] = h->mb.i_mb_xy - 1;
  418. + if( deblock_on_slice_edges || h->mb.slice_table[h->mb.i_mb_left_xy[0]] == h->mb.slice_table[h->mb.i_mb_xy] )
  419. h->mb.i_neighbour |= MB_LEFT;
  420. }
  421.  
  422. @@ -964,14 +964,14 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  423. int s4x4 = h->mb.i_b4_stride;
  424.  
  425. uint8_t (*nnz)[24] = h->mb.non_zero_count;
  426. - int *left_index_table = h->mb.left_index_table;
  427. + const int *left_index_table = h->mb.left_index_table;
  428.  
  429. if( h->mb.i_neighbour & MB_TOP )
  430. CP32( &h->mb.cache.non_zero_count[x264_scan8[0] - 8], &nnz[h->mb.i_mb_top_xy][12] );
  431.  
  432. if( h->mb.i_neighbour & MB_LEFT )
  433. {
  434. - int left = h->mb.i_mb_left_xy;
  435. + int left = h->mb.i_mb_left_xy[0];
  436. h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][left_index_table[4+0]];
  437. h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][left_index_table[4+1]];
  438. h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][left_index_table[4+2]];
  439. @@ -1046,7 +1046,7 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  440. {
  441. uint8_t (*nnz)[24] = h->mb.non_zero_count;
  442. int top = h->mb.i_mb_top_xy;
  443. - int left = h->mb.i_mb_left_xy;
  444. + int left = h->mb.i_mb_left_xy[0];
  445.  
  446. if( (h->mb.i_neighbour & MB_TOP) && h->mb.mb_transform_size[top] )
  447. {
  448. diff --git a/common/mvpred.c b/common/mvpred.c
  449. index a24dde8..c8efe1f 100644
  450. --- a/common/mvpred.c
  451. +++ b/common/mvpred.c
  452. @@ -426,7 +426,7 @@ void x264_mb_predict_mv_ref16x16( x264_t *h, int i_list, int i_ref, int16_t mvc[
  453. }
  454.  
  455. /* spatial predictors */
  456. - SET_MVP( mvr[h->mb.i_mb_left_xy] );
  457. + SET_MVP( mvr[h->mb.i_mb_left_xy[0]] );
  458. SET_MVP( mvr[h->mb.i_mb_top_xy] );
  459. SET_MVP( mvr[h->mb.i_mb_topleft_xy] );
  460. SET_MVP( mvr[h->mb.i_mb_topright_xy] );
  461. diff --git a/encoder/analyse.c b/encoder/analyse.c
  462. index 5419bd1..87125c1 100644
  463. --- a/encoder/analyse.c
  464. +++ b/encoder/analyse.c
  465. @@ -516,7 +516,7 @@ static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int qp )
  466. {
  467. /* Always run in fast-intra mode for subme < 3 */
  468. if( h->mb.i_subpel_refine > 2 &&
  469. - ( IS_INTRA( h->mb.i_mb_type_left ) ||
  470. + ( IS_INTRA( h->mb.i_mb_type_left[0] ) ||
  471. IS_INTRA( h->mb.i_mb_type_top ) ||
  472. IS_INTRA( h->mb.i_mb_type_topleft ) ||
  473. IS_INTRA( h->mb.i_mb_type_topright ) ||
  474. @@ -1296,7 +1296,7 @@ static void x264_mb_analyse_inter_p8x8_mixed_ref( x264_t *h, x264_mb_analysis_t
  475. /* early termination: if 16x16 chose ref 0, then evalute no refs older
  476. * than those used by the neighbors */
  477. if( i_maxref > 0 && (a->l0.me16x16.i_ref == 0 || a->l0.me16x16.i_ref == h->mb.ref_blind_dupe) &&
  478. - h->mb.i_mb_type_top > 0 && h->mb.i_mb_type_left > 0 )
  479. + h->mb.i_mb_type_top > 0 && h->mb.i_mb_type_left[0] > 0 )
  480. {
  481. i_maxref = 0;
  482. CHECK_NEIGHBOUR( -8 - 1 );
  483. @@ -2063,7 +2063,7 @@ static void x264_mb_analyse_inter_b8x8_mixed_ref( x264_t *h, x264_mb_analysis_t
  484. {
  485. x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
  486. if( i_maxref[l] > 0 && lX->me16x16.i_ref == 0 &&
  487. - h->mb.i_mb_type_top > 0 && h->mb.i_mb_type_left > 0 )
  488. + h->mb.i_mb_type_top > 0 && h->mb.i_mb_type_left[0] > 0 )
  489. {
  490. i_maxref[l] = 0;
  491. CHECK_NEIGHBOUR( -8 - 1 );
  492. @@ -2817,7 +2817,7 @@ intra_analysis:
  493. {}
  494. else if( h->param.analyse.i_subpel_refine >= 3 )
  495. analysis.b_try_skip = 1;
  496. - else if( h->mb.i_mb_type_left == P_SKIP ||
  497. + else if( h->mb.i_mb_type_left[0] == P_SKIP ||
  498. h->mb.i_mb_type_top == P_SKIP ||
  499. h->mb.i_mb_type_topleft == P_SKIP ||
  500. h->mb.i_mb_type_topright == P_SKIP )
  501. diff --git a/encoder/cabac.c b/encoder/cabac.c
  502. index 6333737..334318d 100644
  503. --- a/encoder/cabac.c
  504. +++ b/encoder/cabac.c
  505. @@ -79,7 +79,7 @@ static void x264_cabac_mb_type( x264_t *h, x264_cabac_t *cb )
  506. if( h->sh.i_type == SLICE_TYPE_I )
  507. {
  508. int ctx = 0;
  509. - if( (h->mb.i_neighbour & MB_LEFT) && h->mb.i_mb_type_left != I_4x4 )
  510. + if( (h->mb.i_neighbour & MB_LEFT) && h->mb.i_mb_type_left[0] != I_4x4 )
  511. ctx++;
  512. if( (h->mb.i_neighbour & MB_TOP) && h->mb.i_mb_type_top != I_4x4 )
  513. ctx++;
  514. @@ -113,7 +113,7 @@ static void x264_cabac_mb_type( x264_t *h, x264_cabac_t *cb )
  515. else //if( h->sh.i_type == SLICE_TYPE_B )
  516. {
  517. int ctx = 0;
  518. - if( (h->mb.i_neighbour & MB_LEFT) && h->mb.i_mb_type_left != B_SKIP && h->mb.i_mb_type_left != B_DIRECT )
  519. + if( (h->mb.i_neighbour & MB_LEFT) && h->mb.i_mb_type_left[0] != B_SKIP && h->mb.i_mb_type_left[0] != B_DIRECT )
  520. ctx++;
  521. if( (h->mb.i_neighbour & MB_TOP) && h->mb.i_mb_type_top != B_SKIP && h->mb.i_mb_type_top != B_DIRECT )
  522. ctx++;
  523. @@ -198,7 +198,7 @@ static void x264_cabac_mb_intra_chroma_pred_mode( x264_t *h, x264_cabac_t *cb )
  524. int ctx = 0;
  525.  
  526. /* No need to test for I4x4 or I_16x16 as cache_save handle that */
  527. - if( (h->mb.i_neighbour & MB_LEFT) && h->mb.chroma_pred_mode[h->mb.i_mb_left_xy] != 0 )
  528. + if( (h->mb.i_neighbour & MB_LEFT) && h->mb.chroma_pred_mode[h->mb.i_mb_left_xy[0]] != 0 )
  529. ctx++;
  530. if( (h->mb.i_neighbour & MB_TOP) && h->mb.chroma_pred_mode[h->mb.i_mb_top_xy] != 0 )
  531. ctx++;
  532. @@ -280,7 +280,7 @@ static void x264_cabac_mb_qp_delta( x264_t *h, x264_cabac_t *cb )
  533. #if !RDO_SKIP_BS
  534. void x264_cabac_mb_skip( x264_t *h, int b_skip )
  535. {
  536. - int ctx = ((h->mb.i_neighbour & MB_LEFT) && !IS_SKIP( h->mb.i_mb_type_left ))
  537. + int ctx = ((h->mb.i_neighbour & MB_LEFT) && !IS_SKIP( h->mb.i_mb_type_left[0] ))
  538. + ((h->mb.i_neighbour & MB_TOP) && !IS_SKIP( h->mb.i_mb_type_top ))
  539. + (h->sh.i_type == SLICE_TYPE_P ? 11 : 24);
  540. x264_cabac_encode_decision( &h->cabac, ctx, b_skip );
  541. --
  542. 1.7.4
  543.  
  544.  
  545. From 135b93d39e3d8d8540a41bd66d90aa42f7a73ba4 Mon Sep 17 00:00:00 2001
  546. From: Simon Horlick <simonhorlick@gmail.com>
  547. Date: Fri, 14 Jan 2011 21:18:14 +0000
  548. Subject: [PATCH 05/25] Neighbour calculation for mbaff
  549.  
  550. Back up intra borders correctly and make neighbour calculation several times longer.
  551. ---
  552. common/common.h | 9 ++-
  553. common/macroblock.c | 286 +++++++++++++++++++++++++++++++++++++++++++--------
  554. 2 files changed, 249 insertions(+), 46 deletions(-)
  555.  
  556. diff --git a/common/common.h b/common/common.h
  557. index f839e7e..c993857 100644
  558. --- a/common/common.h
  559. +++ b/common/common.h
  560. @@ -549,6 +549,8 @@ struct x264_t
  561. int i_mb_stride;
  562. int i_b8_stride;
  563. int i_b4_stride;
  564. + int left_b8[2];
  565. + int left_b4[2];
  566.  
  567. /* Current index */
  568. int i_mb_x;
  569. @@ -597,6 +599,10 @@ struct x264_t
  570. int i_mb_topleft_xy;
  571. int i_mb_topright_xy;
  572. const int *left_index_table;
  573. + int topleft_partition;
  574. + int intra_border_index;
  575. + int topleft_border_index;
  576. + int topright_border_index;
  577.  
  578. /**** thread synchronization ends here ****/
  579. /* subsequent variables are either thread-local or constant,
  580. @@ -827,7 +833,8 @@ struct x264_t
  581.  
  582. /* Buffers that are allocated per-thread even in sliced threads. */
  583. void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */
  584. - pixel *intra_border_backup[2][2]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
  585. + pixel *intra_border_backup[3][2]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
  586. + pixel *intra_diagonal_backup[5][3];
  587. uint8_t (*deblock_strength[2])[2][4][4];
  588.  
  589. /* CPU functions dependents */
  590. diff --git a/common/macroblock.c b/common/macroblock.c
  591. index aa194a7..60275ae 100644
  592. --- a/common/macroblock.c
  593. +++ b/common/macroblock.c
  594. @@ -314,18 +314,35 @@ void x264_macroblock_cache_free( x264_t *h )
  595. int x264_macroblock_thread_allocate( x264_t *h, int b_lookahead )
  596. {
  597. if( !b_lookahead )
  598. - for( int i = 0; i <= h->param.b_interlaced; i++ )
  599. + {
  600. + for( int i = 0; i <= 2*h->param.b_interlaced; i++ )
  601. {
  602. for( int j = 0; j < 2; j++ )
  603. {
  604. /* shouldn't really be initialized, just silences a valgrind false-positive in predict_8x8_filter_mmx */
  605. CHECKED_MALLOCZERO( h->intra_border_backup[i][j], (h->sps->i_mb_width*16+32) * sizeof(pixel) );
  606. h->intra_border_backup[i][j] += 16;
  607. - h->intra_border_backup[1][j] = h->intra_border_backup[i][j];
  608. + if( !h->param.b_interlaced )
  609. + h->intra_border_backup[1][j] = h->intra_border_backup[i][j];
  610. + }
  611. + }
  612. + for( int i = 0; i < 4*h->mb.b_interlaced; i++ )
  613. + {
  614. + for( int j = 0; j < 3; j++ )
  615. + {
  616. + const int width = 1 + 8; // top left pixel + eight top right pixels (for luma)
  617. + CHECKED_MALLOCZERO( h->intra_diagonal_backup[i][j], (h->sps->i_mb_width*width+32) * sizeof(pixel) );
  618. + h->intra_diagonal_backup[i][j] += 16;
  619. + if( !h->param.b_interlaced )
  620. + h->intra_diagonal_backup[1][j] = h->intra_diagonal_backup[i][j];
  621. }
  622. + }
  623. + for( int i = 0; i <= h->param.b_interlaced; i++ )
  624. + {
  625. CHECKED_MALLOC( h->deblock_strength[i], sizeof(**h->deblock_strength) * h->mb.i_mb_width );
  626. h->deblock_strength[1] = h->deblock_strength[i];
  627. }
  628. + }
  629.  
  630. /* Allocate scratch buffer */
  631. int scratch_size = 0;
  632. @@ -353,12 +370,20 @@ fail:
  633. void x264_macroblock_thread_free( x264_t *h, int b_lookahead )
  634. {
  635. if( !b_lookahead )
  636. + {
  637. for( int i = 0; i <= h->param.b_interlaced; i++ )
  638. - {
  639. x264_free( h->deblock_strength[i] );
  640. + for( int i = 0; i <= 2*h->param.b_interlaced; i++ )
  641. + {
  642. for( int j = 0; j < 2; j++ )
  643. x264_free( h->intra_border_backup[i][j] - 16 );
  644. }
  645. + for( int i = 0; i < 4*h->param.b_interlaced; i++ )
  646. + {
  647. + for( int j = 0; j < 3; j++ )
  648. + x264_free( h->intra_diagonal_backup[i][j] - 16 );
  649. + }
  650. + }
  651. x264_free( h->scratch_buffer );
  652. }
  653.  
  654. @@ -494,14 +519,15 @@ static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int mb_x
  655. {
  656. int w = (i ? 8 : 16);
  657. int i_stride = h->fdec->i_stride[i];
  658. - int i_stride2 = i_stride << b_interlaced;
  659. - int i_pix_offset = b_interlaced
  660. + int i_stride2 = i_stride << h->mb.b_interlaced;
  661. + int i_pix_offset = h->mb.b_interlaced
  662. ? 16 * mb_x + w * (mb_y&~1) * i_stride + (mb_y&1) * i_stride
  663. : 16 * mb_x + w * mb_y * i_stride;
  664. pixel *plane_fdec = &h->fdec->plane[i][i_pix_offset];
  665. - pixel *intra_fdec = &h->intra_border_backup[mb_y&1][i][mb_x*16];
  666. + pixel *intra_fdec = &h->intra_border_backup[h->mb.intra_border_index][i][mb_x*16];
  667. int ref_pix_offset[2] = { i_pix_offset, i_pix_offset };
  668. - if( b_interlaced )
  669. + /* ref_pix_offset[0] references the current field and [1] the opposite field. */
  670. + if( h->mb.b_interlaced )
  671. ref_pix_offset[1] += (1-2*(mb_y&1)) * i_stride;
  672. h->mb.pic.i_stride[i] = i_stride2;
  673. h->mb.pic.p_fenc_plane[i] = &h->fenc->plane[i][i_pix_offset];
  674. @@ -510,11 +536,28 @@ static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int mb_x
  675. h->mc.load_deinterleave_8x8x2_fenc( h->mb.pic.p_fenc[1], h->mb.pic.p_fenc_plane[1], i_stride2 );
  676. memcpy( h->mb.pic.p_fdec[1]-FDEC_STRIDE, intra_fdec, 8*sizeof(pixel) );
  677. memcpy( h->mb.pic.p_fdec[2]-FDEC_STRIDE, intra_fdec+8, 8*sizeof(pixel) );
  678. + if( h->sh.b_mbaff )
  679. + {
  680. + // Top left samples.
  681. + h->mb.pic.p_fdec[1][-FDEC_STRIDE-1] = h->intra_diagonal_backup[h->mb.topleft_border_index][1][mb_x*9];
  682. + h->mb.pic.p_fdec[2][-FDEC_STRIDE-1] = h->intra_diagonal_backup[h->mb.topleft_border_index][2][mb_x*9];
  683. + // Top right samples.
  684. + CP32( &h->mb.pic.p_fdec[1][-FDEC_STRIDE+8], &h->intra_diagonal_backup[h->mb.topright_border_index][1][mb_x*9+1] );
  685. + CP32( &h->mb.pic.p_fdec[2][-FDEC_STRIDE+8], &h->intra_diagonal_backup[h->mb.topright_border_index][2][mb_x*9+1] );
  686. + }
  687. +
  688. }
  689. else
  690. {
  691. h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fenc[0], FENC_STRIDE, h->mb.pic.p_fenc_plane[0], i_stride2, 16 );
  692. - memcpy( h->mb.pic.p_fdec[0]-FDEC_STRIDE, intra_fdec, 24*sizeof(pixel) );
  693. + if( h->sh.b_mbaff )
  694. + {
  695. + memcpy( h->mb.pic.p_fdec[0]-FDEC_STRIDE, intra_fdec, 16*sizeof(pixel) );
  696. + h->mb.pic.p_fdec[0][-FDEC_STRIDE-1] = h->intra_diagonal_backup[h->mb.topleft_border_index][0][mb_x*9];
  697. + CP64( &h->mb.pic.p_fdec[0][-FDEC_STRIDE+16], &h->intra_diagonal_backup[h->mb.topright_border_index][0][mb_x*9+1] );
  698. + }
  699. + else
  700. + memcpy( h->mb.pic.p_fdec[0]-FDEC_STRIDE, intra_fdec, 24*sizeof(pixel) );
  701. }
  702. if( b_interlaced )
  703. {
  704. @@ -571,6 +614,10 @@ static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, i
  705. h->mb.i_mb_xy = mb_y * h->mb.i_mb_stride + mb_x;
  706. h->mb.i_b8_xy = 2*(mb_y * h->mb.i_b8_stride + mb_x);
  707. h->mb.i_b4_xy = 4*(mb_y * h->mb.i_b4_stride + mb_x);
  708. + h->mb.left_b8[0] =
  709. + h->mb.left_b8[1] = -1;
  710. + h->mb.left_b4[0] =
  711. + h->mb.left_b4[1] = -1;
  712. h->mb.i_neighbour = 0;
  713. h->mb.i_neighbour_intra = 0;
  714. h->mb.i_neighbour_frame = 0;
  715. @@ -583,16 +630,105 @@ static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, i
  716. h->mb.i_mb_type_topleft = -1;
  717. h->mb.i_mb_type_topright = -1;
  718. h->mb.left_index_table = left_indices[4];
  719. + h->mb.topleft_partition = 0;
  720. + h->mb.topright_border_index =
  721. + h->mb.topleft_border_index = !(mb_y&1);
  722. + h->mb.intra_border_index = mb_y&1;
  723. +
  724. + int topleft = top - 1;
  725. + int topright = top + 1;
  726. + int left[2];
  727. +
  728. + left[0] = left[1] = h->mb.i_mb_xy - 1;
  729. + h->mb.left_b8[0] = h->mb.left_b8[1] = h->mb.i_b8_xy - 2;
  730. + h->mb.left_b4[0] = h->mb.left_b4[1] = h->mb.i_b4_xy - 4;
  731. +
  732. + if( h->sh.b_mbaff )
  733. + {
  734. + if( mb_y&1 )
  735. + {
  736. + if( mb_x && h->mb.b_interlaced != h->mb.field[h->mb.i_mb_xy-1] )
  737. + {
  738. + left[0] = left[1] = h->mb.i_mb_xy - 1 - h->mb.i_mb_stride;
  739. + h->mb.left_b8[0] = h->mb.left_b8[1] = h->mb.i_b8_xy - 2 - 2*h->mb.i_b8_stride;
  740. + h->mb.left_b4[0] = h->mb.left_b4[1] = h->mb.i_b4_xy - 4 - 4*h->mb.i_b4_stride;
  741. +
  742. + if( h->mb.b_interlaced )
  743. + {
  744. + h->mb.left_index_table = left_indices[3];
  745. + left[1] += h->mb.i_mb_stride;
  746. + h->mb.left_b8[1] += 2*h->mb.i_b8_stride;
  747. + h->mb.left_b4[1] += 4*h->mb.i_b4_stride;
  748. + }
  749. + else
  750. + {
  751. + h->mb.left_index_table = left_indices[1];
  752. + topleft += h->mb.i_mb_stride;
  753. + h->mb.topleft_partition = 1;
  754. + h->mb.topleft_border_index = 3;
  755. + }
  756. + }
  757. + if( h->mb.b_interlaced )
  758. + h->mb.topleft_border_index = 1;
  759. + else
  760. + {
  761. + topright = -1;
  762. + h->mb.intra_border_index = 0;
  763. + }
  764. + h->mb.topright_border_index = 1;
  765. + }
  766. + else
  767. + {
  768. + if( h->mb.b_interlaced && top >= 0 )
  769. + {
  770. + if( !h->mb.field[top] )
  771. + {
  772. + top += h->mb.i_mb_stride;
  773. + h->mb.intra_border_index = 2;
  774. + }
  775. + if( mb_x )
  776. + topleft += h->mb.i_mb_stride*(!h->mb.field[topleft]);
  777. + if( mb_x < h->mb.i_mb_width-1 )
  778. + topright += h->mb.i_mb_stride*(!h->mb.field[topright]);
  779. +
  780. + if( topright >=0 && h->mb.field[topright] )
  781. + h->mb.topright_border_index = 0;
  782. + else
  783. + h->mb.topright_border_index = 2;
  784. + if( topleft >=0 && h->mb.field[topleft] )
  785. + h->mb.topleft_border_index = 0;
  786. + else
  787. + h->mb.topleft_border_index = 2;
  788. + }
  789. + else
  790. + h->mb.intra_border_index = 1;
  791. + if( mb_x && h->mb.b_interlaced != h->mb.field[h->mb.i_mb_xy-1] )
  792. + {
  793. + if( h->mb.b_interlaced )
  794. + {
  795. + h->mb.left_index_table = left_indices[2];
  796. + left[1] += h->mb.i_mb_stride;
  797. + h->mb.left_b8[1] += 2*h->mb.i_b8_stride;
  798. + h->mb.left_b4[1] += 4*h->mb.i_b4_stride;
  799. + }
  800. + else
  801. + h->mb.left_index_table = left_indices[0];
  802. + }
  803. + }
  804. + }
  805.  
  806. if( mb_x > 0 )
  807. {
  808. h->mb.i_neighbour_frame |= MB_LEFT;
  809. - h->mb.i_mb_left_xy[0] = h->mb.i_mb_xy - 1;
  810. + h->mb.i_mb_left_xy[0] = left[0];
  811. + h->mb.i_mb_left_xy[1] = left[1];
  812. h->mb.i_mb_type_left[0] = h->mb.type[h->mb.i_mb_left_xy[0]];
  813. + h->mb.i_mb_type_left[1] = h->mb.type[h->mb.i_mb_left_xy[1]];
  814. if( h->mb.i_mb_xy > h->sh.i_first_mb )
  815. {
  816. h->mb.i_neighbour |= MB_LEFT;
  817.  
  818. + // FIXME: We don't currently support constrained intra + mbaff.
  819. if( !h->param.b_constrained_intra || IS_INTRA( h->mb.i_mb_type_left[0] ) )
  820. h->mb.i_neighbour_intra |= MB_LEFT;
  821. }
  822. @@ -625,12 +761,12 @@ static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, i
  823. }
  824. }
  825.  
  826. - if( mb_x > 0 && top - 1 >= 0 )
  827. + if( mb_x > 0 && topleft >= 0 )
  828. {
  829. h->mb.i_neighbour_frame |= MB_TOPLEFT;
  830. - h->mb.i_mb_topleft_xy = top - 1;
  831. + h->mb.i_mb_topleft_xy = topleft;
  832. h->mb.i_mb_type_topleft = h->mb.type[h->mb.i_mb_topleft_xy];
  833. - if( top - 1 >= h->sh.i_first_mb )
  834. + if( topleft >= h->sh.i_first_mb )
  835. {
  836. h->mb.i_neighbour |= MB_TOPLEFT;
  837.  
  838. @@ -639,12 +775,12 @@ static void inline x264_macroblock_cache_load_neighbours( x264_t *h, int mb_x, i
  839. }
  840. }
  841.  
  842. - if( mb_x < h->mb.i_mb_width - 1 && top + 1 >= 0 )
  843. + if( mb_x < h->mb.i_mb_width - 1 && topright >= 0 )
  844. {
  845. h->mb.i_neighbour_frame |= MB_TOPRIGHT;
  846. - h->mb.i_mb_topright_xy = top + 1;
  847. + h->mb.i_mb_topright_xy = topright;
  848. h->mb.i_mb_type_topright = h->mb.type[h->mb.i_mb_topright_xy];
  849. - if( top + 1 >= h->sh.i_first_mb )
  850. + if( topright >= h->sh.i_first_mb )
  851. {
  852. h->mb.i_neighbour |= MB_TOPRIGHT;
  853.  
  854. @@ -659,9 +795,9 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  855. {
  856. x264_macroblock_cache_load_neighbours( h, mb_x, mb_y );
  857.  
  858. - int left = h->mb.i_mb_left_xy[0];
  859. + int *left = h->mb.i_mb_left_xy;
  860. int top = h->mb.i_mb_top_xy;
  861. - int top_y = mb_y - (1 << h->mb.b_interlaced);
  862. + int top_y = top / h->mb.i_mb_stride;
  863. int s8x8 = h->mb.i_b8_stride;
  864. int s4x4 = h->mb.i_b4_stride;
  865. int top_8x8 = (2*top_y+1) * s8x8 + 2*mb_x;
  866. @@ -715,25 +851,25 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  867.  
  868. if( h->mb.i_neighbour & MB_LEFT )
  869. {
  870. - h->mb.cache.i_cbp_left = cbp[left];
  871. + h->mb.cache.i_cbp_left = cbp[left[0]];
  872.  
  873. /* load intra4x4 */
  874. - h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] = i4x4[left][left_index_table[0]];
  875. - h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] = i4x4[left][left_index_table[1]];
  876. - h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] = i4x4[left][left_index_table[2]];
  877. - h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = i4x4[left][left_index_table[3]];
  878. + h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] = i4x4[left[0]][left_index_table[0]];
  879. + h->mb.cache.intra4x4_pred_mode[x264_scan8[2 ] - 1] = i4x4[left[0]][left_index_table[1]];
  880. + h->mb.cache.intra4x4_pred_mode[x264_scan8[8 ] - 1] = i4x4[left[1]][left_index_table[2]];
  881. + h->mb.cache.intra4x4_pred_mode[x264_scan8[10] - 1] = i4x4[left[1]][left_index_table[3]];
  882.  
  883. /* load non_zero_count */
  884. - h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][left_index_table[4+0]];
  885. - h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][left_index_table[4+1]];
  886. - h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][left_index_table[4+2]];
  887. - h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left][left_index_table[4+3]];
  888. + h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left[0]][left_index_table[4+0]];
  889. + h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left[0]][left_index_table[4+1]];
  890. + h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left[1]][left_index_table[4+2]];
  891. + h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left[1]][left_index_table[4+3]];
  892.  
  893. - h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] = nnz[left][left_index_table[4+4]];
  894. - h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] = nnz[left][left_index_table[4+5]];
  895. + h->mb.cache.non_zero_count[x264_scan8[16+0] - 1] = nnz[left[0]][left_index_table[4+4]];
  896. + h->mb.cache.non_zero_count[x264_scan8[16+2] - 1] = nnz[left[1]][left_index_table[4+5]];
  897.  
  898. - h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] = nnz[left][left_index_table[4+6]];
  899. - h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = nnz[left][left_index_table[4+7]];
  900. + h->mb.cache.non_zero_count[x264_scan8[16+4+0] - 1] = nnz[left[0]][left_index_table[4+6]];
  901. + h->mb.cache.non_zero_count[x264_scan8[16+4+2] - 1] = nnz[left[1]][left_index_table[4+7]];
  902. }
  903. else
  904. {
  905. @@ -758,7 +894,7 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  906. if( h->pps->b_transform_8x8_mode )
  907. {
  908. h->mb.cache.i_neighbour_transform_size =
  909. - ( (h->mb.i_neighbour & MB_LEFT) && h->mb.mb_transform_size[left] )
  910. + ( (h->mb.i_neighbour & MB_LEFT) && h->mb.mb_transform_size[left[0]] )
  911. + ( (h->mb.i_neighbour & MB_TOP) && h->mb.mb_transform_size[top] );
  912. }
  913.  
  914. @@ -771,7 +907,7 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  915. + !!(h->mb.i_neighbour & MB_TOP);
  916. }
  917.  
  918. - if( !h->mb.b_interlaced )
  919. + if( !h->sh.b_mbaff )
  920. {
  921. x264_copy_column8( h->mb.pic.p_fdec[0]-1+ 4*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+ 4*FDEC_STRIDE );
  922. x264_copy_column8( h->mb.pic.p_fdec[0]-1+12*FDEC_STRIDE, h->mb.pic.p_fdec[0]+15+12*FDEC_STRIDE );
  923. @@ -872,10 +1008,10 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  924.  
  925. if( h->mb.i_neighbour & MB_LEFT )
  926. {
  927. - CP16( h->mb.cache.mvd[l][x264_scan8[0 ] - 1], mvd[left][left_index_table[0]] );
  928. - CP16( h->mb.cache.mvd[l][x264_scan8[2 ] - 1], mvd[left][left_index_table[1]] );
  929. - CP16( h->mb.cache.mvd[l][x264_scan8[8 ] - 1], mvd[left][left_index_table[2]] );
  930. - CP16( h->mb.cache.mvd[l][x264_scan8[10] - 1], mvd[left][left_index_table[3]] );
  931. + CP16( h->mb.cache.mvd[l][x264_scan8[0 ] - 1], mvd[left[0]][left_index_table[0]] );
  932. + CP16( h->mb.cache.mvd[l][x264_scan8[2 ] - 1], mvd[left[0]][left_index_table[1]] );
  933. + CP16( h->mb.cache.mvd[l][x264_scan8[8 ] - 1], mvd[left[1]][left_index_table[2]] );
  934. + CP16( h->mb.cache.mvd[l][x264_scan8[10] - 1], mvd[left[1]][left_index_table[3]] );
  935. }
  936. else
  937. for( int i = 0; i < 4; i++ )
  938. @@ -892,7 +1028,7 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  939. {
  940. uint8_t skipbp;
  941. x264_macroblock_cache_skip( h, 0, 0, 4, 4, 0 );
  942. - skipbp = (h->mb.i_neighbour & MB_LEFT) ? h->mb.skipbp[left] : 0;
  943. + skipbp = (h->mb.i_neighbour & MB_LEFT) ? h->mb.skipbp[left[0]] : 0;
  944. h->mb.cache.skip[x264_scan8[0] - 1] = skipbp & 0x2;
  945. h->mb.cache.skip[x264_scan8[8] - 1] = skipbp & 0x8;
  946. skipbp = (h->mb.i_neighbour & MB_TOP) ? h->mb.skipbp[top] : 0;
  947. @@ -971,11 +1107,11 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  948.  
  949. if( h->mb.i_neighbour & MB_LEFT )
  950. {
  951. - int left = h->mb.i_mb_left_xy[0];
  952. - h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left][left_index_table[4+0]];
  953. - h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left][left_index_table[4+1]];
  954. - h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left][left_index_table[4+2]];
  955. - h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left][left_index_table[4+3]];
  956. + int *left = h->mb.i_mb_left_xy;
  957. + h->mb.cache.non_zero_count[x264_scan8[0 ] - 1] = nnz[left[0]][left_index_table[4+0]];
  958. + h->mb.cache.non_zero_count[x264_scan8[2 ] - 1] = nnz[left[0]][left_index_table[4+1]];
  959. + h->mb.cache.non_zero_count[x264_scan8[8 ] - 1] = nnz[left[1]][left_index_table[4+2]];
  960. + h->mb.cache.non_zero_count[x264_scan8[10] - 1] = nnz[left[1]][left_index_table[4+3]];
  961. }
  962.  
  963. for( int l = 0; l <= (h->sh.i_type == SLICE_TYPE_B); l++ )
  964. @@ -1106,20 +1242,80 @@ static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb
  965. int i_pix_offset = b_interlaced
  966. ? 16 * mb_x + w * (mb_y&~1) * i_stride + (mb_y&1) * i_stride
  967. : 16 * mb_x + w * mb_y * i_stride;
  968. + const int intra_diag_width = 8+1; // One top left sample, then eight top right samples.
  969. pixel *intra_fdec = &h->intra_border_backup[mb_y&1][i][mb_x*16];
  970. if( i )
  971. {
  972. + if( h->sh.b_mbaff )
  973. + {
  974. + /* Frame macroblocks use the macroblock directly above for intra
  975. + * prediction. Field macroblock pairs predict from fields of the same
  976. + * parity. However field macroblock pairs predicting from frame pairs
  977. + * use the bottom two rows of the frame for prediction, the penultimate
  978. + * row is stored in intra_border_backup[2]. */
  979. + if( mb_y&1 )
  980. + {
  981. + if( mb_x )
  982. + {
  983. + // Store top left.
  984. + h->intra_diagonal_backup[1][1][mb_x*intra_diag_width] = h->intra_border_backup[1][1][(mb_x-1)*16+7];
  985. + h->intra_diagonal_backup[1][2][mb_x*intra_diag_width] = h->intra_border_backup[1][1][(mb_x-1)*16+8+7];
  986. + h->intra_diagonal_backup[2][1][mb_x*intra_diag_width] = h->intra_border_backup[2][1][(mb_x-1)*16+7];
  987. + h->intra_diagonal_backup[2][2][mb_x*intra_diag_width] = h->intra_border_backup[2][1][(mb_x-1)*16+8+7];
  988. + // Store top right.
  989. + CP32( &h->intra_diagonal_backup[1][1][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*7 );
  990. + CP32( &h->intra_diagonal_backup[1][2][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*7 );
  991. + CP32( &h->intra_diagonal_backup[2][1][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*6 );
  992. + CP32( &h->intra_diagonal_backup[2][2][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*6 );
  993. + }
  994. + memcpy( &h->intra_border_backup[2][i][mb_x*16], h->mb.pic.p_fdec[1]+FDEC_STRIDE*6, 8*sizeof(pixel) );
  995. + memcpy( &h->intra_border_backup[2][i][mb_x*16]+8, h->mb.pic.p_fdec[2]+FDEC_STRIDE*6, 8*sizeof(pixel) );
  996. + }
  997. + else
  998. + {
  999. + if( mb_x )
  1000. + {
  1001. + h->intra_diagonal_backup[0][1][mb_x*intra_diag_width] = h->intra_border_backup[0][1][(mb_x-1)*16+7];
  1002. + h->intra_diagonal_backup[0][2][mb_x*intra_diag_width] = h->intra_border_backup[0][1][(mb_x-1)*16+8+7];
  1003. + CP32( &h->intra_diagonal_backup[0][1][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[1]+FDEC_STRIDE*7 );
  1004. + CP32( &h->intra_diagonal_backup[0][2][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[2]+FDEC_STRIDE*7 );
  1005. + }
  1006. + // Sometimes needed for bottom macroblock of this pair.
  1007. + h->intra_diagonal_backup[3][1][mb_x*intra_diag_width] = h->mb.pic.p_fdec[1][-1+7*FDEC_STRIDE];
  1008. + h->intra_diagonal_backup[3][2][mb_x*intra_diag_width] = h->mb.pic.p_fdec[2][-1+7*FDEC_STRIDE];
  1009. + }
  1010. + }
  1011. h->mc.store_interleave_8x8x2( &h->fdec->plane[1][i_pix_offset], i_stride2, h->mb.pic.p_fdec[1], h->mb.pic.p_fdec[2] );
  1012. memcpy( intra_fdec, h->mb.pic.p_fdec[1]+FDEC_STRIDE*7, 8*sizeof(pixel) );
  1013. memcpy( intra_fdec+8, h->mb.pic.p_fdec[2]+FDEC_STRIDE*7, 8*sizeof(pixel) );
  1014. - twiddle_topleft_pixel( h->mb.pic.p_fdec[1]-FDEC_STRIDE-1, h->mb.pic.p_fdec[1]-FDEC_STRIDE+7, b_interlaced );
  1015. - twiddle_topleft_pixel( h->mb.pic.p_fdec[2]-FDEC_STRIDE-1, h->mb.pic.p_fdec[2]-FDEC_STRIDE+7, b_interlaced );
  1016. + twiddle_topleft_pixel( h->mb.pic.p_fdec[1]-FDEC_STRIDE-1, h->mb.pic.p_fdec[1]-FDEC_STRIDE+7, h->sh.b_mbaff );
  1017. + twiddle_topleft_pixel( h->mb.pic.p_fdec[2]-FDEC_STRIDE-1, h->mb.pic.p_fdec[2]-FDEC_STRIDE+7, h->sh.b_mbaff );
  1018. }
  1019. else
  1020. {
  1021. + if( h->sh.b_mbaff )
  1022. + {
  1023. + if( mb_y&1 )
  1024. + {
  1025. + if( mb_x )
  1026. + {
  1027. + h->intra_diagonal_backup[1][0][mb_x*intra_diag_width] = h->intra_border_backup[1][0][(mb_x-1)*16+15];
  1028. + h->intra_diagonal_backup[2][0][mb_x*intra_diag_width] = h->intra_border_backup[2][0][(mb_x-1)*16+15];
  1029. + CP64( &h->intra_diagonal_backup[1][0][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*15 );
  1030. + CP64( &h->intra_diagonal_backup[2][0][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*14 );
  1031. + }
  1032. + memcpy( &h->intra_border_backup[2][i][mb_x*16], h->mb.pic.p_fdec[0]+FDEC_STRIDE*14, 16*sizeof(pixel) );
  1033. + }
  1034. + else
  1035. + {
  1036. + h->intra_diagonal_backup[0][0][mb_x*intra_diag_width] = h->intra_border_backup[0][0][(mb_x-1)*16+15];
  1037. + CP64( &h->intra_diagonal_backup[0][0][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*15 );
  1038. + h->intra_diagonal_backup[3][0][mb_x*intra_diag_width] = h->mb.pic.p_fdec[0][-1+15*FDEC_STRIDE];
  1039. + }
  1040. + }
  1041. h->mc.copy[PIXEL_16x16]( &h->fdec->plane[0][i_pix_offset], i_stride2, h->mb.pic.p_fdec[0], FDEC_STRIDE, 16 );
  1042. memcpy( intra_fdec, h->mb.pic.p_fdec[0]+FDEC_STRIDE*15, 16*sizeof(pixel) );
  1043. - twiddle_topleft_pixel( h->mb.pic.p_fdec[0]-FDEC_STRIDE-1, h->mb.pic.p_fdec[0]-FDEC_STRIDE+15, b_interlaced );
  1044. + twiddle_topleft_pixel( h->mb.pic.p_fdec[0]-FDEC_STRIDE-1, h->mb.pic.p_fdec[0]-FDEC_STRIDE+15, h->sh.b_mbaff );
  1045. }
  1046. }
  1047.  
  1048. --
  1049. 1.7.4
  1050.  
  1051.  
  1052. From db33884079bf79074a67fac2851d8c9425c45bfa Mon Sep 17 00:00:00 2001
  1053. From: Simon Horlick <simonhorlick@gmail.com>
  1054. Date: Thu, 17 Feb 2011 00:56:59 +0000
  1055. Subject: [PATCH 06/25] Change b_interlaced in store_pic back to its original meaning
  1056.  
  1057. ---
  1058. common/macroblock.c | 10 +++++-----
  1059. 1 files changed, 5 insertions(+), 5 deletions(-)
  1060.  
  1061. diff --git a/common/macroblock.c b/common/macroblock.c
  1062. index 60275ae..63a8933 100644
  1063. --- a/common/macroblock.c
  1064. +++ b/common/macroblock.c
  1065. @@ -1238,15 +1238,15 @@ static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb
  1066. {
  1067. int w = i ? 8 : 16;
  1068. int i_stride = h->fdec->i_stride[i];
  1069. - int i_stride2 = i_stride << b_interlaced;
  1070. - int i_pix_offset = b_interlaced
  1071. + int i_stride2 = i_stride << (b_interlaced && h->mb.b_interlaced);
  1072. + int i_pix_offset = (b_interlaced && h->mb.b_interlaced)
  1073. ? 16 * mb_x + w * (mb_y&~1) * i_stride + (mb_y&1) * i_stride
  1074. : 16 * mb_x + w * mb_y * i_stride;
  1075. const int intra_diag_width = 8+1; // One top left sample, then eight top right samples.
  1076. pixel *intra_fdec = &h->intra_border_backup[mb_y&1][i][mb_x*16];
  1077. if( i )
  1078. {
  1079. - if( h->sh.b_mbaff )
  1080. + if( b_interlaced )
  1081. {
  1082. /* Frame macroblocks use the macroblock directly above for intra
  1083. * prediction. Field macroblock pairs predict from fields of the same
  1084. @@ -1293,7 +1293,7 @@ static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb
  1085. }
  1086. else
  1087. {
  1088. - if( h->sh.b_mbaff )
  1089. + if( b_interlaced )
  1090. {
  1091. if( mb_y&1 )
  1092. {
  1093. @@ -1333,7 +1333,7 @@ void x264_macroblock_cache_save( x264_t *h )
  1094. int8_t *i4x4 = h->mb.intra4x4_pred_mode[i_mb_xy];
  1095. uint8_t *nnz = h->mb.non_zero_count[i_mb_xy];
  1096.  
  1097. - if( h->mb.b_interlaced )
  1098. + if( h->sh.b_mbaff )
  1099. {
  1100. x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 0, 1 );
  1101. x264_macroblock_store_pic( h, h->mb.i_mb_x, h->mb.i_mb_y, 1, 1 );
  1102. --
  1103. 1.7.4
  1104.  
  1105.  
  1106. From b816ff5ba65edb03226237e2fc6fff06d7d9e60a Mon Sep 17 00:00:00 2001
  1107. From: Simon Horlick <simonhorlick@gmail.com>
  1108. Date: Fri, 18 Feb 2011 18:57:15 +0000
  1109. Subject: [PATCH 07/25] Only enable twiddle_topleft in progressive
  1110.  
  1111. ---
  1112. common/macroblock.c | 3 ++-
  1113. 1 files changed, 2 insertions(+), 1 deletions(-)
  1114.  
  1115. diff --git a/common/macroblock.c b/common/macroblock.c
  1116. index 63a8933..f775030 100644
  1117. --- a/common/macroblock.c
  1118. +++ b/common/macroblock.c
  1119. @@ -1313,9 +1313,10 @@ static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb
  1120. h->intra_diagonal_backup[3][0][mb_x*intra_diag_width] = h->mb.pic.p_fdec[0][-1+15*FDEC_STRIDE];
  1121. }
  1122. }
  1123. + else
  1124. + twiddle_topleft_pixel( h->mb.pic.p_fdec[0]-FDEC_STRIDE-1, h->mb.pic.p_fdec[0]-FDEC_STRIDE+15, h->sh.b_mbaff );
  1125. h->mc.copy[PIXEL_16x16]( &h->fdec->plane[0][i_pix_offset], i_stride2, h->mb.pic.p_fdec[0], FDEC_STRIDE, 16 );
  1126. memcpy( intra_fdec, h->mb.pic.p_fdec[0]+FDEC_STRIDE*15, 16*sizeof(pixel) );
  1127. - twiddle_topleft_pixel( h->mb.pic.p_fdec[0]-FDEC_STRIDE-1, h->mb.pic.p_fdec[0]-FDEC_STRIDE+15, h->sh.b_mbaff );
  1128. }
  1129. }
  1130.  
  1131. --
  1132. 1.7.4
  1133.  
  1134.  
  1135. From 2b1351a44177a7dc289c9bfe38c7ddc7b0f9a3a2 Mon Sep 17 00:00:00 2001
  1136. From: Simon Horlick <simonhorlick@gmail.com>
  1137. Date: Wed, 16 Mar 2011 21:34:28 +0000
  1138. Subject: [PATCH 08/25] Initial inter support
  1139.  
  1140. ---
  1141. common/common.h | 5 +
  1142. common/frame.c | 33 +++++++-
  1143. common/frame.h | 3 +
  1144. common/macroblock.c | 217 +++++++++++++++++++++++++++++++++++++++++++-------
  1145. common/mc.c | 37 +++++++--
  1146. common/mvpred.c | 23 +++++-
  1147. common/x86/util.h | 24 ------
  1148. encoder/macroblock.c | 4 +-
  1149. 8 files changed, 277 insertions(+), 69 deletions(-)
  1150.  
  1151. diff --git a/common/common.h b/common/common.h
  1152. index c993857..ef9b35a 100644
  1153. --- a/common/common.h
  1154. +++ b/common/common.h
  1155. @@ -600,6 +600,7 @@ struct x264_t
  1156. int i_mb_topright_xy;
  1157. const int *left_index_table;
  1158. int topleft_partition;
  1159. + int allow_skip;
  1160. int intra_border_index;
  1161. int topleft_border_index;
  1162. int topright_border_index;
  1163. @@ -730,6 +731,10 @@ struct x264_t
  1164. /* neighbor CBPs */
  1165. int i_cbp_top;
  1166. int i_cbp_left;
  1167. +
  1168. + /* extra data required for mbaff in mv prediction */
  1169. + int16_t topright_mv[2][3][2];
  1170. + int8_t topright_ref[2][3];
  1171. } cache;
  1172.  
  1173. /* */
  1174. diff --git a/common/frame.c b/common/frame.c
  1175. index eff8ca5..d04f047 100644
  1176. --- a/common/frame.c
  1177. +++ b/common/frame.c
  1178. @@ -48,7 +48,7 @@ x264_frame_t *x264_frame_new( x264_t *h, int b_fdec )
  1179.  
  1180. int i_mb_count = h->mb.i_mb_count;
  1181. int i_stride, i_width, i_lines;
  1182. - int i_padv = PADV << h->param.b_interlaced;
  1183. + int i_padv = PADV << 2*h->param.b_interlaced;
  1184. int luma_plane_size, chroma_plane_size;
  1185. int align = h->param.cpu&X264_CPU_CACHELINE_64 ? 64 : h->param.cpu&X264_CPU_CACHELINE_32 ? 32 : 16;
  1186. int disalign = h->param.cpu&X264_CPU_ALTIVEC ? 1<<9 : 1<<10;
  1187. @@ -99,21 +99,30 @@ x264_frame_t *x264_frame_new( x264_t *h, int b_fdec )
  1188. chroma_plane_size = (frame->i_stride[1] * (frame->i_lines[1] + i_padv));
  1189.  
  1190. CHECKED_MALLOC( frame->buffer[1], chroma_plane_size * sizeof(pixel) );
  1191. + CHECKED_MALLOC( frame->buffer_fld[1], chroma_plane_size * sizeof(pixel) );
  1192. frame->plane[1] = frame->buffer[1] + frame->i_stride[1] * i_padv/2 + PADH;
  1193. + frame->plane_fld[1] = frame->buffer_fld[1] + frame->i_stride[1] * i_padv/2 + PADH;
  1194.  
  1195. /* all 4 luma planes allocated together, since the cacheline split code
  1196. * requires them to be in-phase wrt cacheline alignment. */
  1197. if( h->param.analyse.i_subpel_refine && b_fdec )
  1198. {
  1199. CHECKED_MALLOC( frame->buffer[0], 4*luma_plane_size * sizeof(pixel) );
  1200. + CHECKED_MALLOC( frame->buffer_fld[0], 4*luma_plane_size * sizeof(pixel) );
  1201. for( int i = 0; i < 4; i++ )
  1202. + {
  1203. frame->filtered[i] = frame->buffer[0] + i*luma_plane_size + frame->i_stride[0] * i_padv + PADH;
  1204. + frame->filtered_fld[i] = frame->buffer_fld[0] + i*luma_plane_size + frame->i_stride[0] * i_padv + PADH;
  1205. + }
  1206. frame->plane[0] = frame->filtered[0];
  1207. + frame->plane_fld[0] = frame->filtered_fld[0];
  1208. }
  1209. else
  1210. {
  1211. CHECKED_MALLOC( frame->buffer[0], luma_plane_size * sizeof(pixel) );
  1212. + CHECKED_MALLOC( frame->buffer_fld[0], luma_plane_size * sizeof(pixel) );
  1213. frame->filtered[0] = frame->plane[0] = frame->buffer[0] + frame->i_stride[0] * i_padv + PADH;
  1214. + frame->filtered_fld[0] = frame->plane_fld[0] = frame->buffer_fld[0] + frame->i_stride[0] * i_padv + PADH;
  1215. }
  1216.  
  1217. frame->b_duplicate = 0;
  1218. @@ -200,7 +209,10 @@ void x264_frame_delete( x264_frame_t *frame )
  1219. if( !frame->b_duplicate )
  1220. {
  1221. for( int i = 0; i < 4; i++ )
  1222. + {
  1223. x264_free( frame->buffer[i] );
  1224. + x264_free( frame->buffer_fld[i] );
  1225. + }
  1226. for( int i = 0; i < 4; i++ )
  1227. x264_free( frame->buffer_lowres[i] );
  1228. for( int i = 0; i < X264_BFRAME_MAX+2; i++ )
  1229. @@ -363,16 +375,25 @@ void x264_frame_expand_border( x264_t *h, x264_frame_t *frame, int mb_y, int b_e
  1230. int padh = PADH;
  1231. int padv = PADV >> !!i;
  1232. // buffer: 2 chroma, 3 luma (rounded to 4) because deblocking goes beyond the top of the mb
  1233. - pixel *pix = frame->plane[i] + X264_MAX(0, (16*mb_y-4)*stride >> !!i);
  1234. if( b_end && !b_start )
  1235. height += 4 >> (!!i + h->sh.b_mbaff);
  1236. + pixel *pix;
  1237. if( h->sh.b_mbaff )
  1238. {
  1239. + // border samples for each field are extended separately
  1240. + pix = frame->plane_fld[i] + X264_MAX(0, (16*mb_y-4)*stride >> !!i);
  1241. plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end, i );
  1242. plane_expand_border( pix+stride, stride*2, width, height, padh, padv, b_start, b_end, i );
  1243. +
  1244. + height = (b_end ? 16*(h->mb.i_mb_height - mb_y) : 32) >> !!i;
  1245. + if( b_end && !b_start )
  1246. + height += 4 >> (!!i);
  1247. + pix = frame->plane[i] + X264_MAX(0, (16*mb_y-4)*stride >> !!i);
  1248. + plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end, i );
  1249. }
  1250. else
  1251. {
  1252. + pix = frame->plane[i] + X264_MAX(0, (16*mb_y-4)*stride >> !!i);
  1253. plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end, i );
  1254. }
  1255. }
  1256. @@ -392,14 +413,16 @@ void x264_frame_expand_border_filtered( x264_t *h, x264_frame_t *frame, int mb_y
  1257. for( int i = 1; i < 4; i++ )
  1258. {
  1259. // buffer: 8 luma, to match the hpel filter
  1260. - pixel *pix = frame->filtered[i] + (16*mb_y - (8 << h->sh.b_mbaff)) * stride - 4;
  1261. + pixel *pix;
  1262. if( h->sh.b_mbaff )
  1263. {
  1264. + pix = frame->filtered_fld[i] + (16*mb_y - 16) * stride - 4;
  1265. plane_expand_border( pix, stride*2, width, height, padh, padv, b_start, b_end, 0 );
  1266. plane_expand_border( pix+stride, stride*2, width, height, padh, padv, b_start, b_end, 0 );
  1267. }
  1268. - else
  1269. - plane_expand_border( pix, stride, width, height, padh, padv, b_start, b_end, 0 );
  1270. +
  1271. + pix = frame->filtered[i] + (16*mb_y - 8) * stride - 4;
  1272. + plane_expand_border( pix, stride, width, height << h->sh.b_mbaff, padh, padv, b_start, b_end, 0 );
  1273. }
  1274. }
  1275.  
  1276. diff --git a/common/frame.h b/common/frame.h
  1277. index 0e0ab3d..8fe0627 100644
  1278. --- a/common/frame.h
  1279. +++ b/common/frame.h
  1280. @@ -72,13 +72,16 @@ typedef struct x264_frame
  1281. int i_width_lowres;
  1282. int i_lines_lowres;
  1283. pixel *plane[2];
  1284. + pixel *plane_fld[2];
  1285. pixel *filtered[4]; /* plane[0], H, V, HV */
  1286. + pixel *filtered_fld[4];
  1287. pixel *lowres[4]; /* half-size copy of input frame: Orig, H, V, HV */
  1288. uint16_t *integral;
  1289.  
  1290. /* for unrestricted mv we allocate more data than needed
  1291. * allocated data are stored in buffer */
  1292. pixel *buffer[4];
  1293. + pixel *buffer_fld[4];
  1294. pixel *buffer_lowres[4];
  1295.  
  1296. x264_weight_t weight[X264_REF_MAX][3]; /* [ref_index][plane] */
  1297. diff --git a/common/macroblock.c b/common/macroblock.c
  1298. index f775030..a441981 100644
  1299. --- a/common/macroblock.c
  1300. +++ b/common/macroblock.c
  1301. @@ -523,7 +523,7 @@ static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int mb_x
  1302. int i_pix_offset = h->mb.b_interlaced
  1303. ? 16 * mb_x + w * (mb_y&~1) * i_stride + (mb_y&1) * i_stride
  1304. : 16 * mb_x + w * mb_y * i_stride;
  1305. - pixel *plane_fdec = &h->fdec->plane[i][i_pix_offset];
  1306. + pixel *plane_fdec = h->mb.b_interlaced ? &h->fdec->plane_fld[i][i_pix_offset] : &h->fdec->plane[i][i_pix_offset];
  1307. pixel *intra_fdec = &h->intra_border_backup[h->mb.intra_border_index][i][mb_x*16];
  1308. int ref_pix_offset[2] = { i_pix_offset, i_pix_offset };
  1309. /* ref_pix_offset[0] references the current field and [1] the opposite field. */
  1310. @@ -570,15 +570,28 @@ static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int mb_x
  1311. else
  1312. h->mb.pic.p_fdec[0][-1+j*FDEC_STRIDE] = plane_fdec[-1+j*i_stride2];
  1313. }
  1314. + pixel *plane_src, **filtered_src;
  1315. for( int j = 0; j < h->mb.pic.i_fref[0]; j++ )
  1316. {
  1317. - h->mb.pic.p_fref[0][j][i?4:0] = &h->fref[0][j >> b_interlaced]->plane[i][ref_pix_offset[j&1]];
  1318. + // Interpolate between pixels in same field.
  1319. + if( h->mb.b_interlaced )
  1320. + {
  1321. + plane_src = h->fref[0][j>>1]->plane_fld[i];
  1322. + filtered_src = h->fref[0][j>>1]->filtered_fld;
  1323. + }
  1324. + else
  1325. + {
  1326. + plane_src = h->fref[0][j]->plane[i];
  1327. + filtered_src = h->fref[0][j]->filtered;
  1328. + }
  1329. + h->mb.pic.p_fref[0][j][i?4:0] = plane_src + ref_pix_offset[j&1];
  1330. +
  1331. if( !i )
  1332. {
  1333. for( int k = 1; k < 4; k++ )
  1334. - h->mb.pic.p_fref[0][j][k] = &h->fref[0][j >> b_interlaced]->filtered[k][ref_pix_offset[j&1]];
  1335. + h->mb.pic.p_fref[0][j][k] = filtered_src[k] + ref_pix_offset[j&1];
  1336. if( h->sh.weight[j][0].weightfn )
  1337. - h->mb.pic.p_fref_w[j] = &h->fenc->weighted[j >> b_interlaced][ref_pix_offset[j&1]];
  1338. + h->mb.pic.p_fref_w[j] = &h->fenc->weighted[j >> h->mb.b_interlaced][ref_pix_offset[j&1]];
  1339. else
  1340. h->mb.pic.p_fref_w[j] = h->mb.pic.p_fref[0][j][0];
  1341. }
  1342. @@ -586,10 +599,21 @@ static void ALWAYS_INLINE x264_macroblock_load_pic_pointers( x264_t *h, int mb_x
  1343. if( h->sh.i_type == SLICE_TYPE_B )
  1344. for( int j = 0; j < h->mb.pic.i_fref[1]; j++ )
  1345. {
  1346. - h->mb.pic.p_fref[1][j][i?4:0] = &h->fref[1][j >> b_interlaced]->plane[i][ref_pix_offset[j&1]];
  1347. + if( h->mb.b_interlaced )
  1348. + {
  1349. + plane_src = h->fref[1][j>>1]->plane_fld[i];
  1350. + filtered_src = h->fref[1][j>>1]->filtered_fld;
  1351. + }
  1352. + else
  1353. + {
  1354. + plane_src = h->fref[1][j]->plane[i];
  1355. + filtered_src = h->fref[1][j]->filtered;
  1356. + }
  1357. + h->mb.pic.p_fref[1][j][i?4:0] = plane_src + ref_pix_offset[j&1];
  1358. +
  1359. if( !i )
  1360. for( int k = 1; k < 4; k++ )
  1361. - h->mb.pic.p_fref[1][j][k] = &h->fref[1][j >> b_interlaced]->filtered[k][ref_pix_offset[j&1]];
  1362. + h->mb.pic.p_fref[1][j][k] = filtered_src[k] + ref_pix_offset[j&1];
  1363. }
  1364. }
  1365.  
  1366. @@ -851,7 +875,9 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1367.  
  1368. if( h->mb.i_neighbour & MB_LEFT )
  1369. {
  1370. - h->mb.cache.i_cbp_left = cbp[left[0]];
  1371. + const int16_t top_luma = (cbp[left[0]] >> (left_index_table[18+0]&(~1))) & 2;
  1372. + const int16_t bot_luma = (cbp[left[1]] >> (left_index_table[18+2]&(~1))) & 2;
  1373. + h->mb.cache.i_cbp_left = (cbp[left[0]] & 0xfff0) | (bot_luma<<2) | top_luma;
  1374.  
  1375. /* load intra4x4 */
  1376. h->mb.cache.intra4x4_pred_mode[x264_scan8[0 ] - 1] = i4x4[left[0]][left_index_table[0]];
  1377. @@ -903,8 +929,8 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1378. h->mb.pic.i_fref[0] = h->i_ref[0] << h->mb.b_interlaced;
  1379. h->mb.pic.i_fref[1] = h->i_ref[1] << h->mb.b_interlaced;
  1380. h->mb.cache.i_neighbour_interlaced =
  1381. - !!(h->mb.i_neighbour & MB_LEFT)
  1382. - + !!(h->mb.i_neighbour & MB_TOP);
  1383. + !!(h->mb.i_neighbour & MB_LEFT && h->mb.field[left[0]])
  1384. + + !!(h->mb.i_neighbour & MB_TOP && h->mb.field[top]);
  1385. }
  1386.  
  1387. if( !h->sh.b_mbaff )
  1388. @@ -941,8 +967,18 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1389. int i8 = x264_scan8[0] - 1 - 1*8;
  1390. if( h->mb.i_neighbour & MB_TOPLEFT )
  1391. {
  1392. - h->mb.cache.ref[l][i8] = ref[top_8x8 - 1];
  1393. - CP32( h->mb.cache.mv[l][i8], mv[top_4x4 - 1] );
  1394. + int y = h->mb.i_mb_topleft_xy / h->mb.i_mb_stride;
  1395. + int ir = 2*(s8x8*y + mb_x-1)+1+s8x8;
  1396. + int iv = 4*(s4x4*y + mb_x-1)+3+3*s4x4;
  1397. + if( h->mb.topleft_partition )
  1398. + {
  1399. + /* Take motion vector from the middle of macroblock instead of
  1400. + * the bottom right as usual. */
  1401. + iv -= 2*s4x4;
  1402. + ir -= s8x8;
  1403. + }
  1404. + h->mb.cache.ref[l][i8] = ref[ir];
  1405. + CP32( h->mb.cache.mv[l][i8], mv[iv] );
  1406. }
  1407. else
  1408. {
  1409. @@ -968,8 +1004,9 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1410. i8 = x264_scan8[0] + 4 - 1*8;
  1411. if( h->mb.i_neighbour & MB_TOPRIGHT )
  1412. {
  1413. - h->mb.cache.ref[l][i8] = ref[top_8x8 + 2];
  1414. - CP32( h->mb.cache.mv[l][i8], mv[top_4x4 + 4] );
  1415. + int y = h->mb.i_mb_topright_xy / h->mb.i_mb_stride;
  1416. + h->mb.cache.ref[l][i8] = ref[2*(s8x8*y + (mb_x+1))+s8x8];
  1417. + CP32( h->mb.cache.mv[l][i8], mv[4*(s4x4*y + (mb_x+1))+3*s4x4] );
  1418. }
  1419. else
  1420. h->mb.cache.ref[l][i8] = -2;
  1421. @@ -977,17 +1014,15 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1422. i8 = x264_scan8[0] - 1;
  1423. if( h->mb.i_neighbour & MB_LEFT )
  1424. {
  1425. - const int ir = h->mb.i_b8_xy - 1;
  1426. - const int iv = h->mb.i_b4_xy - 1;
  1427. - h->mb.cache.ref[l][i8+0*8] =
  1428. - h->mb.cache.ref[l][i8+1*8] = ref[ir + 0*s8x8];
  1429. - h->mb.cache.ref[l][i8+2*8] =
  1430. - h->mb.cache.ref[l][i8+3*8] = ref[ir + 1*s8x8];
  1431. -
  1432. - CP32( h->mb.cache.mv[l][i8+0*8], mv[iv + 0*s4x4] );
  1433. - CP32( h->mb.cache.mv[l][i8+1*8], mv[iv + 1*s4x4] );
  1434. - CP32( h->mb.cache.mv[l][i8+2*8], mv[iv + 2*s4x4] );
  1435. - CP32( h->mb.cache.mv[l][i8+3*8], mv[iv + 3*s4x4] );
  1436. + h->mb.cache.ref[l][i8+0*8] = ref[h->mb.left_b8[0] + 1 + s8x8*((left_index_table[12+0]&~1)>>1)];
  1437. + h->mb.cache.ref[l][i8+1*8] = ref[h->mb.left_b8[0] + 1 + s8x8*((left_index_table[12+1]&~1)>>1)];
  1438. + h->mb.cache.ref[l][i8+2*8] = ref[h->mb.left_b8[1] + 1 + s8x8*((left_index_table[12+2]&~1)>>1)];
  1439. + h->mb.cache.ref[l][i8+3*8] = ref[h->mb.left_b8[1] + 1 + s8x8*((left_index_table[12+3]&~1)>>1)];
  1440. +
  1441. + CP32( h->mb.cache.mv[l][i8+0*8], mv[h->mb.left_b4[0] + 3 + s4x4*left_index_table[12+0]] );
  1442. + CP32( h->mb.cache.mv[l][i8+1*8], mv[h->mb.left_b4[0] + 3 + s4x4*left_index_table[12+1]] );
  1443. + CP32( h->mb.cache.mv[l][i8+2*8], mv[h->mb.left_b4[1] + 3 + s4x4*left_index_table[12+2]] );
  1444. + CP32( h->mb.cache.mv[l][i8+3*8], mv[h->mb.left_b4[1] + 3 + s4x4*left_index_table[12+3]] );
  1445. }
  1446. else
  1447. {
  1448. @@ -998,6 +1033,42 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1449. }
  1450. }
  1451.  
  1452. + /* Extra logic for top right mv in mbaff.
  1453. + * . . . d . . a .
  1454. + * . . . e . . . .
  1455. + * . . . f b . c .
  1456. + * . . . . . . . .
  1457. + *
  1458. + * If the top right of the 4x4 partitions labeled a, b and c in the
  1459. + * above diagram do not exist, but the entries d, e and f exist (in
  1460. + * the macroblock to the left) then use those instead.
  1461. + */
  1462. + if( h->param.b_interlaced )
  1463. + {
  1464. + if( h->mb.i_neighbour & MB_LEFT )
  1465. + {
  1466. + if( h->mb.b_interlaced && !h->mb.field[h->mb.i_mb_xy-1] )
  1467. + {
  1468. + h->mb.cache.topright_ref[l][0] = ref[h->mb.left_b8[0] + 1 + s8x8*0];
  1469. + h->mb.cache.topright_ref[l][1] = ref[h->mb.left_b8[0] + 1 + s8x8*1];
  1470. + h->mb.cache.topright_ref[l][2] = ref[h->mb.left_b8[1] + 1 + s8x8*0];
  1471. + CP32( h->mb.cache.topright_mv[l][0], mv[h->mb.left_b4[0] + 3 + s4x4*(left_index_table[12+0]+1)] );
  1472. + CP32( h->mb.cache.topright_mv[l][1], mv[h->mb.left_b4[0] + 3 + s4x4*(left_index_table[12+1]+1)] );
  1473. + CP32( h->mb.cache.topright_mv[l][2], mv[h->mb.left_b4[1] + 3 + s4x4*(left_index_table[12+2]+1)] );
  1474. + }
  1475. + else if( !h->mb.b_interlaced && h->mb.field[h->mb.i_mb_xy-1] )
  1476. + {
  1477. + // Looking at the bottom field so always take the bottom macroblock of the pair.
  1478. + h->mb.cache.topright_ref[l][0] = ref[h->mb.left_b8[0] + 1 + s8x8*2 + s8x8*left_index_table[12+4]];
  1479. + h->mb.cache.topright_ref[l][1] = ref[h->mb.left_b8[0] + 1 + s8x8*2 + s8x8*left_index_table[12+4]];
  1480. + h->mb.cache.topright_ref[l][2] = ref[h->mb.left_b8[0] + 1 + s8x8*2 + s8x8*left_index_table[12+5]];
  1481. + CP32( h->mb.cache.topright_mv[l][0], mv[h->mb.left_b4[0] + 3 + s4x4*4 + s4x4*left_index_table[12+0]] );
  1482. + CP32( h->mb.cache.topright_mv[l][1], mv[h->mb.left_b4[0] + 3 + s4x4*4 + s4x4*left_index_table[12+1]] );
  1483. + CP32( h->mb.cache.topright_mv[l][2], mv[h->mb.left_b4[0] + 3 + s4x4*4 + s4x4*left_index_table[12+2]] );
  1484. + }
  1485. + }
  1486. + }
  1487. +
  1488. if( h->param.b_cabac )
  1489. {
  1490. uint8_t (*mvd)[8][2] = h->mb.mvd[l];
  1491. @@ -1006,16 +1077,103 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1492. else
  1493. M64( h->mb.cache.mvd[l][x264_scan8[0] - 8] ) = 0;
  1494.  
  1495. - if( h->mb.i_neighbour & MB_LEFT )
  1496. + if( h->mb.cache.ref[l][x264_scan8[0]-1] >= 0 )
  1497. {
  1498. CP16( h->mb.cache.mvd[l][x264_scan8[0 ] - 1], mvd[left[0]][left_index_table[0]] );
  1499. CP16( h->mb.cache.mvd[l][x264_scan8[2 ] - 1], mvd[left[0]][left_index_table[1]] );
  1500. + }
  1501. + else
  1502. + {
  1503. + M16( h->mb.cache.mvd[l][x264_scan8[0]-1+0*8] ) = 0;
  1504. + M16( h->mb.cache.mvd[l][x264_scan8[0]-1+1*8] ) = 0;
  1505. + }
  1506. + if( h->mb.cache.ref[l][x264_scan8[0]-1+2*8] >=0 )
  1507. + {
  1508. CP16( h->mb.cache.mvd[l][x264_scan8[8 ] - 1], mvd[left[1]][left_index_table[2]] );
  1509. CP16( h->mb.cache.mvd[l][x264_scan8[10] - 1], mvd[left[1]][left_index_table[3]] );
  1510. }
  1511. else
  1512. - for( int i = 0; i < 4; i++ )
  1513. - M16( h->mb.cache.mvd[l][x264_scan8[0]-1+i*8] ) = 0;
  1514. + {
  1515. + M16( h->mb.cache.mvd[l][x264_scan8[0]-1+2*8] ) = 0;
  1516. + M16( h->mb.cache.mvd[l][x264_scan8[0]-1+3*8] ) = 0;
  1517. + }
  1518. + }
  1519. +
  1520. + /* If motion vectors are cached from frame macroblocks but this
  1521. + * macroblock is a field macroblock then the motion vector must be
  1522. + * halved. Similarly, motion vectors from field macroblocks are doubled. */
  1523. + if( h->sh.b_mbaff )
  1524. + {
  1525. +#define MAP_MVS\
  1526. + MAP_F2F(mv, ref, x264_scan8[0] - 1 - 1*8, h->mb.i_mb_topleft_xy)\
  1527. + MAP_F2F(mv, ref, x264_scan8[0] + 0 - 1*8, top)\
  1528. + MAP_F2F(mv, ref, x264_scan8[0] + 1 - 1*8, top)\
  1529. + MAP_F2F(mv, ref, x264_scan8[0] + 2 - 1*8, top)\
  1530. + MAP_F2F(mv, ref, x264_scan8[0] + 3 - 1*8, top)\
  1531. + MAP_F2F(mv, ref, x264_scan8[0] + 4 - 1*8, h->mb.i_mb_topright_xy)\
  1532. + MAP_F2F(mv, ref, x264_scan8[0] - 1 + 0*8, left[0])\
  1533. + MAP_F2F(mv, ref, x264_scan8[0] - 1 + 1*8, left[0])\
  1534. + MAP_F2F(mv, ref, x264_scan8[0] - 1 + 2*8, left[1])\
  1535. + MAP_F2F(mv, ref, x264_scan8[0] - 1 + 3*8, left[1])\
  1536. + MAP_F2F(topright_mv, topright_ref, 0, left[0])\
  1537. + MAP_F2F(topright_mv, topright_ref, 1, left[0])\
  1538. + MAP_F2F(topright_mv, topright_ref, 2, left[1])
  1539. +
  1540. + if( h->mb.b_interlaced )
  1541. + {
  1542. +#define MAP_F2F(varmv, varref, index, macroblock)\
  1543. + if( h->mb.cache.varref[l][index] >= 0 && macroblock >= 0 && !h->mb.field[macroblock] )\
  1544. + {\
  1545. + h->mb.cache.varref[l][index] <<= 1;\
  1546. + h->mb.cache.varmv[l][index][1] /= 2;\
  1547. + h->mb.cache.mvd[l][index][1] >>= 1;\
  1548. + }
  1549. + MAP_MVS
  1550. +#undef MAP_F2F
  1551. + }
  1552. + else
  1553. + {
  1554. +#define MAP_F2F(varmv, varref, index, macroblock)\
  1555. + if( h->mb.cache.varref[l][index] >= 0 && macroblock >= 0 && h->mb.field[macroblock] )\
  1556. + {\
  1557. + h->mb.cache.varref[l][index] >>= 1;\
  1558. + h->mb.cache.varmv[l][index][1] <<= 1;\
  1559. + h->mb.cache.mvd[l][index][1] <<= 1;\
  1560. + }
  1561. + MAP_MVS
  1562. +#undef MAP_F2F
  1563. + }
  1564. + }
  1565. + }
  1566. +
  1567. + /* Check whether skip here would cause decoder to predict interlace mode incorrectly. */
  1568. + h->mb.allow_skip = 1;
  1569. + if( h->sh.b_mbaff && (mb_y&1) && IS_SKIP(h->mb.type[h->mb.i_mb_xy - h->mb.i_mb_stride]) )
  1570. + {
  1571. + if( h->mb.i_neighbour & MB_LEFT )
  1572. + {
  1573. + if( h->mb.field[h->mb.i_mb_xy - 1] != h->mb.b_interlaced )
  1574. + h->mb.allow_skip = 0;
  1575. + }
  1576. + else if( h->mb.i_neighbour & MB_TOP )
  1577. + {
  1578. + if( h->mb.field[h->mb.i_mb_top_xy] != h->mb.b_interlaced )
  1579. + h->mb.allow_skip = 0;
  1580. + }
  1581. + else // Frame mb pair is predicted
  1582. + {
  1583. + if( h->mb.b_interlaced )
  1584. + h->mb.allow_skip = 0;
  1585. + }
  1586. + if( !h->mb.allow_skip )
  1587. + {
  1588. + if( IS_SKIP(h->mb.i_type) )
  1589. + {
  1590. + if( h->mb.i_type == P_SKIP )
  1591. + h->mb.i_type = P_L0;
  1592. + else if( h->mb.i_type == B_SKIP )
  1593. + h->mb.i_type = B_DIRECT;
  1594. + }
  1595. }
  1596. }
  1597.  
  1598. @@ -1286,6 +1444,7 @@ static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb
  1599. }
  1600. }
  1601. h->mc.store_interleave_8x8x2( &h->fdec->plane[1][i_pix_offset], i_stride2, h->mb.pic.p_fdec[1], h->mb.pic.p_fdec[2] );
  1602. + h->mc.store_interleave_8x8x2( &h->fdec->plane_fld[1][i_pix_offset], i_stride2, h->mb.pic.p_fdec[1], h->mb.pic.p_fdec[2] );
  1603. memcpy( intra_fdec, h->mb.pic.p_fdec[1]+FDEC_STRIDE*7, 8*sizeof(pixel) );
  1604. memcpy( intra_fdec+8, h->mb.pic.p_fdec[2]+FDEC_STRIDE*7, 8*sizeof(pixel) );
  1605. twiddle_topleft_pixel( h->mb.pic.p_fdec[1]-FDEC_STRIDE-1, h->mb.pic.p_fdec[1]-FDEC_STRIDE+7, h->sh.b_mbaff );
  1606. @@ -1299,12 +1458,13 @@ static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb
  1607. {
  1608. if( mb_x )
  1609. {
  1610. + // Take rightmost sample from top border of left mb to use as topleft here.
  1611. h->intra_diagonal_backup[1][0][mb_x*intra_diag_width] = h->intra_border_backup[1][0][(mb_x-1)*16+15];
  1612. h->intra_diagonal_backup[2][0][mb_x*intra_diag_width] = h->intra_border_backup[2][0][(mb_x-1)*16+15];
  1613. CP64( &h->intra_diagonal_backup[1][0][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*15 );
  1614. CP64( &h->intra_diagonal_backup[2][0][(mb_x-1)*intra_diag_width+1], h->mb.pic.p_fdec[0]+FDEC_STRIDE*14 );
  1615. }
  1616. - memcpy( &h->intra_border_backup[2][i][mb_x*16], h->mb.pic.p_fdec[0]+FDEC_STRIDE*14, 16*sizeof(pixel) );
  1617. + memcpy( &h->intra_border_backup[2][0][mb_x*16], h->mb.pic.p_fdec[0]+FDEC_STRIDE*14, 16*sizeof(pixel) );
  1618. }
  1619. else
  1620. {
  1621. @@ -1316,6 +1476,7 @@ static void ALWAYS_INLINE x264_macroblock_store_pic( x264_t *h, int mb_x, int mb
  1622. else
  1623. twiddle_topleft_pixel( h->mb.pic.p_fdec[0]-FDEC_STRIDE-1, h->mb.pic.p_fdec[0]-FDEC_STRIDE+15, h->sh.b_mbaff );
  1624. h->mc.copy[PIXEL_16x16]( &h->fdec->plane[0][i_pix_offset], i_stride2, h->mb.pic.p_fdec[0], FDEC_STRIDE, 16 );
  1625. + h->mc.copy[PIXEL_16x16]( &h->fdec->plane_fld[0][i_pix_offset], i_stride2, h->mb.pic.p_fdec[0], FDEC_STRIDE, 16 );
  1626. memcpy( intra_fdec, h->mb.pic.p_fdec[0]+FDEC_STRIDE*15, 16*sizeof(pixel) );
  1627. }
  1628. }
  1629. diff --git a/common/mc.c b/common/mc.c
  1630. index 76061c3..e594785 100644
  1631. --- a/common/mc.c
  1632. +++ b/common/mc.c
  1633. @@ -512,22 +512,43 @@ void x264_mc_init( int cpu, x264_mc_functions_t *pf )
  1634. void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
  1635. {
  1636. const int b_interlaced = h->sh.b_mbaff;
  1637. - const int stride = frame->i_stride[0] << b_interlaced;
  1638. + int stride = frame->i_stride[0];
  1639. const int width = frame->i_width[0];
  1640. - int start = (mb_y*16 >> b_interlaced) - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
  1641. - int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
  1642. + int start = mb_y*16 - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
  1643. + int height = (b_end ? frame->i_lines[0] + 16 : (mb_y+b_interlaced)*16) + 8;
  1644. int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
  1645.  
  1646. if( mb_y & b_interlaced )
  1647. return;
  1648.  
  1649. - for( int y = 0; y <= b_interlaced; y++, offs += frame->i_stride[0] )
  1650. + h->mc.hpel_filter(
  1651. + frame->filtered[1] + offs,
  1652. + frame->filtered[2] + offs,
  1653. + frame->filtered[3] + offs,
  1654. + frame->plane[0] + offs,
  1655. + stride, width + 16, height - start,
  1656. + h->scratch_buffer );
  1657. +
  1658. + if( b_interlaced )
  1659. {
  1660. + /* MC must happen between pixels in the same field. */
  1661. + stride = frame->i_stride[0] << 1;
  1662. + start = (mb_y*16 >> 1) - 8;
  1663. + height = ((b_end ? frame->i_lines[0] : mb_y*16) >> 1) + 8;
  1664. + offs = start*stride - 8;
  1665. + h->mc.hpel_filter(
  1666. + frame->filtered_fld[1] + offs,
  1667. + frame->filtered_fld[2] + offs,
  1668. + frame->filtered_fld[3] + offs,
  1669. + frame->plane_fld[0] + offs,
  1670. + stride, width + 16, height - start,
  1671. + h->scratch_buffer );
  1672. + offs += frame->i_stride[0];
  1673. h->mc.hpel_filter(
  1674. - frame->filtered[1] + offs,
  1675. - frame->filtered[2] + offs,
  1676. - frame->filtered[3] + offs,
  1677. - frame->plane[0] + offs,
  1678. + frame->filtered_fld[1] + offs,
  1679. + frame->filtered_fld[2] + offs,
  1680. + frame->filtered_fld[3] + offs,
  1681. + frame->plane_fld[0] + offs,
  1682. stride, width + 16, height - start,
  1683. h->scratch_buffer );
  1684. }
  1685. diff --git a/common/mvpred.c b/common/mvpred.c
  1686. index c8efe1f..278e0ac 100644
  1687. --- a/common/mvpred.c
  1688. +++ b/common/mvpred.c
  1689. @@ -38,12 +38,33 @@ void x264_mb_predict_mv( x264_t *h, int i_list, int idx, int i_width, int16_t mv
  1690. int i_refc = h->mb.cache.ref[i_list][i8 - 8 + i_width];
  1691. int16_t *mv_c = h->mb.cache.mv[i_list][i8 - 8 + i_width];
  1692.  
  1693. + // Partitions not yet reached in scan order are unavailable.
  1694. if( (idx&3) >= 2 + (i_width&1) || i_refc == -2 )
  1695. {
  1696. i_refc = h->mb.cache.ref[i_list][i8 - 8 - 1];
  1697. mv_c = h->mb.cache.mv[i_list][i8 - 8 - 1];
  1698. - }
  1699.  
  1700. + if( h->param.b_interlaced
  1701. + && h->mb.cache.ref[i_list][x264_scan8[0]-1] != -2
  1702. + && h->mb.b_interlaced != h->mb.field[h->mb.i_mb_left_xy[0]] )
  1703. + {
  1704. + if( idx == 2 )
  1705. + {
  1706. + mv_c = h->mb.cache.topright_mv[i_list][0];
  1707. + i_refc = h->mb.cache.topright_ref[i_list][0];
  1708. + }
  1709. + else if( idx == 8 )
  1710. + {
  1711. + mv_c = h->mb.cache.topright_mv[i_list][1];
  1712. + i_refc = h->mb.cache.topright_ref[i_list][1];
  1713. + }
  1714. + else if( idx == 10 )
  1715. + {
  1716. + mv_c = h->mb.cache.topright_mv[i_list][2];
  1717. + i_refc = h->mb.cache.topright_ref[i_list][2];
  1718. + }
  1719. + }
  1720. + }
  1721. if( h->mb.i_partition == D_16x8 )
  1722. {
  1723. if( idx == 0 )
  1724. diff --git a/common/x86/util.h b/common/x86/util.h
  1725. index 6544207..01e54f9 100644
  1726. --- a/common/x86/util.h
  1727. +++ b/common/x86/util.h
  1728. @@ -87,30 +87,6 @@ static ALWAYS_INLINE int x264_predictor_difference_mmxext( int16_t (*mvc)[2], in
  1729. return sum;
  1730. }
  1731.  
  1732. -#define x264_cabac_mvd_sum x264_cabac_mvd_sum_mmxext
  1733. -static ALWAYS_INLINE uint16_t x264_cabac_mvd_sum_mmxext(uint8_t *mvdleft, uint8_t *mvdtop)
  1734. -{
  1735. - static const uint64_t pb_2 = 0x0202020202020202ULL;
  1736. - static const uint64_t pb_32 = 0x2020202020202020ULL;
  1737. - int amvd;
  1738. - asm(
  1739. - "movd %1, %%mm0 \n"
  1740. - "movd %2, %%mm1 \n"
  1741. - "paddb %%mm1, %%mm0 \n"
  1742. - "pxor %%mm2, %%mm2 \n"
  1743. - "movq %%mm0, %%mm1 \n"
  1744. - "pcmpgtb %3, %%mm0 \n"
  1745. - "pcmpgtb %4, %%mm1 \n"
  1746. - "psubb %%mm0, %%mm2 \n"
  1747. - "psubb %%mm1, %%mm2 \n"
  1748. - "movd %%mm2, %0 \n"
  1749. - :"=r"(amvd)
  1750. - :"m"(M16( mvdleft )),"m"(M16( mvdtop )),
  1751. - "m"(pb_2),"m"(pb_32)
  1752. - );
  1753. - return amvd;
  1754. -}
  1755. -
  1756. #define x264_predictor_roundclip x264_predictor_roundclip_mmxext
  1757. static void ALWAYS_INLINE x264_predictor_roundclip_mmxext( int16_t (*dst)[2], int16_t (*mvc)[2], int i_mvc, int mv_x_min, int mv_x_max, int mv_y_min, int mv_y_max )
  1758. {
  1759. diff --git a/encoder/macroblock.c b/encoder/macroblock.c
  1760. index a3fcd61..28609d8 100644
  1761. --- a/encoder/macroblock.c
  1762. +++ b/encoder/macroblock.c
  1763. @@ -609,9 +609,7 @@ void x264_macroblock_encode( x264_t *h )
  1764. return;
  1765. }
  1766.  
  1767. - if( h->sh.b_mbaff
  1768. - && h->mb.i_mb_xy == h->sh.i_first_mb + h->mb.i_mb_stride
  1769. - && IS_SKIP(h->mb.type[h->sh.i_first_mb]) )
  1770. + if( !h->mb.allow_skip )
  1771. {
  1772. /* The first skip is predicted to be a frame mb pair.
  1773. * We don't yet support the aff part of mbaff, so force it to non-skip
  1774. --
  1775. 1.7.4
  1776.  
  1777.  
  1778. From 88aa754aa9a2e3f2907f656e21439e7b93b8cfab Mon Sep 17 00:00:00 2001
  1779. From: Simon Horlick <simonhorlick@gmail.com>
  1780. Date: Wed, 16 Mar 2011 21:34:51 +0000
  1781. Subject: [PATCH 09/25] Copy deblocked pixels to other plane
  1782.  
  1783. ---
  1784. common/deblock.c | 15 +++++++++++++++
  1785. 1 files changed, 15 insertions(+), 0 deletions(-)
  1786.  
  1787. diff --git a/common/deblock.c b/common/deblock.c
  1788. index 0800461..52d410d 100644
  1789. --- a/common/deblock.c
  1790. +++ b/common/deblock.c
  1791. @@ -389,6 +389,21 @@ void x264_frame_deblock_row( x264_t *h, int mb_y )
  1792.  
  1793. #undef FILTER
  1794. }
  1795. +
  1796. + // FIXME: Don't copy the whole frame around.
  1797. + int y = mb_y*16;
  1798. + int start = mb_y == h->i_threadslice_start;
  1799. + int last = mb_y == h->i_threadslice_end - (1 << h->sh.b_mbaff);
  1800. + int height = last ? 32+4 : 32;
  1801. + if( !start ) y -= 4; // Make sure to copy the above four rows of deblocked pixels.
  1802. + for( int i = y; i < y+height; i++ )
  1803. + memcpy( h->fdec->plane_fld[0] + i*stridey, h->fdec->plane[0] + i*stridey, h->mb.i_mb_width*16*sizeof(pixel) );
  1804. +
  1805. + y = mb_y*8;
  1806. + height = last ? 16+2 : 16;
  1807. + if( !start ) y -=2;
  1808. + for( int i = y; i < y+height; i++ )
  1809. + memcpy( h->fdec->plane_fld[1] + i*strideuv, h->fdec->plane[1] + i*strideuv, h->mb.i_mb_width*16*sizeof(pixel) );
  1810. }
  1811.  
  1812. /* For deblock-aware RD.
  1813. --
  1814. 1.7.4
  1815.  
  1816.  
  1817. From d42239c146bd2fc1417987ce1794fbf049796112 Mon Sep 17 00:00:00 2001
  1818. From: Simon Horlick <simonhorlick@gmail.com>
  1819. Date: Sun, 6 Feb 2011 22:58:39 +0000
  1820. Subject: [PATCH 10/25] Fix thread max mv check
  1821.  
  1822. ---
  1823. encoder/analyse.c | 2 +-
  1824. 1 files changed, 1 insertions(+), 1 deletions(-)
  1825.  
  1826. diff --git a/encoder/analyse.c b/encoder/analyse.c
  1827. index 87125c1..4f439d4 100644
  1828. --- a/encoder/analyse.c
  1829. +++ b/encoder/analyse.c
  1830. @@ -460,7 +460,7 @@ static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int qp )
  1831.  
  1832. if( h->param.b_deterministic )
  1833. thread_mvy_range = h->param.analyse.i_mv_range_thread;
  1834. - if( h->mb.b_interlaced )
  1835. + if( h->sh.b_mbaff )
  1836. thread_mvy_range >>= 1;
  1837.  
  1838. x264_analyse_weight_frame( h, pix_y + thread_mvy_range );
  1839. --
  1840. 1.7.4
  1841.  
  1842.  
  1843. From 0ce16f54027e00ee0a9070e807eed3ca570325ad Mon Sep 17 00:00:00 2001
  1844. From: Simon Horlick <simonhorlick@gmail.com>
  1845. Date: Sun, 20 Feb 2011 15:31:55 +0000
  1846. Subject: [PATCH 11/25] Track what interlace decision the decoder is using
  1847.  
  1848. ---
  1849. common/common.h | 1 +
  1850. encoder/cabac.c | 17 ++++++++++++++++-
  1851. encoder/encoder.c | 11 +++++++++++
  1852. 3 files changed, 28 insertions(+), 1 deletions(-)
  1853.  
  1854. diff --git a/common/common.h b/common/common.h
  1855. index ef9b35a..bc14c10 100644
  1856. --- a/common/common.h
  1857. +++ b/common/common.h
  1858. @@ -604,6 +604,7 @@ struct x264_t
  1859. int intra_border_index;
  1860. int topleft_border_index;
  1861. int topright_border_index;
  1862. + int field_decoding_flag;
  1863.  
  1864. /**** thread synchronization ends here ****/
  1865. /* subsequent variables are either thread-local or constant,
  1866. diff --git a/encoder/cabac.c b/encoder/cabac.c
  1867. index 334318d..6138d06 100644
  1868. --- a/encoder/cabac.c
  1869. +++ b/encoder/cabac.c
  1870. @@ -66,6 +66,21 @@ static inline void x264_cabac_mb_type_intra( x264_t *h, x264_cabac_t *cb, int i_
  1871. }
  1872. }
  1873.  
  1874. +static void x264_cabac_field_decoding_flag( x264_t *h, x264_cabac_t *cb )
  1875. +{
  1876. + const int top = h->mb.i_mb_xy - 2*h->mb.i_mb_stride;
  1877. + int ctx = 0;
  1878. + ctx += h->mb.field_decoding_flag & !!h->mb.i_mb_x;
  1879. + ctx += (top >= 0
  1880. + && h->mb.slice_table[top] == h->sh.i_first_mb
  1881. + && h->mb.field[top]);
  1882. +
  1883. + x264_cabac_encode_decision_noup( cb, 70 + ctx, h->mb.b_interlaced );
  1884. +#if !RDO_SKIP_BS
  1885. + h->mb.field_decoding_flag = h->mb.b_interlaced;
  1886. +#endif
  1887. +}
  1888. +
  1889. static void x264_cabac_mb_type( x264_t *h, x264_cabac_t *cb )
  1890. {
  1891. const int i_mb_type = h->mb.i_type;
  1892. @@ -73,7 +88,7 @@ static void x264_cabac_mb_type( x264_t *h, x264_cabac_t *cb )
  1893. if( h->sh.b_mbaff &&
  1894. (!(h->mb.i_mb_y & 1) || IS_SKIP(h->mb.type[h->mb.i_mb_xy - h->mb.i_mb_stride])) )
  1895. {
  1896. - x264_cabac_encode_decision_noup( cb, 70 + h->mb.cache.i_neighbour_interlaced, h->mb.b_interlaced );
  1897. + x264_cabac_field_decoding_flag( h, cb );
  1898. }
  1899.  
  1900. if( h->sh.i_type == SLICE_TYPE_I )
  1901. diff --git a/encoder/encoder.c b/encoder/encoder.c
  1902. index 9f294d1..2525fec 100644
  1903. --- a/encoder/encoder.c
  1904. +++ b/encoder/encoder.c
  1905. @@ -1932,6 +1932,8 @@ static int x264_slice_write( x264_t *h )
  1906. i_mb_x = h->sh.i_first_mb % h->mb.i_mb_width;
  1907. i_skip = 0;
  1908.  
  1909. + h->mb.field_decoding_flag = 0;
  1910. +
  1911. while( (mb_xy = i_mb_x + i_mb_y * h->mb.i_mb_width) <= h->sh.i_last_mb )
  1912. {
  1913. int mb_spos = bs_pos(&h->out.bs) + x264_cabac_pos(&h->cabac);
  1914. @@ -1988,7 +1990,12 @@ static int x264_slice_write( x264_t *h )
  1915. x264_cabac_encode_terminal( &h->cabac );
  1916.  
  1917. if( IS_SKIP( h->mb.i_type ) )
  1918. + {
  1919. + // FIXME: It might be better to change the interlace type
  1920. + // rather than forcing a skip to be non-skip, but this would
  1921. + // require modifying the already saved image data.
  1922. x264_cabac_mb_skip( h, 1 );
  1923. + }
  1924. else
  1925. {
  1926. if( h->sh.i_type != SLICE_TYPE_I )
  1927. @@ -2145,6 +2152,10 @@ static int x264_slice_write( x264_t *h )
  1928. {
  1929. i_mb_y++;
  1930. i_mb_x = 0;
  1931. + if( h->sh.b_mbaff && i_mb_y > 0 )
  1932. + h->mb.field_decoding_flag = h->mb.field[i_mb_x+(i_mb_y-1)*h->mb.i_mb_stride];
  1933. + else
  1934. + h->mb.field_decoding_flag = 0;
  1935. }
  1936. }
  1937. h->out.nal[h->out.i_nal].i_last_mb = h->sh.i_last_mb;
  1938. --
  1939. 1.7.4
  1940.  
  1941.  
  1942. From 02af876c83b5a6fc69a20c1928ad604069e533a7 Mon Sep 17 00:00:00 2001
  1943. From: Simon Horlick <simonhorlick@gmail.com>
  1944. Date: Sun, 20 Feb 2011 15:35:44 +0000
  1945. Subject: [PATCH 12/25] Disallow skip where interlace would be wrong
  1946.  
  1947. ---
  1948. common/macroblock.c | 27 +++++++++++++++++++--------
  1949. 1 files changed, 19 insertions(+), 8 deletions(-)
  1950.  
  1951. diff --git a/common/macroblock.c b/common/macroblock.c
  1952. index a441981..4fe1f82 100644
  1953. --- a/common/macroblock.c
  1954. +++ b/common/macroblock.c
  1955. @@ -1148,6 +1148,17 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1956.  
  1957. /* Check whether skip here would cause decoder to predict interlace mode incorrectly. */
  1958. h->mb.allow_skip = 1;
  1959. + int prevmb = mb_x + h->mb.i_mb_stride*mb_y;
  1960. + if( mb_y&1 )
  1961. + prevmb -= h->mb.i_mb_stride;
  1962. + else if( mb_x )
  1963. + prevmb = (mb_y+1)*h->mb.i_mb_stride + mb_x - 1;
  1964. + else
  1965. + prevmb = (mb_y-1)*h->mb.i_mb_stride + h->mb.i_mb_width;
  1966. + if( h->mb.b_interlaced != h->mb.field_decoding_flag && IS_SKIP(h->mb.type[prevmb]) )
  1967. + {
  1968. + h->mb.allow_skip = 0;
  1969. + }
  1970. if( h->sh.b_mbaff && (mb_y&1) && IS_SKIP(h->mb.type[h->mb.i_mb_xy - h->mb.i_mb_stride]) )
  1971. {
  1972. if( h->mb.i_neighbour & MB_LEFT )
  1973. @@ -1165,15 +1176,15 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  1974. if( h->mb.b_interlaced )
  1975. h->mb.allow_skip = 0;
  1976. }
  1977. - if( !h->mb.allow_skip )
  1978. + }
  1979. + if( !h->mb.allow_skip )
  1980. + {
  1981. + if( IS_SKIP(h->mb.i_type) )
  1982. {
  1983. - if( IS_SKIP(h->mb.i_type) )
  1984. - {
  1985. - if( h->mb.i_type == P_SKIP )
  1986. - h->mb.i_type = P_L0;
  1987. - else if( h->mb.i_type == B_SKIP )
  1988. - h->mb.i_type = B_DIRECT;
  1989. - }
  1990. + if( h->mb.i_type == P_SKIP )
  1991. + h->mb.i_type = P_L0;
  1992. + else if( h->mb.i_type == B_SKIP )
  1993. + h->mb.i_type = B_DIRECT;
  1994. }
  1995. }
  1996.  
  1997. --
  1998. 1.7.4
  1999.  
  2000.  
  2001. From 2a15d908619a780aa5a0f8bd9c6b61ade5d52237 Mon Sep 17 00:00:00 2001
  2002. From: Simon Horlick <simonhorlick@gmail.com>
  2003. Date: Sun, 20 Feb 2011 15:36:29 +0000
  2004. Subject: [PATCH 13/25] CABAC encoding of skips
  2005.  
  2006. ---
  2007. common/common.h | 2 ++
  2008. common/macroblock.c | 29 +++++++++++++++++++++++++++++
  2009. encoder/cabac.c | 14 +++++++-------
  2010. 3 files changed, 38 insertions(+), 7 deletions(-)
  2011.  
  2012. diff --git a/common/common.h b/common/common.h
  2013. index bc14c10..6d81496 100644
  2014. --- a/common/common.h
  2015. +++ b/common/common.h
  2016. @@ -605,6 +605,8 @@ struct x264_t
  2017. int topleft_border_index;
  2018. int topright_border_index;
  2019. int field_decoding_flag;
  2020. + int left_skip;
  2021. + int top_skip;
  2022.  
  2023. /**** thread synchronization ends here ****/
  2024. /* subsequent variables are either thread-local or constant,
  2025. diff --git a/common/macroblock.c b/common/macroblock.c
  2026. index 4fe1f82..c01320c 100644
  2027. --- a/common/macroblock.c
  2028. +++ b/common/macroblock.c
  2029. @@ -1188,6 +1188,35 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  2030. }
  2031. }
  2032.  
  2033. + if( h->param.b_cabac )
  2034. + {
  2035. + /* Neighbours here are calculated based on field_decoding_flag */
  2036. + int left_xy, top_xy;
  2037. + if( h->sh.b_mbaff )
  2038. + {
  2039. + int mb_xy = mb_x + (h->mb.i_mb_y&~1)*h->mb.i_mb_stride;
  2040. + left_xy = mb_xy - 1;
  2041. + if( (mb_y&1) && mb_x > 0 && h->mb.field_decoding_flag == h->mb.field[left_xy] )
  2042. + left_xy += h->mb.i_mb_stride;
  2043. + if( h->mb.field_decoding_flag )
  2044. + {
  2045. + top_xy = mb_xy - h->mb.i_mb_stride;
  2046. + if( !(mb_y&1) && top_xy >= 0 && h->mb.slice_table[top_xy] == h->sh.i_first_mb && h->mb.field[top_xy] )
  2047. + top_xy -= h->mb.i_mb_stride;
  2048. + }
  2049. + else
  2050. + top_xy = mb_x + (mb_y-1)*h->mb.i_mb_stride;
  2051. + }
  2052. + else
  2053. + {
  2054. + left_xy = h->mb.i_mb_xy - 1;
  2055. + top_xy = h->mb.i_mb_xy - h->mb.i_mb_stride;
  2056. + }
  2057. +
  2058. + h->mb.left_skip = mb_x > 0 && h->mb.slice_table[left_xy] == h->sh.i_first_mb && !IS_SKIP( h->mb.type[left_xy] );
  2059. + h->mb.top_skip = top_xy >= 0 && (h->mb.slice_table[top_xy] == h->sh.i_first_mb) && !IS_SKIP( h->mb.type[top_xy] );
  2060. + }
  2061. +
  2062. /* load skip */
  2063. if( h->sh.i_type == SLICE_TYPE_B )
  2064. {
  2065. diff --git a/encoder/cabac.c b/encoder/cabac.c
  2066. index 6138d06..3435048 100644
  2067. --- a/encoder/cabac.c
  2068. +++ b/encoder/cabac.c
  2069. @@ -295,10 +295,10 @@ static void x264_cabac_mb_qp_delta( x264_t *h, x264_cabac_t *cb )
  2070. #if !RDO_SKIP_BS
  2071. void x264_cabac_mb_skip( x264_t *h, int b_skip )
  2072. {
  2073. - int ctx = ((h->mb.i_neighbour & MB_LEFT) && !IS_SKIP( h->mb.i_mb_type_left[0] ))
  2074. - + ((h->mb.i_neighbour & MB_TOP) && !IS_SKIP( h->mb.i_mb_type_top ))
  2075. - + (h->sh.i_type == SLICE_TYPE_P ? 11 : 24);
  2076. - x264_cabac_encode_decision( &h->cabac, ctx, b_skip );
  2077. + int ctx = h->mb.left_skip + h->mb.top_skip;
  2078. + if( h->sh.i_type != SLICE_TYPE_P )
  2079. + ctx += 13;
  2080. + x264_cabac_encode_decision( &h->cabac, 11+ctx, b_skip );
  2081. }
  2082. #endif
  2083.  
  2084. @@ -350,7 +350,7 @@ static void x264_cabac_mb_ref( x264_t *h, x264_cabac_t *cb, int i_list, int idx
  2085. const int i8 = x264_scan8[idx];
  2086. const int i_refa = h->mb.cache.ref[i_list][i8 - 1];
  2087. const int i_refb = h->mb.cache.ref[i_list][i8 - 8];
  2088. - int ctx = 0;
  2089. + int ctx = 0;
  2090.  
  2091. if( i_refa > 0 && !h->mb.cache.skip[i8 - 1] )
  2092. ctx++;
  2093. @@ -423,9 +423,9 @@ static ALWAYS_INLINE int x264_cabac_mb_mvd_cpn( x264_t *h, x264_cabac_t *cb, int
  2094. x264_cabac_encode_bypass( cb, mvd < 0 );
  2095. }
  2096. #endif
  2097. - /* Since we don't need to keep track of MVDs larger than 33, just cap the value.
  2098. + /* Since we don't need to keep track of MVDs larger than 70, just cap the value.
  2099. * This lets us store MVDs as 8-bit values instead of 16-bit. */
  2100. - return X264_MIN( i_abs, 33 );
  2101. + return X264_MIN( i_abs, 70 );
  2102. }
  2103.  
  2104. static NOINLINE uint16_t x264_cabac_mb_mvd( x264_t *h, x264_cabac_t *cb, int i_list, int idx, int width )
  2105. --
  2106. 1.7.4
  2107.  
  2108.  
  2109. From b49fa962fcd25984e5f65a36a2784fbe702bdb4c Mon Sep 17 00:00:00 2001
  2110. From: Simon Horlick <simonhorlick@gmail.com>
  2111. Date: Wed, 16 Mar 2011 21:18:59 +0000
  2112. Subject: [PATCH 14/25] Add mbaff deblock strength calculation
  2113.  
  2114. ---
  2115. common/deblock.c | 38 ++++++++++++++++++++++++++++++++++++++
  2116. common/frame.h | 3 +++
  2117. encoder/encoder.c | 12 +++++++++---
  2118. 3 files changed, 50 insertions(+), 3 deletions(-)
  2119.  
  2120. diff --git a/common/deblock.c b/common/deblock.c
  2121. index 52d410d..55a0154 100644
  2122. --- a/common/deblock.c
  2123. +++ b/common/deblock.c
  2124. @@ -269,6 +269,44 @@ static void deblock_strength_c( uint8_t nnz[X264_SCAN8_SIZE], int8_t ref[2][X264
  2125. }
  2126. }
  2127. }
  2128. +void deblock_strength_mbaff( x264_t *h, uint8_t nnz[X264_SCAN8_SIZE], int8_t ref[2][X264_SCAN8_LUMA_SIZE],
  2129. + int16_t mv[2][X264_SCAN8_LUMA_SIZE][2], uint8_t bs[2][4][4], int mvy_limit, int bframe )
  2130. +{
  2131. + const int vertical = 0, horizontal = 1;
  2132. + int neighbour_field[2];
  2133. + int current_field = h->mb.b_interlaced;
  2134. + neighbour_field[vertical] = h->mb.i_mb_left_xy[0] >= 0 && h->mb.field[h->mb.i_mb_left_xy[0]];
  2135. + neighbour_field[horizontal] = h->mb.i_mb_top_xy >= 0 && h->mb.field[h->mb.i_mb_top_xy];
  2136. +
  2137. + for( int dir = 0; dir < 2; dir++ )
  2138. + {
  2139. + int edge_stride = dir ? 8 : 1;
  2140. + int part_stride = dir ? 1 : 8;
  2141. + for( int edge = 0; edge < 4; edge++ )
  2142. + {
  2143. + for( int i = 0, q = X264_SCAN8_0+edge*edge_stride; i < 4; i++, q += part_stride )
  2144. + {
  2145. + int p = q - edge_stride;
  2146. + if( nnz[q] || nnz[p] )
  2147. + {
  2148. + bs[dir][edge][i] = 2;
  2149. + }
  2150. + else if( (edge == 0 && current_field != neighbour_field[dir]) ||
  2151. + ref[0][q] != ref[0][p] ||
  2152. + abs( mv[0][q][0] - mv[0][p][0] ) >= 4 ||
  2153. + abs( mv[0][q][1] - mv[0][p][1] ) >= mvy_limit ||
  2154. + (bframe && (ref[1][q] != ref[1][p] ||
  2155. + abs( mv[1][q][0] - mv[1][p][0] ) >= 4 ||
  2156. + abs( mv[1][q][1] - mv[1][p][1] ) >= mvy_limit )) )
  2157. + {
  2158. + bs[dir][edge][i] = 1;
  2159. + }
  2160. + else
  2161. + bs[dir][edge][i] = 0;
  2162. + }
  2163. + }
  2164. + }
  2165. +}
  2166.  
  2167. static inline void deblock_edge( x264_t *h, pixel *pix, int i_stride, uint8_t bS[4], int i_qp, int b_chroma, x264_deblock_inter_t pf_inter )
  2168. {
  2169. diff --git a/common/frame.h b/common/frame.h
  2170. index 8fe0627..3296a2c 100644
  2171. --- a/common/frame.h
  2172. +++ b/common/frame.h
  2173. @@ -186,6 +186,9 @@ typedef struct
  2174. int bframe );
  2175. } x264_deblock_function_t;
  2176.  
  2177. +void deblock_strength_mbaff( x264_t *h, uint8_t nnz[X264_SCAN8_SIZE], int8_t ref[2][X264_SCAN8_LUMA_SIZE],
  2178. + int16_t mv[2][X264_SCAN8_LUMA_SIZE][2], uint8_t bs[2][4][4], int mvy_limit, int bframe );
  2179. +
  2180. x264_frame_t *x264_frame_new( x264_t *h, int b_fdec );
  2181. void x264_frame_delete( x264_frame_t *frame );
  2182.  
  2183. diff --git a/encoder/encoder.c b/encoder/encoder.c
  2184. index 2525fec..0319126 100644
  2185. --- a/encoder/encoder.c
  2186. +++ b/encoder/encoder.c
  2187. @@ -2129,14 +2129,20 @@ static int x264_slice_write( x264_t *h )
  2188. /* calculate deblock strength values (actual deblocking is done per-row along with hpel) */
  2189. if( b_deblock )
  2190. {
  2191. - int mvy_limit = 4 >> h->sh.b_mbaff;
  2192. + int mvy_limit = 4 >> h->mb.b_interlaced;
  2193. uint8_t (*bs)[4][4] = h->deblock_strength[h->mb.i_mb_y&1][h->mb.i_mb_x];
  2194. x264_macroblock_cache_load_deblock( h );
  2195. if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
  2196. memset( bs, 3, 2*4*4*sizeof(uint8_t) );
  2197. else
  2198. - h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
  2199. - bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
  2200. + {
  2201. + if( h->sh.b_mbaff )
  2202. + deblock_strength_mbaff( h, h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
  2203. + bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
  2204. + else
  2205. + h->loopf.deblock_strength( h->mb.cache.non_zero_count, h->mb.cache.ref, h->mb.cache.mv,
  2206. + bs, mvy_limit, h->sh.i_type == SLICE_TYPE_B );
  2207. + }
  2208. }
  2209.  
  2210. x264_ratecontrol_mb( h, mb_size );
  2211. --
  2212. 1.7.4
  2213.  
  2214.  
  2215. From 55bb59fb940208de76113875c0744694043a5f4d Mon Sep 17 00:00:00 2001
  2216. From: Simon Horlick <simonhorlick@gmail.com>
  2217. Date: Wed, 16 Mar 2011 21:27:07 +0000
  2218. Subject: [PATCH 15/25] Initial deblocking support
  2219.  
  2220. ---
  2221. common/deblock.c | 13 +++++++--
  2222. common/macroblock.c | 67 ++++++++++++++++++++++++++++++++++----------------
  2223. 2 files changed, 55 insertions(+), 25 deletions(-)
  2224.  
  2225. diff --git a/common/deblock.c b/common/deblock.c
  2226. index 55a0154..48788ae 100644
  2227. --- a/common/deblock.c
  2228. +++ b/common/deblock.c
  2229. @@ -345,13 +345,17 @@ void x264_frame_deblock_row( x264_t *h, int mb_y )
  2230. int b_interlaced = h->sh.b_mbaff;
  2231. int qp_thresh = 15 - X264_MIN( h->sh.i_alpha_c0_offset, h->sh.i_beta_offset ) - X264_MAX( 0, h->param.analyse.i_chroma_qp_offset );
  2232. int stridey = h->fdec->i_stride[0];
  2233. - int stride2y = stridey << b_interlaced;
  2234. int strideuv = h->fdec->i_stride[1];
  2235. - int stride2uv = strideuv << b_interlaced;
  2236. +
  2237. + // Backup mb.b_interlaced because it will be changed in x264_macroblock_cache_load_neighbours_deblock.
  2238. + int interlaced_backup = h->mb.b_interlaced;
  2239.  
  2240. for( int mb_x = 0; mb_x < h->mb.i_mb_width; mb_x += (~b_interlaced | mb_y)&1, mb_y ^= b_interlaced )
  2241. {
  2242. x264_prefetch_fenc( h, h->fdec, mb_x, mb_y );
  2243. + h->mb.i_mb_x = mb_x;
  2244. + h->mb.i_mb_y = mb_y;
  2245. + h->mb.i_mb_xy = mb_x + h->mb.i_mb_stride*mb_y;
  2246. x264_macroblock_cache_load_neighbours_deblock( h, mb_x, mb_y );
  2247.  
  2248. int mb_xy = h->mb.i_mb_xy;
  2249. @@ -361,12 +365,14 @@ void x264_frame_deblock_row( x264_t *h, int mb_y )
  2250.  
  2251. pixel *pixy = h->fdec->plane[0] + 16*mb_y*stridey + 16*mb_x;
  2252. pixel *pixuv = h->fdec->plane[1] + 8*mb_y*strideuv + 16*mb_x;
  2253. - if( mb_y & b_interlaced )
  2254. + if( mb_y & h->mb.b_interlaced )
  2255. {
  2256. pixy -= 15*stridey;
  2257. pixuv -= 7*strideuv;
  2258. }
  2259.  
  2260. + int stride2y = stridey << h->mb.b_interlaced;
  2261. + int stride2uv = strideuv << h->mb.b_interlaced;
  2262. int qp = h->mb.qp[mb_xy];
  2263. int qpc = h->chroma_qp_table[qp];
  2264. int first_edge_only = h->mb.type[mb_xy] == P_SKIP || qp <= qp_thresh;
  2265. @@ -427,6 +433,7 @@ void x264_frame_deblock_row( x264_t *h, int mb_y )
  2266.  
  2267. #undef FILTER
  2268. }
  2269. + h->mb.b_interlaced = interlaced_backup;
  2270.  
  2271. // FIXME: Don't copy the whole frame around.
  2272. int y = mb_y*16;
  2273. diff --git a/common/macroblock.c b/common/macroblock.c
  2274. index c01320c..47bb2ff 100644
  2275. --- a/common/macroblock.c
  2276. +++ b/common/macroblock.c
  2277. @@ -1257,18 +1257,36 @@ void x264_macroblock_cache_load_neighbours_deblock( x264_t *h, int mb_x, int mb_
  2278. int deblock_on_slice_edges = h->sh.i_disable_deblocking_filter_idc != 2;
  2279.  
  2280. h->mb.i_neighbour = 0;
  2281. - h->mb.i_mb_xy = mb_y * h->mb.i_mb_stride + mb_x;
  2282. +
  2283. + if( h->sh.b_mbaff )
  2284. + h->mb.b_interlaced = h->mb.field[h->mb.i_mb_xy];
  2285. +
  2286. + h->mb.i_mb_top_xy = h->mb.i_mb_xy - (h->mb.i_mb_stride << h->mb.b_interlaced);
  2287. + h->mb.i_mb_left_xy[1] =
  2288. + h->mb.i_mb_left_xy[0] = h->mb.i_mb_xy - 1;
  2289. + if( h->sh.b_mbaff )
  2290. + {
  2291. + if( mb_y&1 )
  2292. + {
  2293. + if( h->mb.field[h->mb.i_mb_xy - 1] != h->mb.b_interlaced )
  2294. + h->mb.i_mb_left_xy[0] -= h->mb.i_mb_stride;
  2295. + }
  2296. + else
  2297. + {
  2298. + if( h->mb.b_interlaced && !h->mb.field[h->mb.i_mb_top_xy] )
  2299. + h->mb.i_mb_top_xy += h->mb.i_mb_stride;
  2300. + if( h->mb.field[h->mb.i_mb_xy - 1] != h->mb.b_interlaced )
  2301. + h->mb.i_mb_left_xy[1] += h->mb.i_mb_stride;
  2302. + }
  2303. + }
  2304.  
  2305. if( mb_x > 0 )
  2306. {
  2307. - h->mb.i_mb_left_xy[0] = h->mb.i_mb_xy - 1;
  2308. if( deblock_on_slice_edges || h->mb.slice_table[h->mb.i_mb_left_xy[0]] == h->mb.slice_table[h->mb.i_mb_xy] )
  2309. h->mb.i_neighbour |= MB_LEFT;
  2310. }
  2311. -
  2312. if( mb_y > h->mb.b_interlaced )
  2313. {
  2314. - h->mb.i_mb_top_xy = h->mb.i_mb_xy - (h->mb.i_mb_stride << h->mb.b_interlaced);
  2315. if( deblock_on_slice_edges || h->mb.slice_table[h->mb.i_mb_top_xy] == h->mb.slice_table[h->mb.i_mb_xy] )
  2316. h->mb.i_neighbour |= MB_TOP;
  2317. }
  2318. @@ -1276,7 +1294,7 @@ void x264_macroblock_cache_load_neighbours_deblock( x264_t *h, int mb_x, int mb_
  2319.  
  2320. void x264_macroblock_cache_load_deblock( x264_t *h )
  2321. {
  2322. - if( IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
  2323. + if( !h->sh.b_mbaff && IS_INTRA( h->mb.type[h->mb.i_mb_xy] ) )
  2324. return;
  2325.  
  2326. /* If we have multiple slices and we're deblocking on slice edges, we
  2327. @@ -1291,7 +1309,7 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  2328. h->mb.i_neighbour &= ~old_neighbour;
  2329. if( h->mb.i_neighbour )
  2330. {
  2331. - int top_y = mb_y - (1 << h->mb.b_interlaced);
  2332. + int top_y = h->mb.i_mb_top_xy / h->mb.i_mb_stride;
  2333. int top_8x8 = (2*top_y+1) * h->mb.i_b8_stride + 2*mb_x;
  2334. int top_4x4 = (4*top_y+3) * h->mb.i_b4_stride + 4*mb_x;
  2335. int s8x8 = h->mb.i_b8_stride;
  2336. @@ -1330,17 +1348,15 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  2337. i8 = x264_scan8[0] - 1;
  2338. if( h->mb.i_neighbour & MB_LEFT )
  2339. {
  2340. - int ir = h->mb.i_b8_xy - 1;
  2341. - int iv = h->mb.i_b4_xy - 1;
  2342. h->mb.cache.ref[l][i8+0*8] =
  2343. - h->mb.cache.ref[l][i8+1*8] = ref[ir + 0*s8x8];
  2344. + h->mb.cache.ref[l][i8+1*8] = ref[h->mb.left_b8[0] + 1 + s8x8*left_index_table[12+4]];
  2345. h->mb.cache.ref[l][i8+2*8] =
  2346. - h->mb.cache.ref[l][i8+3*8] = ref[ir + 1*s8x8];
  2347. + h->mb.cache.ref[l][i8+3*8] = ref[h->mb.left_b8[1] + 1 + s8x8*left_index_table[12+5]];
  2348.  
  2349. - CP32( h->mb.cache.mv[l][i8+0*8], mv[iv + 0*s4x4] );
  2350. - CP32( h->mb.cache.mv[l][i8+1*8], mv[iv + 1*s4x4] );
  2351. - CP32( h->mb.cache.mv[l][i8+2*8], mv[iv + 2*s4x4] );
  2352. - CP32( h->mb.cache.mv[l][i8+3*8], mv[iv + 3*s4x4] );
  2353. + CP32( h->mb.cache.mv[l][i8+0*8], mv[h->mb.left_b4[0] + 3 + s4x4*left_index_table[12+0]] );
  2354. + CP32( h->mb.cache.mv[l][i8+1*8], mv[h->mb.left_b4[0] + 3 + s4x4*left_index_table[12+1]] );
  2355. + CP32( h->mb.cache.mv[l][i8+2*8], mv[h->mb.left_b4[1] + 3 + s4x4*left_index_table[12+2]] );
  2356. + CP32( h->mb.cache.mv[l][i8+3*8], mv[h->mb.left_b4[1] + 3 + s4x4*left_index_table[12+3]] );
  2357. }
  2358. }
  2359. }
  2360. @@ -1380,7 +1396,7 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  2361. {
  2362. uint8_t (*nnz)[24] = h->mb.non_zero_count;
  2363. int top = h->mb.i_mb_top_xy;
  2364. - int left = h->mb.i_mb_left_xy[0];
  2365. + int *left = h->mb.i_mb_left_xy;
  2366.  
  2367. if( (h->mb.i_neighbour & MB_TOP) && h->mb.mb_transform_size[top] )
  2368. {
  2369. @@ -1391,15 +1407,22 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  2370. M16( &h->mb.cache.non_zero_count[i8+2] ) = nnz_top1 ? 0x0101 : 0;
  2371. }
  2372.  
  2373. - if( (h->mb.i_neighbour & MB_LEFT) && h->mb.mb_transform_size[left] )
  2374. + if( h->mb.i_neighbour & MB_LEFT )
  2375. {
  2376. + // TODO: Merge code in deblock
  2377. int i8 = x264_scan8[0] - 1;
  2378. - int nnz_left0 = M16( &nnz[left][2] ) | M16( &nnz[left][6] );
  2379. - int nnz_left1 = M16( &nnz[left][10] ) | M16( &nnz[left][14] );
  2380. - h->mb.cache.non_zero_count[i8+8*0] = !!nnz_left0;
  2381. - h->mb.cache.non_zero_count[i8+8*1] = !!nnz_left0;
  2382. - h->mb.cache.non_zero_count[i8+8*2] = !!nnz_left1;
  2383. - h->mb.cache.non_zero_count[i8+8*3] = !!nnz_left1;
  2384. + if( h->mb.mb_transform_size[left[0]] )
  2385. + {
  2386. + int nnz_left0 = M16( &nnz[left[0]][2] ) | M16( &nnz[left[0]][6] );
  2387. + h->mb.cache.non_zero_count[i8+8*0] = !!nnz_left0;
  2388. + h->mb.cache.non_zero_count[i8+8*1] = !!nnz_left0;
  2389. + }
  2390. + if( h->mb.mb_transform_size[left[1]] )
  2391. + {
  2392. + int nnz_left1 = M16( &nnz[left[1]][10] ) | M16( &nnz[left[1]][14] );
  2393. + h->mb.cache.non_zero_count[i8+8*2] = !!nnz_left1;
  2394. + h->mb.cache.non_zero_count[i8+8*3] = !!nnz_left1;
  2395. + }
  2396. }
  2397.  
  2398. if( h->mb.mb_transform_size[h->mb.i_mb_xy] )
  2399. --
  2400. 1.7.4
  2401.  
  2402.  
  2403. From c1420786e639c96f7fd0b11bac69952edeabbafc Mon Sep 17 00:00:00 2001
  2404. From: Simon Horlick <simonhorlick@gmail.com>
  2405. Date: Thu, 17 Mar 2011 18:15:06 +0000
  2406. Subject: [PATCH 16/25] Calculate deblock strength for mbaff
  2407.  
  2408. ---
  2409. common/common.h | 1 +
  2410. common/deblock.c | 1 +
  2411. common/macroblock.c | 5 +++++
  2412. 3 files changed, 7 insertions(+), 0 deletions(-)
  2413.  
  2414. diff --git a/common/common.h b/common/common.h
  2415. index 6d81496..7990cf0 100644
  2416. --- a/common/common.h
  2417. +++ b/common/common.h
  2418. @@ -844,6 +844,7 @@ struct x264_t
  2419. pixel *intra_border_backup[3][2]; /* bottom pixels of the previous mb row, used for intra prediction after the framebuffer has been deblocked */
  2420. pixel *intra_diagonal_backup[5][3];
  2421. uint8_t (*deblock_strength[2])[2][4][4];
  2422. + uint8_t (*deblock_strength_mbaff[2])[2][8]; /* store [field][mb_x][dir][partition], there can be 8 different block strengths in mbaff left/top edges */
  2423.  
  2424. /* CPU functions dependents */
  2425. x264_predict_t predict_16x16[4+3];
  2426. diff --git a/common/deblock.c b/common/deblock.c
  2427. index 48788ae..c1bb10a 100644
  2428. --- a/common/deblock.c
  2429. +++ b/common/deblock.c
  2430. @@ -362,6 +362,7 @@ void x264_frame_deblock_row( x264_t *h, int mb_y )
  2431. int transform_8x8 = h->mb.mb_transform_size[h->mb.i_mb_xy];
  2432. int intra_cur = IS_INTRA( h->mb.type[mb_xy] );
  2433. uint8_t (*bs)[4][4] = h->deblock_strength[mb_y&1][mb_x];
  2434. + uint8_t (*bs_mbaff)[8] = h->deblock_strength_mbaff[mb_y&1][mb_x];
  2435.  
  2436. pixel *pixy = h->fdec->plane[0] + 16*mb_y*stridey + 16*mb_x;
  2437. pixel *pixuv = h->fdec->plane[1] + 8*mb_y*strideuv + 16*mb_x;
  2438. diff --git a/common/macroblock.c b/common/macroblock.c
  2439. index 47bb2ff..3a4c873 100644
  2440. --- a/common/macroblock.c
  2441. +++ b/common/macroblock.c
  2442. @@ -340,7 +340,9 @@ int x264_macroblock_thread_allocate( x264_t *h, int b_lookahead )
  2443. for( int i = 0; i <= h->param.b_interlaced; i++ )
  2444. {
  2445. CHECKED_MALLOC( h->deblock_strength[i], sizeof(**h->deblock_strength) * h->mb.i_mb_width );
  2446. + CHECKED_MALLOC( h->deblock_strength_mbaff[i], sizeof(**h->deblock_strength_mbaff) * h->mb.i_mb_width );
  2447. h->deblock_strength[1] = h->deblock_strength[i];
  2448. + h->deblock_strength_mbaff[1] = h->deblock_strength_mbaff[i];
  2449. }
  2450. }
  2451.  
  2452. @@ -372,7 +374,10 @@ void x264_macroblock_thread_free( x264_t *h, int b_lookahead )
  2453. if( !b_lookahead )
  2454. {
  2455. for( int i = 0; i <= h->param.b_interlaced; i++ )
  2456. + {
  2457. x264_free( h->deblock_strength[i] );
  2458. + x264_free( h->deblock_strength_mbaff[i] );
  2459. + }
  2460. for( int i = 0; i <= 2*h->param.b_interlaced; i++ )
  2461. {
  2462. for( int j = 0; j < 2; j++ )
  2463. --
  2464. 1.7.4
  2465.  
  2466.  
  2467. From 6cda0dc785db073de2c496c93e412bdb54c37a2d Mon Sep 17 00:00:00 2001
  2468. From: Simon Horlick <simonhorlick@gmail.com>
  2469. Date: Wed, 16 Mar 2011 22:06:27 +0000
  2470. Subject: [PATCH 17/25] Left edge deblocking
  2471.  
  2472. ---
  2473. common/deblock.c | 181 ++++++++++++++++++++++++++++++++++++++++++++++++--
  2474. common/macroblock.c | 63 ++++++++++++++++++
  2475. 2 files changed, 236 insertions(+), 8 deletions(-)
  2476.  
  2477. diff --git a/common/deblock.c b/common/deblock.c
  2478. index c1bb10a..be96fc8 100644
  2479. --- a/common/deblock.c
  2480. +++ b/common/deblock.c
  2481. @@ -118,6 +118,128 @@ static inline void deblock_luma_c( pixel *pix, int xstride, int ystride, int alp
  2482. }
  2483. }
  2484. }
  2485. +
  2486. +static inline void deblock_v_luma_mbaff_c( pixel *pix, int stride, int alpha, int beta, int8_t *tc0 )
  2487. +{
  2488. + /* Only filter 8 pixels at a time... */
  2489. + for( int d = 0; d < 8; d++, pix += stride )
  2490. + {
  2491. + int p2 = pix[-3];
  2492. + int p1 = pix[-2];
  2493. + int p0 = pix[-1];
  2494. + int q0 = pix[ 0];
  2495. + int q1 = pix[ 1];
  2496. + int q2 = pix[ 2];
  2497. +
  2498. + if( abs( p0 - q0 ) < alpha && abs( p1 - p0 ) < beta && abs( q1 - q0 ) < beta )
  2499. + {
  2500. + int tc = tc0[d>>1];
  2501. + int delta;
  2502. + if( abs( p2 - p0 ) < beta )
  2503. + {
  2504. + if( tc0[d>>1] )
  2505. + pix[-2] = p1 + x264_clip3( (( p2 + ((p0 + q0 + 1) >> 1)) >> 1) - p1, -tc0[d>>1], tc0[d>>1] );
  2506. + tc++;
  2507. + }
  2508. + if( abs( q2 - q0 ) < beta )
  2509. + {
  2510. + if( tc0[d>>1] )
  2511. + pix[ 1] = q1 + x264_clip3( (( q2 + ((p0 + q0 + 1) >> 1)) >> 1) - q1, -tc0[d>>1], tc0[d>>1] );
  2512. + tc++;
  2513. + }
  2514. +
  2515. + delta = x264_clip3( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
  2516. + pix[-1] = x264_clip_pixel( p0 + delta ); /* p0' */
  2517. + pix[ 0] = x264_clip_pixel( q0 - delta ); /* q0' */
  2518. + }
  2519. + }
  2520. +}
  2521. +
  2522. +static inline void deblock_v_luma_intra_mbaff_c( pixel *pix, int stride, int alpha, int beta )
  2523. +{
  2524. + /* Only filter 8 pixels at a time... */
  2525. + for( int d = 0; d < 8; d++, pix += stride )
  2526. + {
  2527. + /* Filter intra samples normally */
  2528. + int p2 = pix[-3];
  2529. + int p1 = pix[-2];
  2530. + int p0 = pix[-1];
  2531. + int q0 = pix[ 0];
  2532. + int q1 = pix[ 1];
  2533. + int q2 = pix[ 2];
  2534. +
  2535. + if( abs( p0 - q0 ) < alpha && abs( p1 - p0 ) < beta && abs( q1 - q0 ) < beta )
  2536. + {
  2537. + if(abs( p0 - q0 ) < ((alpha >> 2) + 2) )
  2538. + {
  2539. + if( abs( p2 - p0 ) < beta ) /* p0', p1', p2' */
  2540. + {
  2541. + const int p3 = pix[-4];
  2542. + pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
  2543. + pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
  2544. + pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
  2545. + }
  2546. + else /* p0' */
  2547. + pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
  2548. + if( abs( q2 - q0 ) < beta ) /* q0', q1', q2' */
  2549. + {
  2550. + const int q3 = pix[3];
  2551. + pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
  2552. + pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
  2553. + pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
  2554. + }
  2555. + else /* q0' */
  2556. + pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
  2557. + }
  2558. + else /* p0', q0' */
  2559. + {
  2560. + pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
  2561. + pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
  2562. + }
  2563. + }
  2564. + }
  2565. +}
  2566. +
  2567. +static inline void deblock_v_chroma_mbaff_c( pixel *pix, int stride, int alpha, int beta, int8_t *tc0 )
  2568. +{
  2569. + const int xstride = 2;
  2570. + for( int i = 0; i < 4; i++, pix += stride )
  2571. + {
  2572. + /* We don't worry about p2 or q2 */
  2573. + int tc = tc0[i];
  2574. + int p1 = pix[xstride*-2];
  2575. + int p0 = pix[xstride*-1];
  2576. + int q0 = pix[xstride* 0];
  2577. + int q1 = pix[xstride* 1];
  2578. +
  2579. + if( abs( p0 - q0 ) < alpha && abs( p1 - p0 ) < beta && abs( q1 - q0 ) < beta )
  2580. + {
  2581. + int delta = x264_clip3( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
  2582. + pix[xstride*-1] = x264_clip_pixel( p0 + delta ); /* p0' */
  2583. + pix[xstride* 0] = x264_clip_pixel( q0 - delta ); /* q0' */
  2584. + }
  2585. + }
  2586. +}
  2587. +
  2588. +static inline void deblock_v_chroma_intra_mbaff_c( pixel *pix, int stride, int alpha, int beta )
  2589. +{
  2590. + const int xstride = 2;
  2591. + for( int i = 0; i < 4; i++, pix += stride )
  2592. + {
  2593. + int p1 = pix[xstride*-2];
  2594. + int p0 = pix[xstride*-1];
  2595. + int q0 = pix[xstride* 0];
  2596. + int q1 = pix[xstride* 1];
  2597. +
  2598. + if( abs( p0 - q0 ) < alpha && abs( p1 - p0 ) < beta && abs( q1 - q0 ) < beta )
  2599. + {
  2600. + /* p0', q0' */
  2601. + pix[xstride*-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
  2602. + pix[xstride* 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
  2603. + }
  2604. + }
  2605. +}
  2606. +
  2607. static void deblock_v_luma_c( pixel *pix, int stride, int alpha, int beta, int8_t *tc0 )
  2608. {
  2609. deblock_luma_c( pix, stride, 1, alpha, beta, tc0 );
  2610. @@ -392,16 +514,59 @@ void x264_frame_deblock_row( x264_t *h, int mb_y )
  2611.  
  2612. if( h->mb.i_neighbour & MB_LEFT )
  2613. {
  2614. - int qpl = h->mb.qp[h->mb.i_mb_left_xy[0]];
  2615. - int qp_left = (qp + qpl + 1) >> 1;
  2616. - int qpc_left = (h->chroma_qp_table[qp] + h->chroma_qp_table[qpl] + 1) >> 1;
  2617. - int intra_left = IS_INTRA( h->mb.type[h->mb.i_mb_left_xy[0]] );
  2618. - if( intra_cur || intra_left )
  2619. - FILTER( _intra, 0, 0, qp_left, qpc_left );
  2620. + if( b_interlaced && h->mb.field[h->mb.i_mb_left_xy[0]] != h->mb.b_interlaced )
  2621. + {
  2622. + int luma_qp[2];
  2623. + int chroma_qp[2];
  2624. + int left_qp[2];
  2625. + int current_qp = h->mb.qp[mb_xy];
  2626. + left_qp[0] = h->mb.qp[h->mb.i_mb_left_xy[0]];
  2627. + luma_qp[0] = (current_qp + left_qp[0] + 1) >> 1;
  2628. + chroma_qp[0] = (h->chroma_qp_table[current_qp] + h->chroma_qp_table[left_qp[0]] + 1) >> 1;
  2629. + if( bs_mbaff[0][0] == 4)
  2630. + {
  2631. + deblock_edge_intra( h, pixy, 2*stridey, bs_mbaff[0], luma_qp[0], 0, deblock_v_luma_intra_mbaff_c );
  2632. + deblock_edge_intra( h, pixuv, 2*strideuv, bs_mbaff[0], chroma_qp[0], 1, deblock_v_chroma_intra_mbaff_c );
  2633. + deblock_edge_intra( h, pixuv + 1, 2*strideuv, bs_mbaff[0], chroma_qp[0], 1, deblock_v_chroma_intra_mbaff_c );
  2634. + }
  2635. + else
  2636. + {
  2637. + deblock_edge( h, pixy, 2*stridey, bs_mbaff[0], luma_qp[0], 0, deblock_v_luma_mbaff_c );
  2638. + deblock_edge( h, pixuv, 2*strideuv, bs_mbaff[0], chroma_qp[0], 1, deblock_v_chroma_mbaff_c );
  2639. + deblock_edge( h, pixuv + 1, 2*strideuv, bs_mbaff[0], chroma_qp[0], 1, deblock_v_chroma_mbaff_c );
  2640. + }
  2641. +
  2642. + int offy = h->mb.b_interlaced ? 4 : 0;
  2643. + int offuv = h->mb.b_interlaced ? 3 : 0;
  2644. + left_qp[1] = h->mb.qp[h->mb.i_mb_left_xy[1]];
  2645. + luma_qp[1] = (current_qp + left_qp[1] + 1) >> 1;
  2646. + chroma_qp[1] = (h->chroma_qp_table[current_qp] + h->chroma_qp_table[left_qp[1]] + 1) >> 1;
  2647. + if( bs_mbaff[0][4] == 4)
  2648. + {
  2649. + deblock_edge_intra( h, pixy + (stridey<<offy), 2*stridey, bs_mbaff[0]+4, luma_qp[1], 0, deblock_v_luma_intra_mbaff_c );
  2650. + deblock_edge_intra( h, pixuv + (strideuv<<offuv), 2*strideuv, bs_mbaff[0]+4, chroma_qp[1], 1, deblock_v_chroma_intra_mbaff_c );
  2651. + deblock_edge_intra( h, pixuv + 1 + (strideuv<<offuv), 2*strideuv, bs_mbaff[0]+4, chroma_qp[1], 1, deblock_v_chroma_intra_mbaff_c );
  2652. + }
  2653. + else
  2654. + {
  2655. + deblock_edge( h, pixy + (stridey<<offy), 2*stridey, bs_mbaff[0]+4, luma_qp[1], 0, deblock_v_luma_mbaff_c );
  2656. + deblock_edge( h, pixuv + (strideuv<<offuv), 2*strideuv, bs_mbaff[0]+4, chroma_qp[1], 1, deblock_v_chroma_mbaff_c );
  2657. + deblock_edge( h, pixuv + 1 + (strideuv<<offuv), 2*strideuv, bs_mbaff[0]+4, chroma_qp[1], 1, deblock_v_chroma_mbaff_c );
  2658. + }
  2659. + }
  2660. else
  2661. - FILTER( , 0, 0, qp_left, qpc_left );
  2662. - }
  2663. + {
  2664. + int qpl = h->mb.qp[h->mb.i_mb_xy-1];
  2665. + int qp_left = (qp + qpl + 1) >> 1;
  2666. + int qpc_left = (h->chroma_qp_table[qp] + h->chroma_qp_table[qpl] + 1) >> 1;
  2667. + int intra_left = IS_INTRA( h->mb.type[h->mb.i_mb_xy-1] );
  2668.  
  2669. + if( intra_cur || intra_left ) // bs=4
  2670. + FILTER( _intra, 0, 0, qp_left, qpc_left );
  2671. + else
  2672. + FILTER( , 0, 0, qp_left, qpc_left );
  2673. + }
  2674. + }
  2675. if( !first_edge_only )
  2676. {
  2677. if( !transform_8x8 ) FILTER( , 0, 1, qp, qpc );
  2678. diff --git a/common/macroblock.c b/common/macroblock.c
  2679. index 3a4c873..1d72fe8 100644
  2680. --- a/common/macroblock.c
  2681. +++ b/common/macroblock.c
  2682. @@ -1445,6 +1445,69 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  2683. M32( &h->mb.cache.non_zero_count[x264_scan8[0]+8*3] ) = nnzbot;
  2684. }
  2685. }
  2686. +
  2687. + int mb_x = h->mb.i_mb_x;
  2688. + int mb_y = h->mb.i_mb_y;
  2689. + int mb_xy = h->mb.i_mb_xy;
  2690. +
  2691. + // left is wrong without this here
  2692. + x264_macroblock_cache_load_neighbours_deblock( h, mb_x, mb_y );
  2693. +
  2694. + uint8_t (*bs_mbaff)[8] = h->deblock_strength_mbaff[h->mb.i_mb_y&1][h->mb.i_mb_x];
  2695. +
  2696. + int intra_cur = IS_INTRA( h->mb.type[mb_xy] );
  2697. +
  2698. + if( h->mb.i_neighbour & MB_LEFT )
  2699. + {
  2700. + if( h->sh.b_mbaff && h->mb.field[h->mb.i_mb_left_xy[0]] != h->mb.b_interlaced )
  2701. + {
  2702. + static const uint8_t offset[2][2][8] = {
  2703. + { { 0, 0, 0, 0, 1, 1, 1, 1 },
  2704. + { 2, 2, 2, 2, 3, 3, 3, 3 }, },
  2705. + { { 0, 1, 2, 3, 0, 1, 2, 3 },
  2706. + { 0, 1, 2, 3, 0, 1, 2, 3 }, }
  2707. + };
  2708. + uint8_t bS[8];
  2709. +
  2710. + if( intra_cur )
  2711. + bS[0] = bS[1] = bS[2] = bS[3] =
  2712. + bS[4] = bS[5] = bS[6] = bS[7] = 4;
  2713. + else
  2714. + {
  2715. + const uint8_t *off = offset[h->mb.b_interlaced][mb_y&1];
  2716. + uint8_t (*nnz)[24] = h->mb.non_zero_count;
  2717. +
  2718. + for( int i=0; i<8; i++ )
  2719. + {
  2720. + int left = h->mb.i_mb_left_xy[h->mb.b_interlaced ? i>>2 : i&1];
  2721. + int nnz_this = h->mb.cache.non_zero_count[x264_scan8[0]+8*(i>>1)];
  2722. + int nnz_left = nnz[left][3 + 4*off[i]];
  2723. + if( !h->param.b_cabac && h->pps->b_transform_8x8_mode )
  2724. + {
  2725. + int j = off[i]&~1;
  2726. + if( h->mb.mb_transform_size[left] )
  2727. + nnz_left = !!(M16( &nnz[left][2+4*j] ) | M16( &nnz[left][2+4*(1+j)] ));
  2728. + }
  2729. + if( IS_INTRA( h->mb.type[left] ) )
  2730. + bS[i] = 4;
  2731. + else if( nnz_left || nnz_this )
  2732. + bS[i] = 2;
  2733. + else // As left is different interlaced.
  2734. + bS[i] = 1;
  2735. + }
  2736. + }
  2737. +
  2738. + if( h->mb.b_interlaced )
  2739. + {
  2740. + for( int i=0; i<8; i++ ) bs_mbaff[0][i] = bS[i];
  2741. + }
  2742. + else
  2743. + {
  2744. + for( int i=0; i<4; i++ ) bs_mbaff[0][i] = bS[2*i];
  2745. + for( int i=0; i<4; i++ ) bs_mbaff[0][i+4] = bS[1+2*i];
  2746. + }
  2747. + }
  2748. + }
  2749. }
  2750.  
  2751. static void ALWAYS_INLINE twiddle_topleft_pixel( pixel *dst, pixel *src, int b_interlaced )
  2752. --
  2753. 1.7.4
  2754.  
  2755.  
  2756. From 374cba76623f48c2084b89c471e5fdf7951dfc76 Mon Sep 17 00:00:00 2001
  2757. From: Simon Horlick <simonhorlick@gmail.com>
  2758. Date: Wed, 16 Mar 2011 21:24:42 +0000
  2759. Subject: [PATCH 18/25] Top edge deblocking
  2760.  
  2761. ---
  2762. common/deblock.c | 41 ++++++++++++++++++++++++++-------
  2763. common/macroblock.c | 63 +++++++++++++++++++++++++++++++++++++++++++++++++++
  2764. 2 files changed, 95 insertions(+), 9 deletions(-)
  2765.  
  2766. diff --git a/common/deblock.c b/common/deblock.c
  2767. index be96fc8..de9d9fb 100644
  2768. --- a/common/deblock.c
  2769. +++ b/common/deblock.c
  2770. @@ -576,17 +576,40 @@ void x264_frame_deblock_row( x264_t *h, int mb_y )
  2771.  
  2772. if( h->mb.i_neighbour & MB_TOP )
  2773. {
  2774. - int qpt = h->mb.qp[h->mb.i_mb_top_xy];
  2775. - int qp_top = (qp + qpt + 1) >> 1;
  2776. - int qpc_top = (h->chroma_qp_table[qp] + h->chroma_qp_table[qpt] + 1) >> 1;
  2777. - int intra_top = IS_INTRA( h->mb.type[h->mb.i_mb_top_xy] );
  2778. - if( ~b_interlaced & (intra_cur | intra_top) )
  2779. - FILTER( _intra, 1, 0, qp_top, qpc_top );
  2780. + if( b_interlaced && !(mb_y&1) && !h->mb.b_interlaced && h->mb.field[h->mb.i_mb_top_xy] )
  2781. + {
  2782. + /* Need to filter both fields (even for frame macroblocks) */
  2783. + /* Filter top two rows using the top and then bottom macroblocks of the above pair. */
  2784. + int mbn_xy = mb_xy - 2 * h->mb.i_mb_stride;
  2785. +
  2786. + for(int j=0; j<2; j++, mbn_xy += h->mb.i_mb_stride)
  2787. + {
  2788. + int qpt = h->mb.qp[mbn_xy];
  2789. + int qp_top = (qp + qpt + 1) >> 1;
  2790. + int qpc_top = (h->chroma_qp_table[qp] + h->chroma_qp_table[qpt] + 1) >> 1;
  2791. +
  2792. + deblock_edge( h, pixy + j*stridey, 2* stridey, bs_mbaff[1]+4*j, qp_top, 0, deblock_v_luma_c );
  2793. + deblock_edge( h, pixuv + j*strideuv, 2*strideuv, bs_mbaff[1]+4*j, qpc_top, 1, deblock_v_chroma_c );
  2794. + }
  2795. + }
  2796. else
  2797. {
  2798. - if( intra_top )
  2799. - M32( bs[1][0] ) = 0x03030303;
  2800. - FILTER( , 1, 0, qp_top, qpc_top );
  2801. + int qpt = h->mb.qp[h->mb.i_mb_top_xy];
  2802. + int qp_top = (qp + qpt + 1) >> 1;
  2803. + int qpc_top = (h->chroma_qp_table[qp] + h->chroma_qp_table[qpt] + 1) >> 1;
  2804. + int intra_top = IS_INTRA( h->mb.type[h->mb.i_mb_top_xy] );
  2805. +
  2806. + if( (!b_interlaced || (!h->mb.b_interlaced && !h->mb.field[h->mb.i_mb_top_xy]))
  2807. + && (intra_cur || intra_top) )
  2808. + {
  2809. + FILTER( _intra, 1, 0, qp_top, qpc_top );
  2810. + }
  2811. + else
  2812. + {
  2813. + if( intra_top )
  2814. + M32( bs[1][0] ) = 0x03030303;
  2815. + FILTER( , 1, 0, qp_top, qpc_top );
  2816. + }
  2817. }
  2818. }
  2819.  
  2820. diff --git a/common/macroblock.c b/common/macroblock.c
  2821. index 1d72fe8..0da5958 100644
  2822. --- a/common/macroblock.c
  2823. +++ b/common/macroblock.c
  2824. @@ -1508,6 +1508,69 @@ void x264_macroblock_cache_load_deblock( x264_t *h )
  2825. }
  2826. }
  2827. }
  2828. +
  2829. + if( h->mb.i_neighbour & MB_TOP )
  2830. + {
  2831. + if( h->sh.b_mbaff && !(mb_y&1) && !h->mb.b_interlaced && h->mb.field[h->mb.i_mb_top_xy] )
  2832. + {
  2833. + /* Need to filter both fields (even for frame macroblocks) */
  2834. + /* Filter top two rows using the top and then bottom macroblocks of the above pair. */
  2835. + int mbn_xy = mb_xy - 2 * h->mb.i_mb_stride;
  2836. +
  2837. + for(int j=0; j<2; j++, mbn_xy += h->mb.i_mb_stride)
  2838. + {
  2839. + int mbn_intra = IS_INTRA( h->mb.type[mbn_xy] );
  2840. + uint8_t (*nnz)[24] = h->mb.non_zero_count;
  2841. +
  2842. + uint32_t nnz_top[4];
  2843. + uint32_t nnz_cur[4];
  2844. + nnz_top[0] = nnz[mbn_xy][3*4+0];
  2845. + nnz_top[1] = nnz[mbn_xy][3*4+1];
  2846. + nnz_top[2] = nnz[mbn_xy][3*4+2];
  2847. + nnz_top[3] = nnz[mbn_xy][3*4+3];
  2848. + nnz_cur[0] = h->mb.cache.non_zero_count[x264_scan8[0]+0];
  2849. + nnz_cur[1] = h->mb.cache.non_zero_count[x264_scan8[0]+1];
  2850. + nnz_cur[2] = h->mb.cache.non_zero_count[x264_scan8[0]+2];
  2851. + nnz_cur[3] = h->mb.cache.non_zero_count[x264_scan8[0]+3];
  2852. +
  2853. + /* Munge NNZ for cavlc + 8x8dct */
  2854. + if( !h->param.b_cabac && h->pps->b_transform_8x8_mode )
  2855. + {
  2856. + int top = mbn_xy;
  2857. + if( (h->mb.i_neighbour & MB_TOP) && h->mb.mb_transform_size[top] )
  2858. + {
  2859. + int nnz_top0 = M16( &nnz[top][8] ) | M16( &nnz[top][12] );
  2860. + int nnz_top1 = M16( &nnz[top][10] ) | M16( &nnz[top][14] );
  2861. + nnz_top[0] = nnz_top[1] = nnz_top0 ? 0x0101 : 0;
  2862. + nnz_top[2] = nnz_top[3] = nnz_top1 ? 0x0101 : 0;
  2863. + }
  2864. + if( h->mb.mb_transform_size[h->mb.i_mb_xy] )
  2865. + {
  2866. + int nnz0 = M16( &h->mb.cache.non_zero_count[x264_scan8[ 0]] ) | M16( &h->mb.cache.non_zero_count[x264_scan8[ 2]] );
  2867. + int nnz1 = M16( &h->mb.cache.non_zero_count[x264_scan8[ 4]] ) | M16( &h->mb.cache.non_zero_count[x264_scan8[ 6]] );
  2868. + nnz_cur[0] = nnz_cur[1] = !!nnz0;
  2869. + nnz_cur[2] = nnz_cur[3] = !!nnz1;
  2870. + }
  2871. + }
  2872. +
  2873. + uint8_t bS[4];
  2874. + if( intra_cur || mbn_intra )
  2875. + bS[0] = bS[1] = bS[2] = bS[3] = 3;
  2876. + else
  2877. + {
  2878. + for( int i = 0; i < 4; i++ )
  2879. + {
  2880. + if( nnz_cur[i] || nnz_top[i] )
  2881. + bS[i] = 2;
  2882. + else
  2883. + bS[i] = 1;
  2884. + }
  2885. + }
  2886. + for( int i=0; i<4; i++ )
  2887. + bs_mbaff[1][i+4*j] = bS[i];
  2888. + }
  2889. + }
  2890. + }
  2891. }
  2892.  
  2893. static void ALWAYS_INLINE twiddle_topleft_pixel( pixel *dst, pixel *src, int b_interlaced )
  2894. --
  2895. 1.7.4
  2896.  
  2897.  
  2898. From 388145f81f75bfcd549335774457ad7d7fcd475f Mon Sep 17 00:00:00 2001
  2899. From: Simon Horlick <simonhorlick@gmail.com>
  2900. Date: Tue, 15 Mar 2011 01:39:49 +0000
  2901. Subject: [PATCH 19/25] Use both left macroblocks for ref_idx calculation
  2902.  
  2903. ---
  2904. common/macroblock.c | 58 ++++++++++++++++++++++++++++++++++++++++++++------
  2905. 1 files changed, 51 insertions(+), 7 deletions(-)
  2906.  
  2907. diff --git a/common/macroblock.c b/common/macroblock.c
  2908. index 0da5958..5b42966 100644
  2909. --- a/common/macroblock.c
  2910. +++ b/common/macroblock.c
  2911. @@ -1229,14 +1229,58 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  2912. h->mb.dist_scale_factor = h->mb.dist_scale_factor_buf[h->mb.b_interlaced&(mb_y&1)];
  2913. if( h->param.b_cabac )
  2914. {
  2915. - uint8_t skipbp;
  2916. x264_macroblock_cache_skip( h, 0, 0, 4, 4, 0 );
  2917. - skipbp = (h->mb.i_neighbour & MB_LEFT) ? h->mb.skipbp[left[0]] : 0;
  2918. - h->mb.cache.skip[x264_scan8[0] - 1] = skipbp & 0x2;
  2919. - h->mb.cache.skip[x264_scan8[8] - 1] = skipbp & 0x8;
  2920. - skipbp = (h->mb.i_neighbour & MB_TOP) ? h->mb.skipbp[top] : 0;
  2921. - h->mb.cache.skip[x264_scan8[0] - 8] = skipbp & 0x4;
  2922. - h->mb.cache.skip[x264_scan8[4] - 8] = skipbp & 0x8;
  2923. +
  2924. + if( h->mb.i_neighbour & MB_LEFT )
  2925. + {
  2926. + if( h->mb.skipbp[left[0]] == 0xf )
  2927. + h->mb.cache.skip[x264_scan8[0] - 1] = 1;
  2928. + else if( h->mb.partition[left[0]] == D_8x8 )
  2929. + {
  2930. + int off = 1+(left_index_table[18]&~1);
  2931. + h->mb.cache.skip[x264_scan8[0] - 1] = (h->mb.skipbp[left[0]] >> off) & 1;
  2932. + }
  2933. + else
  2934. + h->mb.cache.skip[x264_scan8[0] - 1] = h->mb.skipbp[left[0]] & 0x2;
  2935. +
  2936. + if( h->mb.skipbp[left[1]] == 0xf )
  2937. + h->mb.cache.skip[x264_scan8[8] - 1] = 1;
  2938. + else if( h->mb.partition[left[1]] == D_8x8 )
  2939. + {
  2940. + int off = 1+(left_index_table[20]&~1);
  2941. + h->mb.cache.skip[x264_scan8[8] - 1] = (h->mb.skipbp[left[1]] >> off) & 1;
  2942. + }
  2943. + else
  2944. + h->mb.cache.skip[x264_scan8[8] - 1] = h->mb.skipbp[left[1]] & 0x8;
  2945. + }
  2946. + else
  2947. + {
  2948. + h->mb.cache.skip[x264_scan8[0] - 1 + 0*8] = 0;
  2949. + h->mb.cache.skip[x264_scan8[0] - 1 + 1*8] = 0;
  2950. + h->mb.cache.skip[x264_scan8[0] - 1 + 2*8] = 0;
  2951. + h->mb.cache.skip[x264_scan8[0] - 1 + 3*8] = 0;
  2952. + }
  2953. +
  2954. + if( h->mb.i_neighbour & MB_TOP )
  2955. + {
  2956. + if( h->mb.skipbp[top] == 0xf )
  2957. + {
  2958. + h->mb.cache.skip[x264_scan8[0] - 8] = 1;
  2959. + h->mb.cache.skip[x264_scan8[4] - 8] = 1;
  2960. + }
  2961. + else if( h->mb.partition[top] == D_8x8 )
  2962. + {
  2963. + h->mb.cache.skip[x264_scan8[0] - 8] = h->mb.skipbp[top] & 0x4;
  2964. + h->mb.cache.skip[x264_scan8[4] - 8] = h->mb.skipbp[top] & 0x8;
  2965. + }
  2966. + else
  2967. + {
  2968. + h->mb.cache.skip[x264_scan8[0] - 8] = 0;
  2969. + h->mb.cache.skip[x264_scan8[4] - 8] = 0;
  2970. + }
  2971. + }
  2972. + else
  2973. + M32( &h->mb.cache.skip[x264_scan8[0] - 8] ) = 0;
  2974. }
  2975. }
  2976.  
  2977. --
  2978. 1.7.4
  2979.  
  2980.  
  2981. From e30f270b1830c3fc2b12014aa85d6dd959e3c4d3 Mon Sep 17 00:00:00 2001
  2982. From: Simon Horlick <simonhorlick@gmail.com>
  2983. Date: Tue, 15 Mar 2011 01:14:16 +0000
  2984. Subject: [PATCH 20/25] Fix min/max mv calculation
  2985.  
  2986. ---
  2987. common/macroblock.c | 5 +++++
  2988. encoder/analyse.c | 12 ++++++++----
  2989. 2 files changed, 13 insertions(+), 4 deletions(-)
  2990.  
  2991. diff --git a/common/macroblock.c b/common/macroblock.c
  2992. index 5b42966..00e9403 100644
  2993. --- a/common/macroblock.c
  2994. +++ b/common/macroblock.c
  2995. @@ -841,6 +841,11 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  2996.  
  2997. const int *left_index_table = h->mb.left_index_table;
  2998.  
  2999. + int my = h->mb.i_mb_y >> h->mb.b_interlaced;
  3000. + int mb_height = h->mb.i_mb_height >> h->mb.b_interlaced;
  3001. + h->mb.mv_min[1] = 4*( -16*my - 24 );
  3002. + h->mb.mv_max[1] = 4*( 16*( mb_height - my - 1 ) + 24 );
  3003. +
  3004. /* load cache */
  3005. if( h->mb.i_neighbour & MB_TOP )
  3006. {
  3007. diff --git a/encoder/analyse.c b/encoder/analyse.c
  3008. index 4f439d4..d004c66 100644
  3009. --- a/encoder/analyse.c
  3010. +++ b/encoder/analyse.c
  3011. @@ -443,13 +443,11 @@ static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int qp )
  3012. h->mb.mv_max_fpel[0] = (h->mb.mv_max_spel[0]>>2) - i_fpel_border;
  3013. if( h->mb.i_mb_x == 0 )
  3014. {
  3015. - int mb_y = h->mb.i_mb_y >> h->sh.b_mbaff;
  3016. - int mb_height = h->sps->i_mb_height >> h->sh.b_mbaff;
  3017. int thread_mvy_range = i_fmv_range;
  3018.  
  3019. if( h->i_thread_frames > 1 )
  3020. {
  3021. - int pix_y = (h->mb.i_mb_y | h->mb.b_interlaced) * 16;
  3022. + int pix_y = (h->mb.i_mb_y | h->sh.b_mbaff) * 16;
  3023. int thresh = pix_y + h->param.analyse.i_mv_range_thread;
  3024. for( int i = (h->sh.i_type == SLICE_TYPE_B); i >= 0; i-- )
  3025. for( int j = 0; j < h->i_ref[i]; j++ )
  3026. @@ -465,7 +463,8 @@ static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int qp )
  3027.  
  3028. x264_analyse_weight_frame( h, pix_y + thread_mvy_range );
  3029. }
  3030. -
  3031. + int mb_y = h->mb.i_mb_y >> h->sh.b_mbaff;
  3032. + int mb_height = h->sps->i_mb_height >> h->sh.b_mbaff;
  3033. h->mb.mv_min[1] = 4*( -16*mb_y - 24 );
  3034. h->mb.mv_max[1] = 4*( 16*( mb_height - mb_y - 1 ) + 24 );
  3035. h->mb.mv_min_spel[1] = x264_clip3( h->mb.mv_min[1], -i_fmv_range, i_fmv_range );
  3036. @@ -473,6 +472,11 @@ static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int qp )
  3037. h->mb.mv_max_spel[1] = X264_MIN( h->mb.mv_max_spel[1], thread_mvy_range*4 );
  3038. h->mb.mv_min_fpel[1] = (h->mb.mv_min_spel[1]>>2) + i_fpel_border;
  3039. h->mb.mv_max_fpel[1] = (h->mb.mv_max_spel[1]>>2) - i_fpel_border;
  3040. +
  3041. + mb_y = h->mb.i_mb_y >> h->mb.b_interlaced;
  3042. + mb_height = h->sps->i_mb_height >> h->mb.b_interlaced;
  3043. + h->mb.mv_min[1] = 4*( -16*mb_y - 24 );
  3044. + h->mb.mv_max[1] = 4*( 16*( mb_height - mb_y - 1 ) + 24 );
  3045. }
  3046. #undef CLIP_FMV
  3047.  
  3048. --
  3049. 1.7.4
  3050.  
  3051.  
  3052. From 962a6ce9a008d7239c11920635e570903bed7a85 Mon Sep 17 00:00:00 2001
  3053. From: Simon Horlick <simonhorlick@gmail.com>
  3054. Date: Tue, 15 Mar 2011 01:15:06 +0000
  3055. Subject: [PATCH 21/25] Calculate bipred POCs correctly
  3056.  
  3057. ---
  3058. common/common.h | 4 +-
  3059. common/macroblock.c | 67 ++++++++++++++++++++++++++-------------------------
  3060. 2 files changed, 36 insertions(+), 35 deletions(-)
  3061.  
  3062. diff --git a/common/common.h b/common/common.h
  3063. index 7990cf0..c7670e7 100644
  3064. --- a/common/common.h
  3065. +++ b/common/common.h
  3066. @@ -756,9 +756,9 @@ struct x264_t
  3067. int i_chroma_lambda2_offset;
  3068.  
  3069. /* B_direct and weighted prediction */
  3070. - int16_t dist_scale_factor_buf[2][X264_REF_MAX*2][4];
  3071. + int16_t dist_scale_factor_buf[2][2][X264_REF_MAX*2][4];
  3072. int16_t (*dist_scale_factor)[4];
  3073. - int8_t bipred_weight_buf[2][X264_REF_MAX*2][4];
  3074. + int8_t bipred_weight_buf[2][2][X264_REF_MAX*2][4];
  3075. int8_t (*bipred_weight)[4];
  3076. /* maps fref1[0]'s ref indices into the current list0 */
  3077. #define map_col_to_list0(col) h->mb.map_col_to_list0[(col)+2]
  3078. diff --git a/common/macroblock.c b/common/macroblock.c
  3079. index 00e9403..f6406fe 100644
  3080. --- a/common/macroblock.c
  3081. +++ b/common/macroblock.c
  3082. @@ -1230,8 +1230,8 @@ void x264_macroblock_cache_load( x264_t *h, int mb_x, int mb_y )
  3083. /* load skip */
  3084. if( h->sh.i_type == SLICE_TYPE_B )
  3085. {
  3086. - h->mb.bipred_weight = h->mb.bipred_weight_buf[h->mb.b_interlaced&(mb_y&1)];
  3087. - h->mb.dist_scale_factor = h->mb.dist_scale_factor_buf[h->mb.b_interlaced&(mb_y&1)];
  3088. + h->mb.bipred_weight = h->mb.bipred_weight_buf[h->mb.b_interlaced][h->mb.b_interlaced&(mb_y&1)];
  3089. + h->mb.dist_scale_factor = h->mb.dist_scale_factor_buf[h->mb.b_interlaced][h->mb.b_interlaced&(mb_y&1)];
  3090. if( h->param.b_cabac )
  3091. {
  3092. x264_macroblock_cache_skip( h, 0, 0, 4, 4, 0 );
  3093. @@ -1903,42 +1903,43 @@ void x264_macroblock_cache_save( x264_t *h )
  3094.  
  3095. void x264_macroblock_bipred_init( x264_t *h )
  3096. {
  3097. - for( int field = 0; field <= h->sh.b_mbaff; field++ )
  3098. - for( int i_ref0 = 0; i_ref0 < (h->i_ref[0]<<h->sh.b_mbaff); i_ref0++ )
  3099. - {
  3100. - x264_frame_t *l0 = h->fref[0][i_ref0>>h->sh.b_mbaff];
  3101. - int poc0 = l0->i_poc + l0->i_delta_poc[field^(i_ref0&1)];
  3102. - for( int i_ref1 = 0; i_ref1 < (h->i_ref[1]<<h->sh.b_mbaff); i_ref1++ )
  3103. + for( int mbfield = 0; mbfield <= h->sh.b_mbaff; mbfield++ )
  3104. + for( int field = 0; field <= h->sh.b_mbaff; field++ )
  3105. + for( int i_ref0 = 0; i_ref0 < (h->i_ref[0]<<mbfield); i_ref0++ )
  3106. {
  3107. - int dist_scale_factor;
  3108. - x264_frame_t *l1 = h->fref[1][i_ref1>>h->sh.b_mbaff];
  3109. - int poc1 = l1->i_poc + l1->i_delta_poc[field^(i_ref1&1)];
  3110. - int cur_poc = h->fdec->i_poc + h->fdec->i_delta_poc[field];
  3111. - int td = x264_clip3( poc1 - poc0, -128, 127 );
  3112. - if( td == 0 /* || pic0 is a long-term ref */ )
  3113. - dist_scale_factor = 256;
  3114. - else
  3115. + x264_frame_t *l0 = h->fref[0][i_ref0>>mbfield];
  3116. + int poc0 = l0->i_poc + mbfield*l0->i_delta_poc[field^(i_ref0&1)];
  3117. + for( int i_ref1 = 0; i_ref1 < (h->i_ref[1]<<mbfield); i_ref1++ )
  3118. {
  3119. - int tb = x264_clip3( cur_poc - poc0, -128, 127 );
  3120. - int tx = (16384 + (abs(td) >> 1)) / td;
  3121. - dist_scale_factor = x264_clip3( (tb * tx + 32) >> 6, -1024, 1023 );
  3122. - }
  3123. + int dist_scale_factor;
  3124. + x264_frame_t *l1 = h->fref[1][i_ref1>>mbfield];
  3125. + int cur_poc = h->fdec->i_poc + h->fdec->i_delta_poc[field];
  3126. + int poc1 = l1->i_poc + mbfield*l1->i_delta_poc[field^(i_ref1&1)];
  3127. + int td = x264_clip3( poc1 - poc0, -128, 127 );
  3128. + if( td == 0 /* || pic0 is a long-term ref */ )
  3129. + dist_scale_factor = 256;
  3130. + else
  3131. + {
  3132. + int tb = x264_clip3( cur_poc - poc0, -128, 127 );
  3133. + int tx = (16384 + (abs(td) >> 1)) / td;
  3134. + dist_scale_factor = x264_clip3( (tb * tx + 32) >> 6, -1024, 1023 );
  3135. + }
  3136.  
  3137. - h->mb.dist_scale_factor_buf[field][i_ref0][i_ref1] = dist_scale_factor;
  3138. + h->mb.dist_scale_factor_buf[mbfield][field][i_ref0][i_ref1] = dist_scale_factor;
  3139.  
  3140. - dist_scale_factor >>= 2;
  3141. - if( h->param.analyse.b_weighted_bipred
  3142. - && dist_scale_factor >= -64
  3143. - && dist_scale_factor <= 128 )
  3144. - {
  3145. - h->mb.bipred_weight_buf[field][i_ref0][i_ref1] = 64 - dist_scale_factor;
  3146. - // ssse3 implementation of biweight doesn't support the extrema.
  3147. - // if we ever generate them, we'll have to drop that optimization.
  3148. - assert( dist_scale_factor >= -63 && dist_scale_factor <= 127 );
  3149. + dist_scale_factor >>= 2;
  3150. + if( h->param.analyse.b_weighted_bipred
  3151. + && dist_scale_factor >= -64
  3152. + && dist_scale_factor <= 128 )
  3153. + {
  3154. + h->mb.bipred_weight_buf[mbfield][field][i_ref0][i_ref1] = 64 - dist_scale_factor;
  3155. + // ssse3 implementation of biweight doesn't support the extrema.
  3156. + // if we ever generate them, we'll have to drop that optimization.
  3157. + assert( dist_scale_factor >= -63 && dist_scale_factor <= 127 );
  3158. + }
  3159. + else
  3160. + h->mb.bipred_weight_buf[mbfield][field][i_ref0][i_ref1] = 32;
  3161. }
  3162. - else
  3163. - h->mb.bipred_weight_buf[field][i_ref0][i_ref1] = 32;
  3164. }
  3165. - }
  3166. }
  3167.  
  3168. --
  3169. 1.7.4
  3170.  
  3171.  
  3172. From a9f5b05e3dc0c7482e6661a1ee8d457f1e35ee75 Mon Sep 17 00:00:00 2001
  3173. From: Simon Horlick <simonhorlick@gmail.com>
  3174. Date: Tue, 15 Mar 2011 01:16:20 +0000
  3175. Subject: [PATCH 22/25] Direct temporal
  3176.  
  3177. ---
  3178. common/mvpred.c | 100 +++++++++++++++++++++++++++++++++++++++++--------------
  3179. 1 files changed, 75 insertions(+), 25 deletions(-)
  3180.  
  3181. diff --git a/common/mvpred.c b/common/mvpred.c
  3182. index 278e0ac..28eabca 100644
  3183. --- a/common/mvpred.c
  3184. +++ b/common/mvpred.c
  3185. @@ -182,50 +182,100 @@ void x264_mb_predict_mv_pskip( x264_t *h, int16_t mv[2] )
  3186.  
  3187. static int x264_mb_predict_mv_direct16x16_temporal( x264_t *h )
  3188. {
  3189. - int i_mb_4x4 = 16 * h->mb.i_mb_stride * h->mb.i_mb_y + 4 * h->mb.i_mb_x;
  3190. - int i_mb_8x8 = 4 * h->mb.i_mb_stride * h->mb.i_mb_y + 2 * h->mb.i_mb_x;
  3191. - const int type_col = h->fref[1][0]->mb_type[h->mb.i_mb_xy];
  3192. - const int partition_col = h->fref[1][0]->mb_partition[h->mb.i_mb_xy];
  3193. -
  3194. - x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
  3195. -
  3196. - h->mb.i_partition = partition_col;
  3197. -
  3198. - if( IS_INTRA( type_col ) )
  3199. + int mb_x = h->mb.i_mb_x;
  3200. + int mb_y = h->mb.i_mb_y;
  3201. + int mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
  3202. + int type_col[2] = { h->fref[1][0]->mb_type[mb_xy], h->fref[1][0]->mb_type[mb_xy] };
  3203. + int partition_col[2] = { h->fref[1][0]->mb_partition[mb_xy], h->fref[1][0]->mb_partition[mb_xy] };
  3204. + int col_parity = abs(h->fref[1][0]->i_poc - h->fdec->i_poc)
  3205. + >= abs(h->fref[1][0]->i_poc + h->sh.i_delta_poc_bottom - h->fdec->i_poc);
  3206. + int preshift = h->mb.b_interlaced;
  3207. + int postshift = h->mb.b_interlaced;
  3208. + int offset = 1;
  3209. + int yshift = 1;
  3210. + h->mb.i_partition = partition_col[0];
  3211. + if( h->param.b_interlaced && h->fref[1][0]->field[mb_xy] != h->mb.b_interlaced )
  3212. {
  3213. - x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
  3214. - x264_macroblock_cache_mv( h, 0, 0, 4, 4, 0, 0 );
  3215. - x264_macroblock_cache_mv( h, 0, 0, 4, 4, 1, 0 );
  3216. - return 1;
  3217. + if( h->mb.b_interlaced )
  3218. + {
  3219. + mb_y = h->mb.i_mb_y&~1;
  3220. + mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
  3221. + type_col[0] = h->fref[1][0]->mb_type[mb_xy];
  3222. + type_col[1] = h->fref[1][0]->mb_type[mb_xy + h->mb.i_mb_stride];
  3223. + partition_col[0] = h->fref[1][0]->mb_partition[mb_xy];
  3224. + partition_col[1] = h->fref[1][0]->mb_partition[mb_xy + h->mb.i_mb_stride];
  3225. + preshift = 0;
  3226. + yshift = 0;
  3227. +
  3228. + if( (IS_INTRA(type_col[0]) || partition_col[0] == D_16x16)
  3229. + && (IS_INTRA(type_col[1]) || partition_col[1] == D_16x16)
  3230. + && partition_col[0] != D_8x8 )
  3231. + h->mb.i_partition = D_16x8;
  3232. + else
  3233. + h->mb.i_partition = D_8x8;
  3234. + }
  3235. + else
  3236. + {
  3237. + mb_y = (h->mb.i_mb_y&~1) + col_parity;
  3238. + mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
  3239. + type_col[0] = type_col[1] = h->fref[1][0]->mb_type[mb_xy];
  3240. + partition_col[0] = partition_col[1] = h->fref[1][0]->mb_partition[mb_xy];
  3241. + preshift = 1;
  3242. + yshift = 2;
  3243. + h->mb.i_partition = partition_col[0];
  3244. + }
  3245. + offset = 0;
  3246. }
  3247. + int i_mb_4x4 = 16 * h->mb.i_mb_stride * mb_y + 4 * mb_x;
  3248. + int i_mb_8x8 = 4 * h->mb.i_mb_stride * mb_y + 2 * mb_x;
  3249. +
  3250. + x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, 0 );
  3251.  
  3252. /* Don't do any checks other than the ones we have to, based
  3253. * on the size of the colocated partitions.
  3254. * Depends on the enum order: D_8x8, D_16x8, D_8x16, D_16x16 */
  3255. - int max_i8 = (D_16x16 - partition_col) + 1;
  3256. - int step = (partition_col == D_16x8) + 1;
  3257. - int width = 4 >> ((D_16x16 - partition_col)&1);
  3258. - int height = 4 >> ((D_16x16 - partition_col)>>1);
  3259. -
  3260. + int max_i8 = (D_16x16 - h->mb.i_partition) + 1;
  3261. + int step = (h->mb.i_partition == D_16x8) + 1;
  3262. + int width = 4 >> ((D_16x16 - h->mb.i_partition)&1);
  3263. + int height = 4 >> ((D_16x16 - h->mb.i_partition)>>1);
  3264. for( int i8 = 0; i8 < max_i8; i8 += step )
  3265. {
  3266. int x8 = i8&1;
  3267. int y8 = i8>>1;
  3268. - int i_part_8x8 = i_mb_8x8 + x8 + y8 * h->mb.i_b8_stride;
  3269. +
  3270. + if( IS_INTRA( type_col[y8] ) )
  3271. + {
  3272. + x264_macroblock_cache_ref( h, 2*x8, 2*y8, width, height, 0, 0 );
  3273. + x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 0, 0 );
  3274. + x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 1, 0 );
  3275. + continue;
  3276. + }
  3277. +
  3278. + int yM = 3*y8;
  3279. + if( h->param.b_interlaced && h->fref[1][0]->field[mb_xy] != h->mb.b_interlaced )
  3280. + {
  3281. + if( h->mb.b_interlaced )
  3282. + yM = y8*6;
  3283. + else
  3284. + yM = 2*(h->mb.i_mb_y&1) + y8;
  3285. + }
  3286. +
  3287. + int i_part_8x8 = i_mb_8x8 + x8 + (yM>>1) * h->mb.i_b8_stride;
  3288. int i_ref1_ref = h->fref[1][0]->ref[0][i_part_8x8];
  3289. - int i_ref = (map_col_to_list0(i_ref1_ref>>h->sh.b_mbaff) << h->sh.b_mbaff) + (i_ref1_ref&h->sh.b_mbaff);
  3290. + int i_ref = (map_col_to_list0(i_ref1_ref>>preshift) << postshift) + (offset&i_ref1_ref&h->mb.b_interlaced);
  3291.  
  3292. if( i_ref >= 0 )
  3293. {
  3294. int dist_scale_factor = h->mb.dist_scale_factor[i_ref][0];
  3295. - int16_t *mv_col = h->fref[1][0]->mv[0][i_mb_4x4 + 3*x8 + 3*y8 * h->mb.i_b4_stride];
  3296. + int16_t *mv_col = h->fref[1][0]->mv[0][i_mb_4x4 + 3*x8 + yM * h->mb.i_b4_stride];
  3297. + int16_t mv_y = (mv_col[1]<<yshift)/2;
  3298. int l0x = ( dist_scale_factor * mv_col[0] + 128 ) >> 8;
  3299. - int l0y = ( dist_scale_factor * mv_col[1] + 128 ) >> 8;
  3300. - if( h->param.i_threads > 1 && (l0y > h->mb.mv_max_spel[1] || l0y-mv_col[1] > h->mb.mv_max_spel[1]) )
  3301. + int l0y = ( dist_scale_factor * mv_y + 128 ) >> 8;
  3302. + if( h->param.i_threads > 1 && (l0y > h->mb.mv_max_spel[1] || l0y-mv_y > h->mb.mv_max_spel[1]) )
  3303. return 0;
  3304. x264_macroblock_cache_ref( h, 2*x8, 2*y8, width, height, 0, i_ref );
  3305. x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 0, pack16to32_mask(l0x, l0y) );
  3306. - x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_col[1]) );
  3307. + x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 1, pack16to32_mask(l0x-mv_col[0], l0y-mv_y) );
  3308. }
  3309. else
  3310. {
  3311. --
  3312. 1.7.4
  3313.  
  3314.  
  3315. From fe257a3f3ad1b5121c52999f1db6727aa50082c5 Mon Sep 17 00:00:00 2001
  3316. From: Simon Horlick <simonhorlick@gmail.com>
  3317. Date: Tue, 15 Mar 2011 01:17:01 +0000
  3318. Subject: [PATCH 23/25] Direct spatial
  3319.  
  3320. ---
  3321. common/mvpred.c | 78 +++++++++++++++++++++++++++++++++++++++++++-----------
  3322. 1 files changed, 62 insertions(+), 16 deletions(-)
  3323.  
  3324. diff --git a/common/mvpred.c b/common/mvpred.c
  3325. index 28eabca..f25fa03 100644
  3326. --- a/common/mvpred.c
  3327. +++ b/common/mvpred.c
  3328. @@ -295,15 +295,6 @@ static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
  3329. {
  3330. int8_t ref[2];
  3331. ALIGNED_ARRAY_8( int16_t, mv,[2],[2] );
  3332. - const int8_t *l1ref0 = &h->fref[1][0]->ref[0][h->mb.i_b8_xy];
  3333. - const int8_t *l1ref1 = &h->fref[1][0]->ref[1][h->mb.i_b8_xy];
  3334. - const int16_t (*l1mv[2])[2] = { (const int16_t (*)[2]) &h->fref[1][0]->mv[0][h->mb.i_b4_xy],
  3335. - (const int16_t (*)[2]) &h->fref[1][0]->mv[1][h->mb.i_b4_xy] };
  3336. - const int type_col = h->fref[1][0]->mb_type[h->mb.i_mb_xy];
  3337. - const int partition_col = h->fref[1][0]->mb_partition[h->mb.i_mb_xy];
  3338. -
  3339. - h->mb.i_partition = partition_col;
  3340. -
  3341. for( int i_list = 0; i_list < 2; i_list++ )
  3342. {
  3343. int i_refa = h->mb.cache.ref[i_list][X264_SCAN8_0 - 1];
  3344. @@ -348,6 +339,49 @@ static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
  3345. ref[i_list] = i_ref;
  3346. }
  3347.  
  3348. + int mb_x = h->mb.i_mb_x;
  3349. + int mb_y = h->mb.i_mb_y;
  3350. + int mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
  3351. + int type_col[2] = { h->fref[1][0]->mb_type[mb_xy], h->fref[1][0]->mb_type[mb_xy] };
  3352. + int partition_col[2] = { h->fref[1][0]->mb_partition[mb_xy], h->fref[1][0]->mb_partition[mb_xy] };
  3353. + int col_parity = abs(h->fref[1][0]->i_poc - h->fdec->i_poc)
  3354. + >= abs(h->fref[1][0]->i_poc + h->sh.i_delta_poc_bottom - h->fdec->i_poc);
  3355. + h->mb.i_partition = partition_col[0];
  3356. + if( h->param.b_interlaced && h->fref[1][0]->field[mb_xy] != h->mb.b_interlaced )
  3357. + {
  3358. + if( h->mb.b_interlaced )
  3359. + {
  3360. + mb_y = h->mb.i_mb_y&~1;
  3361. + mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
  3362. + type_col[0] = h->fref[1][0]->mb_type[mb_xy];
  3363. + type_col[1] = h->fref[1][0]->mb_type[mb_xy + h->mb.i_mb_stride];
  3364. + partition_col[0] = h->fref[1][0]->mb_partition[mb_xy];
  3365. + partition_col[1] = h->fref[1][0]->mb_partition[mb_xy + h->mb.i_mb_stride];
  3366. +
  3367. + if( (IS_INTRA(type_col[0]) || partition_col[0] == D_16x16)
  3368. + && (IS_INTRA(type_col[1]) || partition_col[1] == D_16x16)
  3369. + && partition_col[0] != D_8x8 )
  3370. + h->mb.i_partition = D_16x8;
  3371. + else
  3372. + h->mb.i_partition = D_8x8;
  3373. + }
  3374. + else
  3375. + {
  3376. + mb_y = (h->mb.i_mb_y&~1) + col_parity;
  3377. + mb_xy = mb_x + h->mb.i_mb_stride * mb_y;
  3378. + type_col[0] = type_col[1] = h->fref[1][0]->mb_type[mb_xy];
  3379. + partition_col[0] = partition_col[1] = h->fref[1][0]->mb_partition[mb_xy];
  3380. + h->mb.i_partition = partition_col[0];
  3381. + }
  3382. + }
  3383. + int i_mb_4x4 = 16 * h->mb.i_mb_stride * mb_y + 4 * mb_x;
  3384. + int i_mb_8x8 = 4 * h->mb.i_mb_stride * mb_y + 2 * mb_x;
  3385. +
  3386. + int8_t *l1ref0 = &h->fref[1][0]->ref[0][i_mb_8x8];
  3387. + int8_t *l1ref1 = &h->fref[1][0]->ref[1][i_mb_8x8];
  3388. + int16_t (*l1mv[2])[2] = { (int16_t (*)[2]) &h->fref[1][0]->mv[0][i_mb_4x4],
  3389. + (int16_t (*)[2]) &h->fref[1][0]->mv[1][i_mb_4x4] };
  3390. +
  3391. if( (M16( ref ) & 0x8080) == 0x8080 ) /* if( ref[0] < 0 && ref[1] < 0 ) */
  3392. {
  3393. x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, 0 );
  3394. @@ -367,24 +401,33 @@ static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
  3395. return 0;
  3396. }
  3397.  
  3398. - if( !M64( mv ) || IS_INTRA( type_col ) || (ref[0]&&ref[1]) )
  3399. + if( !M64( mv ) || (ref[0]&&ref[1]) )
  3400. return 1;
  3401.  
  3402. /* Don't do any checks other than the ones we have to, based
  3403. * on the size of the colocated partitions.
  3404. * Depends on the enum order: D_8x8, D_16x8, D_8x16, D_16x16 */
  3405. - int max_i8 = (D_16x16 - partition_col) + 1;
  3406. - int step = (partition_col == D_16x8) + 1;
  3407. - int width = 4 >> ((D_16x16 - partition_col)&1);
  3408. - int height = 4 >> ((D_16x16 - partition_col)>>1);
  3409. + int max_i8 = (D_16x16 - h->mb.i_partition) + 1;
  3410. + int step = (h->mb.i_partition == D_16x8) + 1;
  3411. + int width = 4 >> ((D_16x16 - h->mb.i_partition)&1);
  3412. + int height = 4 >> ((D_16x16 - h->mb.i_partition)>>1);
  3413.  
  3414. /* col_zero_flag */
  3415. for( int i8 = 0; i8 < max_i8; i8 += step )
  3416. {
  3417. const int x8 = i8&1;
  3418. const int y8 = i8>>1;
  3419. - const int o8 = x8 + y8 * h->mb.i_b8_stride;
  3420. - const int o4 = 3*(x8 + y8 * h->mb.i_b4_stride);
  3421. + int yM = 3*y8;
  3422. + if( h->param.b_interlaced && h->fref[1][0]->field[mb_xy] != h->mb.b_interlaced )
  3423. + {
  3424. + if( h->mb.b_interlaced )
  3425. + yM = y8*6;
  3426. + else
  3427. + yM = 2*(h->mb.i_mb_y&1) + y8;
  3428. + }
  3429. + int o8 = x8 + (yM>>1) * h->mb.i_b8_stride;
  3430. + int o4 = 3*x8 + yM * h->mb.i_b4_stride;
  3431. +
  3432. int idx;
  3433. if( l1ref0[o8] == 0 )
  3434. idx = 0;
  3435. @@ -393,6 +436,9 @@ static int x264_mb_predict_mv_direct16x16_spatial( x264_t *h )
  3436. else
  3437. continue;
  3438.  
  3439. + if( IS_INTRA( type_col[y8] ) )
  3440. + continue;
  3441. +
  3442. if( abs( l1mv[idx][o4][0] ) <= 1 && abs( l1mv[idx][o4][1] ) <= 1 )
  3443. {
  3444. if( ref[0] == 0 ) x264_macroblock_cache_mv( h, 2*x8, 2*y8, width, height, 0, 0 );
  3445. --
  3446. 1.7.4
  3447.  
  3448.  
  3449. From 53c6284008aa1471a62de4f6da6d587698a817c2 Mon Sep 17 00:00:00 2001
  3450. From: Simon Horlick <simonhorlick@gmail.com>
  3451. Date: Thu, 17 Mar 2011 17:39:18 +0000
  3452. Subject: [PATCH 24/25] Fix non-determinism with AQ
  3453.  
  3454. ---
  3455. encoder/ratecontrol.c | 4 ++--
  3456. 1 files changed, 2 insertions(+), 2 deletions(-)
  3457.  
  3458. diff --git a/encoder/ratecontrol.c b/encoder/ratecontrol.c
  3459. index e1a673f..bcbcb02 100644
  3460. --- a/encoder/ratecontrol.c
  3461. +++ b/encoder/ratecontrol.c
  3462. @@ -219,10 +219,10 @@ static ALWAYS_INLINE uint32_t ac_energy_plane( x264_t *h, int mb_x, int mb_y, x2
  3463. {
  3464. int w = i ? 8 : 16;
  3465. int stride = frame->i_stride[i];
  3466. - int offset = h->mb.b_interlaced
  3467. + int offset = h->sh.b_mbaff
  3468. ? 16 * mb_x + w * (mb_y&~1) * stride + (mb_y&1) * stride
  3469. : 16 * mb_x + w * mb_y * stride;
  3470. - stride <<= h->mb.b_interlaced;
  3471. + stride <<= h->sh.b_mbaff;
  3472. if( i )
  3473. {
  3474. ALIGNED_ARRAY_16( pixel, pix,[FENC_STRIDE*8] );
  3475. --
  3476. 1.7.4
  3477.  
  3478.  
  3479. From 7e2d83f1ef2dc762ffe6880ee54686088aeff660 Mon Sep 17 00:00:00 2001
  3480. From: Simon Horlick <simonhorlick@gmail.com>
  3481. Date: Mon, 14 Mar 2011 02:54:30 +0000
  3482. Subject: [PATCH 25/25] Adaptive mbaff with vsad decision
  3483.  
  3484. ---
  3485. encoder/encoder.c | 21 ++++++++++++++++++++-
  3486. 1 files changed, 20 insertions(+), 1 deletions(-)
  3487.  
  3488. diff --git a/encoder/encoder.c b/encoder/encoder.c
  3489. index 0319126..d885f31 100644
  3490. --- a/encoder/encoder.c
  3491. +++ b/encoder/encoder.c
  3492. @@ -1873,6 +1873,25 @@ static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
  3493. x264_macroblock_slice_init( h );
  3494. }
  3495.  
  3496. +static int field_vsad( x264_t *h, int mb_x, int mb_y )
  3497. +{
  3498. + int score_field = 0;
  3499. + int score_frame = 0;
  3500. + int stride = h->fenc->i_stride[0];
  3501. + uint8_t *fenc = h->fenc->plane[0] + 16 * (mb_x + mb_y * stride);
  3502. +
  3503. + for( int i = 1; i < 16; i++ )
  3504. + for( int j = 0; j < 16; j++ )
  3505. + score_frame += abs(fenc[i*stride+j] - fenc[(i-1)*stride+j]);
  3506. + for( int i = 2; i < 16; i+=2 )
  3507. + for( int j = 0; j < 16; j++ )
  3508. + score_field += abs(fenc[i*stride+j] - fenc[(i-2)*stride+j]);
  3509. + for( int i = 3; i < 16; i+=2 )
  3510. + for( int j = 0; j < 16; j++ )
  3511. + score_field += abs(fenc[i*stride+j] - fenc[(i-2)*stride+j]);
  3512. + return (score_field < score_frame);
  3513. +}
  3514. +
  3515. static int x264_slice_write( x264_t *h )
  3516. {
  3517. int i_skip;
  3518. @@ -1967,7 +1986,7 @@ static int x264_slice_write( x264_t *h )
  3519. if( h->mb.b_adaptive_mbaff )
  3520. {
  3521. if( !(i_mb_y&1) )
  3522. - h->mb.b_interlaced = 1;
  3523. + h->mb.b_interlaced = field_vsad( h, i_mb_x, i_mb_y );
  3524. x264_zigzag_init( h->param.cpu, &h->zigzagf, h->mb.b_interlaced );
  3525. }
  3526. h->mb.field[mb_xy] = h->mb.b_interlaced;
  3527. --
  3528. 1.7.4
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!