API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Jun 28th, 2017
490
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
C++ 40.86 KB | None | 0 0
raw download clone embed print report
  1. diff --git a/common/common.h b/common/common.h
  2. index 4307619..5c0ef1d 100644
  3. --- a/common/common.h
  4. +++ b/common/common.h
  5. @@ -59,12 +59,13 @@ do {\
  6.  #define X264_PCM_COST (384*BIT_DEPTH+16)
  7.  #define X264_LOOKAHEAD_MAX 250
  8.  #define QP_BD_OFFSET (6*(BIT_DEPTH-8))
  9. -#define QP_MAX (51+QP_BD_OFFSET)
  10. -#define QP_MAX_MAX (51+2*6)
  11. -#define LAMBDA_MAX (91 << (BIT_DEPTH-8))
  12. +#define QP_MAX_SPEC (51+QP_BD_OFFSET)
  13. +#define QP_MAX (QP_MAX_SPEC+24)
  14. +#define QP_MAX_MAX (51+2*6+24)
  15.  #define PIXEL_MAX ((1 << BIT_DEPTH)-1)
  16.  // arbitrary, but low because SATD scores are 1/4 normal
  17.  #define X264_LOOKAHEAD_QP (12+QP_BD_OFFSET)
  18. +#define SPEC_QP(x) X264_MIN((x), QP_MAX_SPEC)
  19.  
  20.  // number of pixels (per thread) in progress at any given time.
  21.  // 16 for the macroblock in progress + 3 for deblocking + 3 for motion compensation filter + 2 for extra safety
  22. @@ -458,12 +459,11 @@ struct x264_t
  23.      udctcoef        (*quant8_mf[2])[64];     /* [2][52][64] */
  24.      udctcoef        (*quant4_bias[4])[16];   /* [4][52][16] */
  25.      udctcoef        (*quant8_bias[2])[64];   /* [2][52][64] */
  26. +    udctcoef        (*nr_offset_emergency)[3][64];
  27.  
  28. -    /* mv/ref cost arrays.  Indexed by lambda instead of
  29. -     * qp because, due to rounding, some quantizers share
  30. -     * lambdas.  This saves memory. */
  31. -    uint16_t *cost_mv[LAMBDA_MAX+1];
  32. -    uint16_t *cost_mv_fpel[LAMBDA_MAX+1][4];
  33. +    /* mv/ref cost arrays. */
  34. +    uint16_t *cost_mv[QP_MAX+1];
  35. +    uint16_t *cost_mv_fpel[QP_MAX+1][4];
  36.  
  37.      const uint8_t   *chroma_qp_table; /* includes both the nonlinear luma->chroma mapping and chroma_qp_offset */
  38.  
  39. @@ -811,9 +811,14 @@ struct x264_t
  40.  
  41.      } stat;
  42.  
  43. -    ALIGNED_16( uint32_t nr_residual_sum[2][64] );
  44. -    ALIGNED_16( udctcoef nr_offset[2][64] );
  45. -    uint32_t        nr_count[2];
  46. +    /* 0 = luma 4x4, 1 = luma 8x8, 2 = chroma 4x4 */
  47. +    udctcoef (*nr_offset)[64];
  48. +    uint32_t (*nr_residual_sum)[64];
  49. +    uint32_t *nr_count;
  50. +
  51. +    ALIGNED_16( udctcoef nr_offset_denoise[3][64] );
  52. +    ALIGNED_16( uint32_t nr_residual_sum_buf[2][3][64] );
  53. +    uint32_t nr_count_buf[2][3];
  54.  
  55.      /* Buffers that are allocated per-thread even in sliced threads. */
  56.      void *scratch_buffer; /* for any temporary storage that doesn't want repeated malloc */
  57. diff --git a/common/macroblock.c b/common/macroblock.c
  58. index 5c76d3f..8013957 100644
  59. --- a/common/macroblock.c
  60. +++ b/common/macroblock.c
  61. @@ -1171,7 +1171,7 @@ void x264_macroblock_cache_save( x264_t *h )
  62.          if( h->mb.i_type != I_16x16 && h->mb.i_cbp_luma == 0 && h->mb.i_cbp_chroma == 0 )
  63.              h->mb.i_qp = h->mb.i_last_qp;
  64.          h->mb.qp[i_mb_xy] = h->mb.i_qp;
  65. -        h->mb.i_last_dqp = h->mb.i_qp - h->mb.i_last_qp;
  66. +        h->mb.i_last_dqp = SPEC_QP( h->mb.i_qp ) - SPEC_QP( h->mb.i_last_qp );
  67.          h->mb.i_last_qp = h->mb.i_qp;
  68.      }
  69.  
  70. diff --git a/common/quant.c b/common/quant.c
  71. index 816e60a..36e59ee 100644
  72. --- a/common/quant.c
  73. +++ b/common/quant.c
  74. @@ -143,7 +143,7 @@ static void dequant_4x4_dc( dctcoef dct[16], int dequant_mf[6][16], int i_qp )
  75.  
  76.  static void x264_denoise_dct( dctcoef *dct, uint32_t *sum, udctcoef *offset, int size )
  77.  {
  78. -    for( int i = 1; i < size; i++ )
  79. +    for( int i = 0; i < size; i++ )
  80.      {
  81.          int level = dct[i];
  82.          int sign = level>>31;
  83. diff --git a/common/set.c b/common/set.c
  84. index 92b6b7a..c4d5549 100644
  85. --- a/common/set.c
  86. +++ b/common/set.c
  87. @@ -23,6 +23,8 @@
  88.   * For more information, contact us at licensing@x264.com.
  89.   *****************************************************************************/
  90.  
  91. +#define _ISOC99_SOURCE
  92. +#include <math.h>
  93.  #include "common.h"
  94.  
  95.  #define SHIFT(x,s) ((s)<=0 ? (x)<<-(s) : ((x)+(1<<((s)-1)))>>(s))
  96. @@ -146,21 +148,22 @@ int x264_cqm_init( x264_t *h )
  97.                       quant8_mf[i_list][q][i] = DIV(def_quant8[q][i] * 16, h->pps->scaling_list[4+i_list][i]);
  98.              }
  99.      }
  100. -    for( int q = 0; q < QP_MAX+1; q++ )
  101. +    for( int qm = 0; qm <= QP_MAX; qm++ )
  102.      {
  103.          int j;
  104. +        int q = SPEC_QP( qm );
  105.          for( int i_list = 0; i_list < 4; i_list++ )
  106.              for( int i = 0; i < 16; i++ )
  107.              {
  108. -                h->unquant4_mf[i_list][q][i] = (1ULL << (q/6 + 15 + 8)) / quant4_mf[i_list][q%6][i];
  109. -                h->quant4_mf[i_list][q][i] = j = SHIFT(quant4_mf[i_list][q%6][i], q/6 - 1);
  110. +                h->unquant4_mf[i_list][qm][i] = (1ULL << (q/6 + 15 + 8)) / quant4_mf[i_list][q%6][i];
  111. +                h->quant4_mf[i_list][qm][i] = j = SHIFT(quant4_mf[i_list][q%6][i], q/6 - 1);
  112.                  if( !j )
  113.                  {
  114.                      min_qp_err = X264_MIN( min_qp_err, q );
  115.                      continue;
  116.                  }
  117.                  // round to nearest, unless that would cause the deadzone to be negative
  118. -                h->quant4_bias[i_list][q][i] = X264_MIN( DIV(deadzone[i_list]<<10, j), (1<<15)/j );
  119. +                h->quant4_bias[i_list][qm][i] = X264_MIN( DIV(deadzone[i_list]<<10, j), (1<<15)/j );
  120.                  if( j > 0xffff && q > max_qp_err && (i_list == CQM_4IY || i_list == CQM_4PY) )
  121.                      max_qp_err = q;
  122.                  if( j > 0xffff && q > max_chroma_qp_err && (i_list == CQM_4IC || i_list == CQM_4PC) )
  123. @@ -170,20 +173,62 @@ int x264_cqm_init( x264_t *h )
  124.              for( int i_list = 0; i_list < 2; i_list++ )
  125.                  for( int i = 0; i < 64; i++ )
  126.                  {
  127. -                    h->unquant8_mf[i_list][q][i] = (1ULL << (q/6 + 16 + 8)) / quant8_mf[i_list][q%6][i];
  128. +                    h->unquant8_mf[i_list][qm][i] = (1ULL << (q/6 + 16 + 8)) / quant8_mf[i_list][q%6][i];
  129.                      j = SHIFT(quant8_mf[i_list][q%6][i], q/6);
  130. -                    h->quant8_mf[i_list][q][i] = (uint16_t)j;
  131. +                    h->quant8_mf[i_list][qm][i] = (uint16_t)j;
  132.  
  133.                      if( !j )
  134.                      {
  135.                          min_qp_err = X264_MIN( min_qp_err, q );
  136.                          continue;
  137.                      }
  138. -                    h->quant8_bias[i_list][q][i] = X264_MIN( DIV(deadzone[i_list]<<10, j), (1<<15)/j );
  139. +                    h->quant8_bias[i_list][qm][i] = X264_MIN( DIV(deadzone[i_list]<<10, j), (1<<15)/j );
  140.                      if( j > 0xffff && q > max_qp_err )
  141.                          max_qp_err = q;
  142.                  }
  143.      }
  144. +    x264_emms();
  145. +    CHECKED_MALLOC( h->nr_offset_emergency, sizeof(*h->nr_offset_emergency)*(QP_MAX-QP_MAX_SPEC) );
  146. +    for( int q = 0; q < QP_MAX - QP_MAX_SPEC; q++ )
  147. +        for( int cat = 0; cat <= 2; cat++ )
  148. +        {
  149. +            int dct8x8 = cat == 1;
  150. +            int size = dct8x8 ? 64 : 16;
  151. +            udctcoef *nr_offset = h->nr_offset_emergency[q][cat];
  152. +            int lowest_dc = (QP_MAX-QP_MAX_SPEC)/2;
  153. +
  154. +            for( int i = 0; i < size; i++ )
  155. +            {
  156. +                /* True "emergency mode": remove all DCT coefficients */
  157. +                if( q == QP_MAX - QP_MAX_SPEC - 1 )
  158. +                {
  159. +                    nr_offset[i] = (1 << (7 + BIT_DEPTH)) - 1;
  160. +                    continue;
  161. +                }
  162. +
  163. +                double pos;
  164. +                if( i == 0 )
  165. +                {
  166. +                    /* Only reduce DC at very high QPs */
  167. +                    if( q < lowest_dc )
  168. +                    {
  169. +                        nr_offset[i] = 0;
  170. +                        continue;
  171. +                    }
  172. +                    pos = (double)(q-lowest_dc+1) / (QP_MAX - QP_MAX_SPEC - lowest_dc);
  173. +                }
  174. +                else
  175. +                    pos = (double)(q+1) / (QP_MAX - QP_MAX_SPEC);
  176. +
  177. +                /* Exponentially interpolate between "start" and "end" */
  178. +                double start = dct8x8 ? h->unquant8_mf[CQM_8PY][QP_MAX_SPEC][i] / 4096.0
  179. +                                      : h->unquant4_mf[CQM_4PY][QP_MAX_SPEC][i] / 4096.0;
  180. +                double end = start * 32;
  181. +                double scale = log(end / start);
  182. +                double bias = exp(scale * pos) * start;
  183. +                nr_offset[i] = bias + 0.5;
  184. +            }
  185. +        }
  186.  
  187.      if( !h->mb.b_lossless && max_qp_err >= h->param.rc.i_qp_min )
  188.      {
  189. @@ -233,6 +278,7 @@ void x264_cqm_delete( x264_t *h )
  190.  {
  191.      CQM_DELETE( 4, 4 );
  192.      CQM_DELETE( 8, 2 );
  193. +    x264_free( h->nr_offset_emergency );
  194.  }
  195.  
  196.  static int x264_cqm_parse_jmlist( x264_t *h, const char *buf, const char *name,
  197. diff --git a/common/x86/quant-a.asm b/common/x86/quant-a.asm
  198. index 5d7a15e..e09f6c9 100644
  199. --- a/common/x86/quant-a.asm
  200. +++ b/common/x86/quant-a.asm
  201. @@ -699,8 +699,7 @@ DEQUANT_DC sse2
  202.  ; void denoise_dct( int32_t *dct, uint32_t *sum, uint32_t *offset, int size )
  203.  ;-----------------------------------------------------------------------------
  204.  %macro DENOISE_DCT 1-2 0
  205. -cglobal denoise_dct_%1, 4,5,%2
  206. -    mov       r4d, [r0] ; backup DC coefficient
  207. +cglobal denoise_dct_%1, 4,4,%2
  208.      pxor      m6, m6
  209.  .loop:
  210.      sub       r3, mmsize/2
  211. @@ -727,8 +726,7 @@ cglobal denoise_dct_%1, 4,5,%2
  212.      mova      [r1+r3*4+0*mmsize], m4
  213.      mova      [r1+r3*4+1*mmsize], m5
  214.      jg .loop
  215. -    mov       [r0], r4d ; restore DC coefficient
  216. -    RET
  217. +    REP_RET
  218.  %endmacro
  219.  
  220.  %define PABSD PABSD_MMX
  221. @@ -749,8 +747,7 @@ DENOISE_DCT ssse3, 8
  222.  ; void denoise_dct( int16_t *dct, uint32_t *sum, uint16_t *offset, int size )
  223.  ;-----------------------------------------------------------------------------
  224.  %macro DENOISE_DCT 1-2 0
  225. -cglobal denoise_dct_%1, 4,5,%2
  226. -    movzx     r4d, word [r0]
  227. +cglobal denoise_dct_%1, 4,4,%2
  228.      pxor      m6, m6
  229.  .loop:
  230.      sub       r3, mmsize
  231. @@ -781,8 +778,7 @@ cglobal denoise_dct_%1, 4,5,%2
  232.      mova      [r1+r3*4+2*mmsize], m5
  233.      mova      [r1+r3*4+3*mmsize], m3
  234.      jg .loop
  235. -    mov       [r0], r4w
  236. -    RET
  237. +    REP_RET
  238.  %endmacro
  239.  
  240.  %define PABSW PABSW_MMX
  241. diff --git a/encoder/analyse.c b/encoder/analyse.c
  242. index a997425..d2ec63f 100644
  243. --- a/encoder/analyse.c
  244. +++ b/encoder/analyse.c
  245. @@ -147,18 +147,26 @@ const uint16_t x264_lambda_tab[QP_MAX_MAX+1] = {
  246.    25,  29,  32,  36,  40,  45,  51,  57, /* 40-47 */
  247.    64,  72,  81,  91, 102, 114, 128, 144, /* 48-55 */
  248.   161, 181, 203, 228, 256, 287, 323, 362, /* 56-63 */
  249. + 406, 456, 512, 575, 645, 724, 813, 912, /* 64-71 */
  250. +1024,1149,1290,1448,1625,1825,2048,2299, /* 72-79 */
  251. +2048,2299,2580,2896,3251,3649,4096,4598, /* 80-87 */
  252.  };
  253.  
  254.  /* lambda2 = pow(lambda,2) * .9 * 256 */
  255. +/* Capped to avoid overflow */
  256.  const int x264_lambda2_tab[QP_MAX_MAX+1] = {
  257. -     14,     18,     22,      28,      36,      45,      57,      72, /*  0- 7 */
  258. -     91,    115,    145,     182,     230,     290,     365,     460, /*  8-15 */
  259. -    580,    731,    921,    1161,    1462,    1843,    2322,    2925, /* 16-23 */
  260. -   3686,   4644,   5851,    7372,    9289,   11703,   14745,   18578, /* 24-31 */
  261. -  23407,  29491,  37156,   46814,   58982,   74313,   93628,  117964, /* 32-39 */
  262. - 148626, 187257, 235929,  297252,  374514,  471859,  594505,  749029, /* 40-47 */
  263. - 943718,1189010,1498059, 1887436, 2378021, 2996119, 3774873, 4756042, /* 48-55 */
  264. -5992238,7549747,9512085,11984476,15099494,19024170,23968953,30198988, /* 56-63 */
  265. +       14,       18,       22,       28,       36,       45,      57,      72, /*  0- 7 */
  266. +       91,      115,      145,      182,      230,      290,     365,     460, /*  8-15 */
  267. +      580,      731,      921,     1161,     1462,     1843,    2322,    2925, /* 16-23 */
  268. +     3686,     4644,     5851,     7372,     9289,    11703,   14745,   18578, /* 24-31 */
  269. +    23407,    29491,    37156,    46814,    58982,    74313,   93628,  117964, /* 32-39 */
  270. +   148626,   187257,   235929,   297252,   374514,   471859,  594505,  749029, /* 40-47 */
  271. +   943718,  1189010,  1498059,  1887436,  2378021,  2996119, 3774873, 4756042, /* 48-55 */
  272. +  5992238,  7549747,  9512085, 11984476, 15099494, 19024170,23968953,30198988, /* 56-63 */
  273. + 38048341, 47937906, 60397977, 76096683, 95875813,120795955,                   /* 64-69 */
  274. +134217727,134217727,134217727,134217727,134217727,134217727,                   /* 70-75 */
  275. +134217727,134217727,134217727,134217727,134217727,134217727,                   /* 76-81 */
  276. +134217727,134217727,134217727,134217727,134217727,134217727,                   /* 82-87 */
  277.  };
  278.  
  279.  const uint8_t x264_exp2_lut[64] = {
  280. @@ -196,29 +204,40 @@ const float x264_log2_lz_lut[32] = {
  281.  // I'm just matching the behaviour of deadzone quant.
  282.  static const int x264_trellis_lambda2_tab[2][QP_MAX_MAX+1] = {
  283.      // inter lambda = .85 * .85 * 2**(qp/3. + 10 - LAMBDA_BITS)
  284. -    {      46,      58,      73,      92,     117,     147,
  285. -          185,     233,     294,     370,     466,     587,
  286. -          740,     932,    1174,    1480,    1864,    2349,
  287. -         2959,    3728,    4697,    5918,    7457,    9395,
  288. -        11837,   14914,   18790,   23674,   29828,   37581,
  289. -        47349,   59656,   75163,   94699,  119313,  150326,
  290. -       189399,  238627,  300652,  378798,  477255,  601304,
  291. -       757596,  954511, 1202608, 1515192, 1909022, 2405217,
  292. -      3030384, 3818045, 4810435, 6060769, 7636091, 9620872,
  293. -     12121539,15272182,19241743,24243077,30544363,38483486,
  294. -     48486154,61088726,76966972,96972308 },
  295. +    {
  296. +               46,       58,       73,       92,      117,      147,
  297. +              185,      233,      294,      370,      466,      587,
  298. +              740,      932,     1174,     1480,     1864,     2349,
  299. +             2959,     3728,     4697,     5918,     7457,     9395,
  300. +            11837,    14914,    18790,    23674,    29828,    37581,
  301. +            47349,    59656,    75163,    94699,   119313,   150326,
  302. +           189399,   238627,   300652,   378798,   477255,   601304,
  303. +           757596,   954511,  1202608,  1515192,  1909022,  2405217,
  304. +          3030384,  3818045,  4810435,  6060769,  7636091,  9620872,
  305. +         12121539, 15272182, 19241743, 24243077, 30544363, 38483486,
  306. +         48486154, 61088726, 76966972, 96972308,
  307. +        122177453,134217727,134217727,134217727,134217727,134217727,
  308. +        134217727,134217727,134217727,134217727,134217727,134217727,
  309. +        134217727,134217727,134217727,134217727,134217727,134217727
  310. +    },
  311.      // intra lambda = .65 * .65 * 2**(qp/3. + 10 - LAMBDA_BITS)
  312. -    {      27,      34,      43,      54,      68,      86,
  313. -          108,     136,     172,     216,     273,     343,
  314. -          433,     545,     687,     865,    1090,    1374,
  315. -         1731,    2180,    2747,    3461,    4361,    5494,
  316. -         6922,    8721,   10988,   13844,   17442,   21976,
  317. -        27688,   34885,   43953,   55377,   69771,   87906,
  318. -       110755,  139543,  175813,  221511,  279087,  351627,
  319. -       443023,  558174,  703255,  886046, 1116348, 1406511,
  320. -      1772093, 2232697, 2813022, 3544186, 4465396, 5626046,
  321. -      7088374, 8930791,11252092,14176748,17861583,22504184,
  322. -     28353495,35723165,45008368,56706990 }
  323. +    {
  324. +               27,       34,       43,       54,       68,       86,
  325. +              108,      136,      172,      216,      273,      343,
  326. +              433,      545,      687,      865,     1090,     1374,
  327. +             1731,     2180,     2747,     3461,     4361,     5494,
  328. +             6922,     8721,    10988,    13844,    17442,    21976,
  329. +            27688,    34885,    43953,    55377,    69771,    87906,
  330. +           110755,   139543,   175813,   221511,   279087,   351627,
  331. +           443023,   558174,   703255,   886046,  1116348,  1406511,
  332. +          1772093,  2232697,  2813022,  3544186,  4465396,  5626046,
  333. +          7088374,  8930791, 11252092, 14176748, 17861583, 22504184,
  334. +         28353495, 35723165, 45008368, 56706990,
  335. +         71446330, 90016736,113413980,134217727,134217727,134217727,
  336. +        134217727,134217727,134217727,134217727,134217727,134217727,
  337. +        134217727,134217727,134217727,134217727,134217727,134217727,
  338. +        134217727,134217727,134217727,134217727,134217727,134217727
  339. +    }
  340.  };
  341.  
  342.  static const uint16_t x264_chroma_lambda2_offset_tab[] = {
  343. @@ -247,35 +266,35 @@ static const uint8_t i_sub_mb_p_cost_table[4] = {
  344.  
  345.  static void x264_analyse_update_cache( x264_t *h, x264_mb_analysis_t *a );
  346.  
  347. -static uint16_t x264_cost_ref[LAMBDA_MAX+1][3][33];
  348. +static uint16_t x264_cost_ref[QP_MAX+1][3][33];
  349.  static UNUSED x264_pthread_mutex_t cost_ref_mutex = X264_PTHREAD_MUTEX_INITIALIZER;
  350.  
  351.  int x264_analyse_init_costs( x264_t *h, int qp )
  352.  {
  353.      int lambda = x264_lambda_tab[qp];
  354. -    if( h->cost_mv[lambda] )
  355. +    if( h->cost_mv[qp] )
  356.          return 0;
  357.      /* factor of 4 from qpel, 2 from sign, and 2 because mv can be opposite from mvp */
  358. -    CHECKED_MALLOC( h->cost_mv[lambda], (4*4*2048 + 1) * sizeof(uint16_t) );
  359. -    h->cost_mv[lambda] += 2*4*2048;
  360. +    CHECKED_MALLOC( h->cost_mv[qp], (4*4*2048 + 1) * sizeof(uint16_t) );
  361. +    h->cost_mv[qp] += 2*4*2048;
  362.      for( int i = 0; i <= 2*4*2048; i++ )
  363.      {
  364. -        h->cost_mv[lambda][-i] =
  365. -        h->cost_mv[lambda][i]  = lambda * (log2f(i+1)*2 + 0.718f + !!i) + .5f;
  366. +        h->cost_mv[qp][-i] =
  367. +        h->cost_mv[qp][i]  = X264_MIN( lambda * (log2f(i+1)*2 + 0.718f + !!i) + .5f, (1<<16)-1 );
  368.      }
  369.      x264_pthread_mutex_lock( &cost_ref_mutex );
  370.      for( int i = 0; i < 3; i++ )
  371.          for( int j = 0; j < 33; j++ )
  372. -            x264_cost_ref[lambda][i][j] = i ? lambda * bs_size_te( i, j ) : 0;
  373. +            x264_cost_ref[qp][i][j] = X264_MIN( i ? lambda * bs_size_te( i, j ) : 0, (1<<16)-1 );
  374.      x264_pthread_mutex_unlock( &cost_ref_mutex );
  375. -    if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->cost_mv_fpel[lambda][0] )
  376. +    if( h->param.analyse.i_me_method >= X264_ME_ESA && !h->cost_mv_fpel[qp][0] )
  377.      {
  378.          for( int j = 0; j < 4; j++ )
  379.          {
  380. -            CHECKED_MALLOC( h->cost_mv_fpel[lambda][j], (4*2048 + 1) * sizeof(uint16_t) );
  381. -            h->cost_mv_fpel[lambda][j] += 2*2048;
  382. +            CHECKED_MALLOC( h->cost_mv_fpel[qp][j], (4*2048 + 1) * sizeof(uint16_t) );
  383. +            h->cost_mv_fpel[qp][j] += 2*2048;
  384.              for( int i = -2*2048; i < 2*2048; i++ )
  385. -                h->cost_mv_fpel[lambda][j][i] = h->cost_mv[lambda][i*4+j];
  386. +                h->cost_mv_fpel[qp][j][i] = h->cost_mv[qp][i*4+j];
  387.          }
  388.      }
  389.      return 0;
  390. @@ -285,7 +304,7 @@ fail:
  391.  
  392.  void x264_analyse_free_costs( x264_t *h )
  393.  {
  394. -    for( int i = 0; i < LAMBDA_MAX+1; i++ )
  395. +    for( int i = 0; i < QP_MAX+1; i++ )
  396.      {
  397.          if( h->cost_mv[i] )
  398.              x264_free( h->cost_mv[i] - 2*4*2048 );
  399. @@ -326,16 +345,16 @@ void x264_analyse_weight_frame( x264_t *h, int end )
  400.  /* initialize an array of lambda*nbits for all possible mvs */
  401.  static void x264_mb_analyse_load_costs( x264_t *h, x264_mb_analysis_t *a )
  402.  {
  403. -    a->p_cost_mv = h->cost_mv[a->i_lambda];
  404. -    a->p_cost_ref[0] = x264_cost_ref[a->i_lambda][x264_clip3(h->sh.i_num_ref_idx_l0_active-1,0,2)];
  405. -    a->p_cost_ref[1] = x264_cost_ref[a->i_lambda][x264_clip3(h->sh.i_num_ref_idx_l1_active-1,0,2)];
  406. +    a->p_cost_mv = h->cost_mv[a->i_qp];
  407. +    a->p_cost_ref[0] = x264_cost_ref[a->i_qp][x264_clip3(h->sh.i_num_ref_idx_l0_active-1,0,2)];
  408. +    a->p_cost_ref[1] = x264_cost_ref[a->i_qp][x264_clip3(h->sh.i_num_ref_idx_l1_active-1,0,2)];
  409.  }
  410.  
  411.  static void x264_mb_analyse_init_qp( x264_t *h, x264_mb_analysis_t *a, int i_qp )
  412.  {
  413.      /* conduct the analysis using this lamda and QP */
  414.      a->i_qp = h->mb.i_qp = i_qp;
  415. -    h->mb.i_chroma_qp = h->chroma_qp_table[i_qp];
  416. +    h->mb.i_chroma_qp = h->chroma_qp_table[SPEC_QP( i_qp )];
  417.  
  418.      a->i_lambda = x264_lambda_tab[i_qp];
  419.      a->i_lambda2 = x264_lambda2_tab[i_qp];
  420. @@ -343,10 +362,11 @@ static void x264_mb_analyse_init_qp( x264_t *h, x264_mb_analysis_t *a, int i_qp
  421.      h->mb.b_trellis = h->param.analyse.i_trellis > 1 && a->i_mbrd;
  422.      if( h->param.analyse.i_trellis )
  423.      {
  424. +        int effective_chroma_qp = h->mb.i_chroma_qp + X264_MAX( i_qp - QP_MAX_SPEC, 0 );
  425.          h->mb.i_trellis_lambda2[0][0] = x264_trellis_lambda2_tab[0][h->mb.i_qp];
  426.          h->mb.i_trellis_lambda2[0][1] = x264_trellis_lambda2_tab[1][h->mb.i_qp];
  427. -        h->mb.i_trellis_lambda2[1][0] = x264_trellis_lambda2_tab[0][h->mb.i_chroma_qp];
  428. -        h->mb.i_trellis_lambda2[1][1] = x264_trellis_lambda2_tab[1][h->mb.i_chroma_qp];
  429. +        h->mb.i_trellis_lambda2[1][0] = x264_trellis_lambda2_tab[0][effective_chroma_qp];
  430. +        h->mb.i_trellis_lambda2[1][1] = x264_trellis_lambda2_tab[1][effective_chroma_qp];
  431.      }
  432.      h->mb.i_psy_rd_lambda = a->i_lambda;
  433.      /* Adjusting chroma lambda based on QP offset hurts PSNR but improves visual quality. */
  434. @@ -366,7 +386,6 @@ static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int i_qp )
  435.      x264_mb_analyse_init_qp( h, a, i_qp );
  436.  
  437.      h->mb.b_transform_8x8 = 0;
  438. -    h->mb.b_noise_reduction = 0;
  439.  
  440.      /* I: Intra part */
  441.      a->i_satd_i16x16 =
  442. @@ -384,6 +403,21 @@ static void x264_mb_analyse_init( x264_t *h, x264_mb_analysis_t *a, int i_qp )
  443.          a->i_mbrd ? 2 :
  444.          !h->param.analyse.i_trellis && !h->param.analyse.i_noise_reduction;
  445.  
  446. +    if( h->mb.i_qp > QP_MAX_SPEC )
  447. +    {
  448. +        h->nr_offset = h->nr_offset_emergency[h->mb.i_qp-QP_MAX_SPEC-1];
  449. +        h->nr_residual_sum = h->nr_residual_sum_buf[1];
  450. +        h->nr_count = h->nr_count_buf[1];
  451. +        h->mb.b_noise_reduction = 1;
  452. +    }
  453. +    else
  454. +    {
  455. +        h->nr_offset = h->nr_offset_denoise;
  456. +        h->nr_residual_sum = h->nr_residual_sum_buf[0];
  457. +        h->nr_count = h->nr_count_buf[0];
  458. +        h->mb.b_noise_reduction = 0;
  459. +    }
  460. +
  461.      /* II: Inter part P/B frame */
  462.      if( h->sh.i_type != SLICE_TYPE_I )
  463.      {
  464. @@ -2731,6 +2765,7 @@ void x264_macroblock_analyse( x264_t *h )
  465.      int i_cost = COST_MAX;
  466.  
  467.      h->mb.i_qp = x264_ratecontrol_mb_qp( h );
  468. +
  469.      /* If the QP of this MB is within 1 of the previous MB, code the same QP as the previous MB,
  470.       * to lower the bit cost of the qp_delta.  Don't do this if QPRD is enabled. */
  471.      if( h->param.rc.i_aq_mode && h->param.analyse.i_subpel_refine < 10 && abs(h->mb.i_qp - h->mb.i_last_qp) == 1 )
  472. @@ -3476,7 +3511,8 @@ intra_analysis:
  473.          x264_mb_analyse_qp_rd( h, &analysis );
  474.  
  475.      h->mb.b_trellis = h->param.analyse.i_trellis;
  476. -    h->mb.b_noise_reduction = !!h->param.analyse.i_noise_reduction;
  477. +    h->mb.b_noise_reduction = h->mb.b_noise_reduction || (!!h->param.analyse.i_noise_reduction && !IS_INTRA( h->mb.i_type ));
  478. +
  479.      if( !IS_SKIP(h->mb.i_type) && h->mb.i_psy_trellis && h->param.analyse.i_trellis == 1 )
  480.          x264_psy_trellis_init( h, 0 );
  481.      if( h->mb.b_trellis == 1 || h->mb.b_noise_reduction )
  482. diff --git a/encoder/cabac.c b/encoder/cabac.c
  483. index 5539b4b..a2d13f0 100644
  484. --- a/encoder/cabac.c
  485. +++ b/encoder/cabac.c
  486. @@ -246,7 +246,7 @@ static void x264_cabac_mb_cbp_chroma( x264_t *h, x264_cabac_t *cb )
  487.  
  488.  static void x264_cabac_mb_qp_delta( x264_t *h, x264_cabac_t *cb )
  489.  {
  490. -    int i_dqp = h->mb.i_qp - h->mb.i_last_qp;
  491. +    int i_dqp = SPEC_QP( h->mb.i_qp) - SPEC_QP( h->mb.i_last_qp );
  492.      int ctx;
  493.  
  494.      /* Avoid writing a delta quant if we have an empty i16x16 block, e.g. in a completely flat background area */
  495. diff --git a/encoder/cavlc.c b/encoder/cavlc.c
  496. index ca35210..7224a9d 100644
  497. --- a/encoder/cavlc.c
  498. +++ b/encoder/cavlc.c
  499. @@ -208,7 +208,7 @@ static const uint8_t ct_index[17] = {0,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,3};
  500.  static void cavlc_qp_delta( x264_t *h )
  501.  {
  502.      bs_t *s = &h->out.bs;
  503. -    int i_dqp = h->mb.i_qp - h->mb.i_last_qp;
  504. +    int i_dqp = SPEC_QP( h->mb.i_qp ) - SPEC_QP( h->mb.i_last_qp );
  505.  
  506.      /* Avoid writing a delta quant if we have an empty i16x16 block, e.g. in a completely flat background area */
  507.      if( h->mb.i_type == I_16x16 && !(h->mb.i_cbp_luma | h->mb.i_cbp_chroma)
  508. diff --git a/encoder/encoder.c b/encoder/encoder.c
  509. index a6b53fd..89f1f39 100644
  510. --- a/encoder/encoder.c
  511. +++ b/encoder/encoder.c
  512. @@ -163,7 +163,7 @@ static void x264_slice_header_init( x264_t *h, x264_slice_header_t *sh,
  513.      sh->i_cabac_init_idc = param->i_cabac_init_idc;
  514.  
  515.      sh->i_qp = i_qp;
  516. -    sh->i_qp_delta = i_qp - pps->i_pic_init_qp;
  517. +    sh->i_qp_delta = SPEC_QP( sh->i_qp ) - pps->i_pic_init_qp;
  518.      sh->b_sp_for_swidth = 0;
  519.      sh->i_qs_delta = 0;
  520.  
  521. @@ -1050,7 +1050,8 @@ x264_t *x264_encoder_open( x264_param_t *param )
  522.          p += sprintf( p, " none!" );
  523.      x264_log( h, X264_LOG_INFO, "%s\n", buf );
  524.  
  525. -    for( qp = h->param.rc.i_qp_min; qp <= h->param.rc.i_qp_max; qp++ )
  526. +    int qp_max = h->param.rc.i_qp_max == QP_MAX_SPEC ? QP_MAX : h->param.rc.i_qp_max;
  527. +    for( qp = h->param.rc.i_qp_min; qp <= qp_max; qp++ )
  528.          if( x264_analyse_init_costs( h, qp ) )
  529.              goto fail;
  530.      if( x264_analyse_init_costs( h, X264_LOOKAHEAD_QP ) )
  531. @@ -1058,7 +1059,7 @@ x264_t *x264_encoder_open( x264_param_t *param )
  532.  
  533.      static const uint16_t cost_mv_correct[7] = { 24, 47, 95, 189, 379, 757, 1515 };
  534.      /* Checks for known miscompilation issues. */
  535. -    if( h->cost_mv[x264_lambda_tab[X264_LOOKAHEAD_QP]][2013] != cost_mv_correct[BIT_DEPTH-8] )
  536. +    if( h->cost_mv[X264_LOOKAHEAD_QP][2013] != cost_mv_correct[BIT_DEPTH-8] )
  537.      {
  538.          x264_log( h, X264_LOG_ERROR, "MV cost test failed: x264 has been miscompiled!\n" );
  539.          goto fail;
  540. @@ -1879,7 +1880,7 @@ static int x264_slice_write( x264_t *h )
  541.      if( h->sh.i_first_mb != h->i_threadslice_start * h->mb.i_mb_width )
  542.      {
  543.          h->sh.i_qp = h->mb.i_last_qp;
  544. -        h->sh.i_qp_delta = h->sh.i_qp - h->pps->i_pic_init_qp;
  545. +        h->sh.i_qp_delta = SPEC_QP( h->sh.i_qp ) - h->pps->i_pic_init_qp;
  546.      }
  547.  
  548.      x264_slice_header_write( &h->out.bs, &h->sh, h->i_nal_ref_idc );
  549. diff --git a/encoder/macroblock.c b/encoder/macroblock.c
  550. index 39fed61..11a63d9 100644
  551. --- a/encoder/macroblock.c
  552. +++ b/encoder/macroblock.c
  553. @@ -84,6 +84,8 @@ static inline void dct2x2dc( dctcoef d[4], dctcoef dct4x4[4][16] )
  554.  static ALWAYS_INLINE int x264_quant_4x4( x264_t *h, dctcoef dct[16], int i_qp, int ctx_block_cat, int b_intra, int idx )
  555.  {
  556.      int i_quant_cat = b_intra ? CQM_4IY : CQM_4PY;
  557. +    if( h->mb.b_noise_reduction && ctx_block_cat != DCT_LUMA_AC )
  558. +        h->quantf.denoise_dct( dct, h->nr_residual_sum[0], h->nr_offset[0], 16 );
  559.      if( h->mb.b_trellis )
  560.          return x264_quant_4x4_trellis( h, dct, i_quant_cat, i_qp, ctx_block_cat, b_intra, 0, idx );
  561.      else
  562. @@ -93,6 +95,8 @@ static ALWAYS_INLINE int x264_quant_4x4( x264_t *h, dctcoef dct[16], int i_qp, i
  563.  static ALWAYS_INLINE int x264_quant_8x8( x264_t *h, dctcoef dct[64], int i_qp, int b_intra, int idx )
  564.  {
  565.      int i_quant_cat = b_intra ? CQM_8IY : CQM_8PY;
  566. +    if( h->mb.b_noise_reduction )
  567. +        h->quantf.denoise_dct( dct, h->nr_residual_sum[1], h->nr_offset[1], 64 );
  568.      if( h->mb.b_trellis )
  569.          return x264_quant_8x8_trellis( h, dct, i_quant_cat, i_qp, b_intra, idx );
  570.      else
  571. @@ -115,6 +119,7 @@ void x264_mb_encode_i4x4( x264_t *h, int idx, int i_qp )
  572.      int nz;
  573.      pixel *p_src = &h->mb.pic.p_fenc[0][block_idx_xy_fenc[idx]];
  574.      pixel *p_dst = &h->mb.pic.p_fdec[0][block_idx_xy_fdec[idx]];
  575. +    int i_qp_spec = SPEC_QP( i_qp );
  576.      ALIGNED_ARRAY_16( dctcoef, dct4x4,[16] );
  577.  
  578.      if( h->mb.b_lossless )
  579. @@ -133,7 +138,7 @@ void x264_mb_encode_i4x4( x264_t *h, int idx, int i_qp )
  580.      {
  581.          h->mb.i_cbp_luma |= 1<<(idx>>2);
  582.          h->zigzagf.scan_4x4( h->dct.luma4x4[idx], dct4x4 );
  583. -        h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4IY], i_qp );
  584. +        h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4IY], i_qp_spec );
  585.          h->dctf.add4x4_idct( p_dst, dct4x4 );
  586.      }
  587.  }
  588. @@ -161,6 +166,7 @@ void x264_mb_encode_i8x8( x264_t *h, int idx, int i_qp )
  589.      int nz;
  590.      pixel *p_src = &h->mb.pic.p_fenc[0][8*x + 8*y*FENC_STRIDE];
  591.      pixel *p_dst = &h->mb.pic.p_fdec[0][8*x + 8*y*FDEC_STRIDE];
  592. +    int i_qp_spec = SPEC_QP( i_qp );
  593.      ALIGNED_ARRAY_16( dctcoef, dct8x8,[64] );
  594.  
  595.      if( h->mb.b_lossless )
  596. @@ -178,7 +184,7 @@ void x264_mb_encode_i8x8( x264_t *h, int idx, int i_qp )
  597.      {
  598.          h->mb.i_cbp_luma |= 1<<idx;
  599.          h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8 );
  600. -        h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8IY], i_qp );
  601. +        h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8IY], i_qp_spec );
  602.          h->dctf.add8x8_idct8( p_dst, dct8x8 );
  603.          STORE_8x8_NNZ( s8, 1 );
  604.      }
  605. @@ -196,6 +202,7 @@ static void x264_mb_encode_i16x16( x264_t *h, int i_qp )
  606.  
  607.      int nz;
  608.      int decimate_score = h->mb.b_dct_decimate ? 0 : 9;
  609. +    int i_qp_spec = SPEC_QP( i_qp );
  610.  
  611.      if( h->mb.b_lossless )
  612.      {
  613. @@ -218,6 +225,8 @@ static void x264_mb_encode_i16x16( x264_t *h, int i_qp )
  614.      for( int i = 0; i < 16; i++ )
  615.      {
  616.          /* copy dc coeff */
  617. +        if( h->mb.b_noise_reduction )
  618. +            h->quantf.denoise_dct( dct4x4[i], h->nr_residual_sum[0], h->nr_offset[0], 16 );
  619.          dct_dc4x4[block_idx_xy_1d[i]] = dct4x4[i][0];
  620.          dct4x4[i][0] = 0;
  621.  
  622. @@ -227,7 +236,7 @@ static void x264_mb_encode_i16x16( x264_t *h, int i_qp )
  623.          if( nz )
  624.          {
  625.              h->zigzagf.scan_4x4( h->dct.luma4x4[i], dct4x4[i] );
  626. -            h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IY], i_qp );
  627. +            h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IY], i_qp_spec );
  628.              if( decimate_score < 6 ) decimate_score += h->quantf.decimate_score15( h->dct.luma4x4[i] );
  629.              h->mb.i_cbp_luma = 0xf;
  630.          }
  631. @@ -254,7 +263,7 @@ static void x264_mb_encode_i16x16( x264_t *h, int i_qp )
  632.  
  633.          /* output samples to fdec */
  634.          h->dctf.idct4x4dc( dct_dc4x4 );
  635. -        h->quantf.dequant_4x4_dc( dct_dc4x4, h->dequant4_mf[CQM_4IY], i_qp );  /* XXX not inversed */
  636. +        h->quantf.dequant_4x4_dc( dct_dc4x4, h->dequant4_mf[CQM_4IY], i_qp_spec );  /* XXX not inversed */
  637.          if( h->mb.i_cbp_luma )
  638.              for( int i = 0; i < 16; i++ )
  639.                  dct4x4[i][0] = dct_dc4x4[block_idx_xy_1d[i]];
  640. @@ -326,13 +335,15 @@ void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qp )
  641.  {
  642.      int nz, nz_dc;
  643.      int b_decimate = b_inter && h->mb.b_dct_decimate;
  644. +    int i_qp_spec = SPEC_QP( i_qp );
  645.      ALIGNED_ARRAY_16( dctcoef, dct2x2,[4] );
  646.      h->mb.i_cbp_chroma = 0;
  647. +    h->nr_count[2] += h->mb.b_noise_reduction * 4;
  648.  
  649.      /* Early termination: check variance of chroma residual before encoding.
  650.       * Don't bother trying early termination at low QPs.
  651.       * Values are experimentally derived. */
  652. -    if( b_decimate && i_qp >= (h->mb.b_trellis ? 12 : 18) )
  653. +    if( b_decimate && i_qp >= (h->mb.b_trellis ? 12 : 18) && !h->mb.b_noise_reduction )
  654.      {
  655.          int thresh = (x264_lambda2_tab[i_qp] + 32) >> 6;
  656.          int ssd[2];
  657. @@ -363,11 +374,11 @@ void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qp )
  658.  
  659.                      if( nz_dc )
  660.                      {
  661. -                        if( !x264_mb_optimize_chroma_dc( h, b_inter, i_qp, dct2x2 ) )
  662. +                        if( !x264_mb_optimize_chroma_dc( h, b_inter, i_qp_spec, dct2x2 ) )
  663.                              continue;
  664.                          h->mb.cache.non_zero_count[x264_scan8[25]+ch] = 1;
  665.                          zigzag_scan_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );
  666. -                        idct_dequant_2x2_dconly( dct2x2, dct2x2, h->dequant4_mf[CQM_4IC + b_inter], i_qp );
  667. +                        idct_dequant_2x2_dconly( dct2x2, dct2x2, h->dequant4_mf[CQM_4IC + b_inter], i_qp_spec );
  668.                          h->dctf.add8x8_idct_dc( h->mb.pic.p_fdec[1+ch], dct2x2 );
  669.                          h->mb.i_cbp_chroma = 1;
  670.                      }
  671. @@ -401,6 +412,9 @@ void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qp )
  672.          }
  673.  
  674.          h->dctf.sub8x8_dct( dct4x4, p_src, p_dst );
  675. +        if( h->mb.b_noise_reduction )
  676. +            for( int i = 0; i < 4; i++ )
  677. +                h->quantf.denoise_dct( dct4x4[i], h->nr_residual_sum[2], h->nr_offset[2], 16 );
  678.          dct2x2dc( dct2x2, dct4x4 );
  679.          /* calculate dct coeffs */
  680.          for( int i = 0; i < 4; i++ )
  681. @@ -414,7 +428,7 @@ void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qp )
  682.              {
  683.                  nz_ac = 1;
  684.                  h->zigzagf.scan_4x4( h->dct.luma4x4[16+i+ch*4], dct4x4[i] );
  685. -                h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IC + b_inter], i_qp );
  686. +                h->quantf.dequant_4x4( dct4x4[i], h->dequant4_mf[CQM_4IC + b_inter], i_qp_spec );
  687.                  if( b_decimate )
  688.                      i_decimate_score += h->quantf.decimate_score15( h->dct.luma4x4[16+i+ch*4] );
  689.              }
  690. @@ -443,7 +457,7 @@ void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qp )
  691.              }
  692.              /* DC-only */
  693.              zigzag_scan_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );
  694. -            idct_dequant_2x2_dconly( dct2x2, dct2x2, h->dequant4_mf[CQM_4IC + b_inter], i_qp );
  695. +            idct_dequant_2x2_dconly( dct2x2, dct2x2, h->dequant4_mf[CQM_4IC + b_inter], i_qp_spec );
  696.              h->dctf.add8x8_idct_dc( p_dst, dct2x2 );
  697.          }
  698.          else
  699. @@ -452,7 +466,7 @@ void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qp )
  700.              if( nz_dc )
  701.              {
  702.                  zigzag_scan_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );
  703. -                idct_dequant_2x2_dc( dct2x2, dct4x4, h->dequant4_mf[CQM_4IC + b_inter], i_qp );
  704. +                idct_dequant_2x2_dc( dct2x2, dct4x4, h->dequant4_mf[CQM_4IC + b_inter], i_qp_spec );
  705.              }
  706.              h->dctf.add8x8_idct( p_dst, dct4x4 );
  707.          }
  708. @@ -587,6 +601,7 @@ void x264_macroblock_encode( x264_t *h )
  709.      int b_decimate = h->mb.b_dct_decimate;
  710.      int b_force_no_skip = 0;
  711.      int nz;
  712. +    int i_qp_spec = SPEC_QP( i_qp );
  713.      h->mb.i_cbp_luma = 0;
  714.      h->mb.cache.non_zero_count[x264_scan8[24]] = 0;
  715.  
  716. @@ -748,8 +763,6 @@ void x264_macroblock_encode( x264_t *h )
  717.  
  718.              for( int idx = 0; idx < 4; idx++ )
  719.              {
  720. -                if( h->mb.b_noise_reduction )
  721. -                    h->quantf.denoise_dct( dct8x8[idx], h->nr_residual_sum[1], h->nr_offset[1], 64 );
  722.                  nz = x264_quant_8x8( h, dct8x8[idx], i_qp, 0, idx );
  723.  
  724.                  if( nz )
  725. @@ -782,7 +795,7 @@ void x264_macroblock_encode( x264_t *h )
  726.  
  727.                      if( h->mb.i_cbp_luma&(1<<idx) )
  728.                      {
  729. -                        h->quantf.dequant_8x8( dct8x8[idx], h->dequant8_mf[CQM_8PY], i_qp );
  730. +                        h->quantf.dequant_8x8( dct8x8[idx], h->dequant8_mf[CQM_8PY], i_qp_spec );
  731.                          h->dctf.add8x8_idct8( &h->mb.pic.p_fdec[0][8*x + 8*y*FDEC_STRIDE], dct8x8[idx] );
  732.                          STORE_8x8_NNZ( s8, 1 );
  733.                      }
  734. @@ -807,15 +820,13 @@ void x264_macroblock_encode( x264_t *h )
  735.                  {
  736.                      int idx = i8x8 * 4 + i4x4;
  737.  
  738. -                    if( h->mb.b_noise_reduction )
  739. -                        h->quantf.denoise_dct( dct4x4[idx], h->nr_residual_sum[0], h->nr_offset[0], 16 );
  740.                      nz = x264_quant_4x4( h, dct4x4[idx], i_qp, DCT_LUMA_4x4, 0, idx );
  741.                      h->mb.cache.non_zero_count[x264_scan8[idx]] = nz;
  742.  
  743.                      if( nz )
  744.                      {
  745.                          h->zigzagf.scan_4x4( h->dct.luma4x4[idx], dct4x4[idx] );
  746. -                        h->quantf.dequant_4x4( dct4x4[idx], h->dequant4_mf[CQM_4PY], i_qp );
  747. +                        h->quantf.dequant_4x4( dct4x4[idx], h->dequant4_mf[CQM_4PY], i_qp_spec );
  748.                          if( b_decimate && i_decimate_8x8 < 6 )
  749.                              i_decimate_8x8 += h->quantf.decimate_score16( h->dct.luma4x4[idx] );
  750.                          cbp = 1;
  751. @@ -1019,12 +1030,16 @@ int x264_macroblock_probe_skip( x264_t *h, int b_bidir )
  752.  
  753.  void x264_noise_reduction_update( x264_t *h )
  754.  {
  755. -    for( int cat = 0; cat < 2; cat++ )
  756. +    h->nr_offset = h->nr_offset_denoise;
  757. +    h->nr_residual_sum = h->nr_residual_sum_buf[0];
  758. +    h->nr_count = h->nr_count_buf[0];
  759. +    for( int cat = 0; cat < 3; cat++ )
  760.      {
  761. -        int size = cat ? 64 : 16;
  762. -        const uint16_t *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
  763. +        int dct8x8 = cat == 1;
  764. +        int size = dct8x8 ? 64 : 16;
  765. +        const uint16_t *weight = dct8x8 ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
  766.  
  767. -        if( h->nr_count[cat] > (cat ? (1<<16) : (1<<18)) )
  768. +        if( h->nr_count[cat] > (dct8x8 ? (1<<16) : (1<<18)) )
  769.          {
  770.              for( int i = 0; i < size; i++ )
  771.                  h->nr_residual_sum[cat][i] >>= 1;
  772. @@ -1036,6 +1051,9 @@ void x264_noise_reduction_update( x264_t *h )
  773.                  ((uint64_t)h->param.analyse.i_noise_reduction * h->nr_count[cat]
  774.                   + h->nr_residual_sum[cat][i]/2)
  775.                / ((uint64_t)h->nr_residual_sum[cat][i] * weight[i]/256 + 1);
  776. +
  777. +        /* Don't denoise DC coefficients */
  778. +        h->nr_offset[cat][0] = 0;
  779.      }
  780.  }
  781.  
  782. @@ -1054,6 +1072,7 @@ void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
  783.      int b_decimate = h->mb.b_dct_decimate;
  784.      int nnz8x8 = 0;
  785.      int nz;
  786. +    int i_qp_spec = SPEC_QP( i_qp );
  787.  
  788.      if( !h->mb.b_skip_mc )
  789.          x264_mb_mc_8x8( h, i8 );
  790. @@ -1101,7 +1120,7 @@ void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
  791.  
  792.                  if( nnz8x8 )
  793.                  {
  794. -                    h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8PY], i_qp );
  795. +                    h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8PY], i_qp_spec );
  796.                      h->dctf.add8x8_idct8( p_fdec, dct8x8 );
  797.                      STORE_8x8_NNZ( s8, 1 );
  798.                  }
  799. @@ -1123,7 +1142,7 @@ void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
  800.                  if( nz )
  801.                  {
  802.                      h->zigzagf.scan_4x4( h->dct.luma4x4[i8*4+i4], dct4x4[i4] );
  803. -                    h->quantf.dequant_4x4( dct4x4[i4], h->dequant4_mf[CQM_4PY], i_qp );
  804. +                    h->quantf.dequant_4x4( dct4x4[i4], h->dequant4_mf[CQM_4PY], i_qp_spec );
  805.                      if( b_decimate )
  806.                          i_decimate_8x8 += h->quantf.decimate_score16( h->dct.luma4x4[i8*4+i4] );
  807.                      nnz8x8 = 1;
  808. @@ -1146,8 +1165,9 @@ void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
  809.              ALIGNED_ARRAY_16( dctcoef, dct4x4,[16] );
  810.              p_fenc = h->mb.pic.p_fenc[1+ch] + 4*x + 4*y*FENC_STRIDE;
  811.              p_fdec = h->mb.pic.p_fdec[1+ch] + 4*x + 4*y*FDEC_STRIDE;
  812. -
  813.              h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
  814. +            if( h->mb.b_noise_reduction );
  815. +                h->quantf.denoise_dct( dct4x4, h->nr_residual_sum[2], h->nr_offset[2], 16 );
  816.              dct4x4[0] = 0;
  817.  
  818.              if( h->mb.b_trellis )
  819. @@ -1159,7 +1179,7 @@ void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
  820.              if( nz )
  821.              {
  822.                  h->zigzagf.scan_4x4( h->dct.luma4x4[16+i8+ch*4], dct4x4 );
  823. -                h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp );
  824. +                h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp_spec );
  825.                  h->dctf.add4x4_idct( p_fdec, dct4x4 );
  826.              }
  827.          }
  828. diff --git a/encoder/me.c b/encoder/me.c
  829. index 90f7dfd..9ca04d0 100644
  830. --- a/encoder/me.c
  831. +++ b/encoder/me.c
  832. @@ -597,7 +597,7 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc,
  833.              int delta = x264_pixel_size[sad_size].w;
  834.              int16_t *xs = h->scratch_buffer;
  835.              int xn;
  836. -            uint16_t *cost_fpel_mvx = h->cost_mv_fpel[x264_lambda_tab[h->mb.i_qp]][-m->mvp[0]&3] + (-m->mvp[0]>>2);
  837. +            uint16_t *cost_fpel_mvx = h->cost_mv_fpel[h->mb.i_qp][-m->mvp[0]&3] + (-m->mvp[0]>>2);
  838.  
  839.              h->pixf.sad_x4[sad_size]( zero, p_fenc, p_fenc+delta,
  840.                  p_fenc+delta*FENC_STRIDE, p_fenc+delta+delta*FENC_STRIDE,
  841. diff --git a/encoder/me.h b/encoder/me.h
  842. index fd99a5d..58cd3e0 100644
  843. --- a/encoder/me.h
  844. +++ b/encoder/me.h
  845. @@ -71,7 +71,7 @@ void x264_me_refine_bidir_rd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_wei
  846.  void x264_me_refine_bidir_satd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight );
  847.  uint64_t x264_rd_cost_part( x264_t *h, int i_lambda2, int i8, int i_pixel );
  848.  
  849. -extern uint16_t *x264_cost_mv_fpel[LAMBDA_MAX+1][4];
  850. +extern uint16_t *x264_cost_mv_fpel[QP_MAX+1][4];
  851.  
  852.  #define COPY1_IF_LT(x,y)\
  853.  if((y)<(x))\
  854. diff --git a/encoder/ratecontrol.c b/encoder/ratecontrol.c
  855. index 8f7ef9a..e727ab0 100644
  856. --- a/encoder/ratecontrol.c
  857. +++ b/encoder/ratecontrol.c
  858. @@ -1441,8 +1441,14 @@ int x264_ratecontrol_mb_qp( x264_t *h )
  859.      x264_emms();
  860.      float qp = h->rc->qpm;
  861.      if( h->param.rc.i_aq_mode )
  862. -        /* MB-tree currently doesn't adjust quantizers in unreferenced frames. */
  863. -        qp += h->fdec->b_kept_as_ref ? h->fenc->f_qp_offset[h->mb.i_mb_xy] : h->fenc->f_qp_offset_aq[h->mb.i_mb_xy];
  864. +    {
  865. +         /* MB-tree currently doesn't adjust quantizers in unreferenced frames. */
  866. +        float qp_offset = h->fdec->b_kept_as_ref ? h->fenc->f_qp_offset[h->mb.i_mb_xy] : h->fenc->f_qp_offset_aq[h->mb.i_mb_xy];
  867. +        /* Scale AQ's effect towards zero in emergency mode. */
  868. +        if( qp > QP_MAX_SPEC )
  869. +            qp_offset *= (QP_MAX - qp) / (QP_MAX_SPEC - QP_MAX);
  870. +        qp += qp_offset;
  871. +    }
  872.      return x264_clip3( qp + .5, h->param.rc.i_qp_min, h->param.rc.i_qp_max );
  873.  }
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!