API tools faq
paste
Advertisement
Guest User

Dark Shikari

a guest
Sep 12th, 2008
100
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 9.42 KB | None | 0 0
raw download clone embed print report
  1. Index: encoder/macroblock.c
  2. ===================================================================
  3. --- encoder/macroblock.c (revision 736)
  4. +++ encoder/macroblock.c (working copy)
  5. @@ -96,7 +96,10 @@
  6. h->dctf.sub4x4_dct( dct4x4, p_src, p_dst );
  7.  
  8. if( h->mb.b_trellis )
  9. + {
  10. x264_quant_4x4_trellis( h, dct4x4, CQM_4IY, i_qscale, DCT_LUMA_4x4, 1 );
  11. + x264_quant_4x4_qns( h, dct4x4, i_qscale, idx, 1, DCT_LUMA_4x4 );
  12. + }
  13. else
  14. h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
  15.  
  16. @@ -118,7 +121,10 @@
  17. h->dctf.sub8x8_dct8( dct8x8, p_src, p_dst );
  18.  
  19. if( h->mb.b_trellis )
  20. + {
  21. x264_quant_8x8_trellis( h, dct8x8, CQM_8IY, i_qscale, 1 );
  22. + x264_quant_8x8_qns( h, dct8x8, i_qscale, idx, 1 );
  23. + }
  24. else
  25. h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8IY][i_qscale], h->quant8_bias[CQM_8IY][i_qscale] );
  26.  
  27. @@ -158,7 +164,10 @@
  28.  
  29. /* quant/scan/dequant */
  30. if( h->mb.b_trellis )
  31. + {
  32. x264_quant_4x4_trellis( h, dct4x4[1+i], CQM_4IY, i_qscale, DCT_LUMA_AC, 1 );
  33. + x264_quant_4x4_qns( h, dct4x4[1+i], i_qscale, i, 1, DCT_LUMA_AC );
  34. + }
  35. else
  36. h->quantf.quant_4x4( dct4x4[1+i], h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
  37.  
  38. @@ -425,7 +434,10 @@
  39. if( h->mb.b_noise_reduction )
  40. x264_denoise_dct( h, (int16_t*)dct8x8[idx] );
  41. if( h->mb.b_trellis )
  42. + {
  43. x264_quant_8x8_trellis( h, dct8x8[idx], CQM_8PY, i_qp, 0 );
  44. + x264_quant_8x8_qns( h, dct8x8[idx], i_qp, idx, 0 );
  45. + }
  46. else
  47. h->quantf.quant_8x8( dct8x8[idx], h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
  48.  
  49. @@ -477,7 +489,10 @@
  50. if( h->mb.b_noise_reduction )
  51. x264_denoise_dct( h, (int16_t*)dct4x4[idx] );
  52. if( h->mb.b_trellis )
  53. + {
  54. x264_quant_4x4_trellis( h, dct4x4[idx], CQM_4PY, i_qp, DCT_LUMA_4x4, 0 );
  55. + x264_quant_4x4_qns( h, dct4x4[idx], i_qp, idx, 0, DCT_LUMA_4x4 );
  56. + }
  57. else
  58. h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
  59.  
  60. Index: encoder/rdo.c
  61. ===================================================================
  62. --- encoder/rdo.c (revision 736)
  63. +++ encoder/rdo.c (working copy)
  64. @@ -503,7 +503,183 @@
  65. }
  66. }
  67.  
  68. +inline int variance( uint8_t *pix, int x, int y, int start )
  69. +{
  70. + uint8_t *p_src = pix + (x + start) + FENC_STRIDE * (y + start);
  71. + int sum = 0; int ssd = 0; int i,j;
  72. + for(i = 0; i < 3; i++)
  73. + for(j = 0; j < 3; j++)
  74. + {
  75. + int cur = p_src[j+i*FENC_STRIDE];
  76. + sum += cur; ssd += cur*cur;
  77. + }
  78. + return (9*ssd - (sum * sum));
  79. +}
  80.  
  81. +uint64_t weighted_error8( x264_t *h, int16_t dct[8][8], uint16_t inv_variance[8][8], int i_qp, int idx, int i_quant_cat)
  82. +{
  83. + uint8_t idct[8*FDEC_STRIDE];
  84. + int16_t ndct[8][8];
  85. + int i,j;
  86. + uint8_t *decpix = h->mb.pic.p_fdec[0]+(8*(idx&1))+FDEC_STRIDE*(((idx>>1)&1)<<3);
  87. + h->mc.copy[PIXEL_8x8]( idct, FDEC_STRIDE, decpix, FDEC_STRIDE, 8 );
  88. + memcpy( ndct, dct, sizeof(ndct) );
  89. + h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct );
  90. + h->quantf.dequant_8x8( ndct, h->dequant8_mf[CQM_8PY], i_qp );
  91. + h->dctf.add8x8_idct8( idct, ndct );
  92. + uint64_t error = 0;
  93. + uint8_t *pix = h->mb.pic.p_fenc[0]+(8*(idx&1))+FENC_STRIDE*(((idx>>1)&1)<<3);
  94. + for( i = 0; i < 8; i++ )
  95. + for( j = 0; j < 8; j++ )
  96. + {
  97. + int pix_error = idct[i+j*FDEC_STRIDE] - pix[i+j*FENC_STRIDE];
  98. + error += (pix_error*pix_error)*inv_variance[j][i];
  99. + }
  100. + x264_cabac_t cabac_tmp = h->cabac;
  101. + block_residual_write_cabac(h,&cabac_tmp,DCT_LUMA_8x8,idx,h->dct.luma8x8[idx],64);
  102. + error *= 38;
  103. + uint64_t bits = cabac_tmp.f8_bits_encoded;
  104. + bits *= lambda2_tab[i_quant_cat][i_qp];
  105. + error += (bits + 8) >> 4;
  106. + return error;
  107. +}
  108. +
  109. +uint64_t weighted_error4( x264_t *h, int16_t dct[4][4], uint16_t inv_variance[4][4], int i_qp, int idx, int i_quant_cat, int i_ctxBlockCat )
  110. +{
  111. + uint8_t idct[4*FDEC_STRIDE];
  112. + int16_t ndct[4][4];
  113. + int i,j;
  114. + uint8_t *decpix = h->mb.pic.p_fdec[0] + 4*block_idx_x[idx] + 4*FDEC_STRIDE*block_idx_y[idx];
  115. + h->mc.copy[PIXEL_4x4]( idct, FDEC_STRIDE, decpix, FDEC_STRIDE, 4 );
  116. + memcpy( ndct, dct, sizeof(ndct) );
  117. + h->zigzagf.scan_4x4( h->dct.block[idx].luma4x4, dct );
  118. + h->quantf.dequant_4x4( ndct, h->dequant4_mf[CQM_4PY], i_qp );
  119. + if( i_ctxBlockCat == DCT_LUMA_AC )
  120. + ndct[0][0] = dct[0][0];
  121. + h->dctf.add4x4_idct( idct, ndct );
  122. + uint64_t error = 0;
  123. + uint8_t *pix = h->mb.pic.p_fenc[0] + 4*block_idx_x[idx] + 4*FENC_STRIDE*block_idx_y[idx];
  124. + for( i = 0; i < 4; i++ )
  125. + for( j = 0; j < 4; j++ )
  126. + {
  127. + int pix_error = idct[i+j*FDEC_STRIDE] - pix[i+j*FENC_STRIDE];
  128. + error += ((pix_error*pix_error)*inv_variance[j][i]);
  129. + }
  130. + x264_cabac_t cabac_tmp = h->cabac;
  131. + block_residual_write_cabac(h,&cabac_tmp,i_ctxBlockCat,idx,h->dct.block[idx].luma4x4,16 - (i_ctxBlockCat == DCT_LUMA_AC));
  132. + error *= 38;
  133. + uint64_t bits = cabac_tmp.f8_bits_encoded;
  134. + bits *= lambda2_tab[i_quant_cat][i_qp];
  135. + error += (bits + 8) >> 4;
  136. + return error;
  137. +}
  138. +
  139. +void x264_quant_8x8_qns( x264_t *h, int16_t dct[8][8], int i_qp, int idx, int i_quant_cat )
  140. +{
  141. + int x,y,start;
  142. + uint16_t inv_variance[8][8];
  143. + int size = 8;
  144. + uint8_t *pix = h->mb.pic.p_fenc[0]+(8*(idx&1))+FENC_STRIDE*(((idx>>1)&1)<<3);
  145. + for(x = 0; x < size; x++)
  146. + for(y = 0; y < size; y++)
  147. + {
  148. + if(x == 0 || y == 0) start = 0;
  149. + else if(x == size - 1 || y == size - 1) start = -2;
  150. + else start = -1;
  151. + inv_variance[y][x] = 256 + 65536 / (variance(pix,x,y,start) + 180);
  152. + inv_variance[y][x] = 800 - inv_variance[y][x];
  153. + }
  154. + int i;
  155. + int runs = 0;
  156. + int changed[64];
  157. + for( ; runs < 1000; runs++)
  158. + {
  159. + int best_change_i = 0;
  160. + int64_t best_change = 0;
  161. + int best_change_amount = 0;
  162. + uint64_t orig_error = weighted_error8(h,dct,inv_variance,i_qp,idx,i_quant_cat);
  163. + for( i = 0; i < 64; i++ )
  164. + {
  165. + // if(changed[i]) continue;
  166. + int change = -1;
  167. + for(; change <= 1; change += 2)
  168. + {
  169. + dct[0][i] += change;
  170. + uint64_t new_error = weighted_error8(h,dct,inv_variance,i_qp,idx,i_quant_cat);
  171. + int64_t cur_change = orig_error - new_error;
  172. + if(cur_change > best_change)
  173. + {
  174. + dct[0][i] -= change;
  175. + best_change = cur_change;
  176. + best_change_i = i;
  177. + best_change_amount = change;
  178. + break;
  179. + }
  180. + dct[0][i] -= change;
  181. + }
  182. + }
  183. + if(best_change == 0) break;
  184. + else
  185. + {
  186. + changed[best_change_i] = 1;
  187. + dct[0][best_change_i] += best_change_amount;
  188. + }
  189. + }
  190. +}
  191. +
  192. +void x264_quant_4x4_qns( x264_t *h, int16_t dct[4][4], int i_qp, int idx, int i_quant_cat, int i_ctxBlockCat )
  193. +{
  194. + int x,y,start;
  195. + uint16_t inv_variance[4][4];
  196. + int size = 4;
  197. + uint8_t *pix = h->mb.pic.p_fenc[0] + 4*block_idx_x[idx] + 4*FENC_STRIDE*block_idx_y[idx];
  198. + for(x = 0; x < size; x++)
  199. + for(y = 0; y < size; y++)
  200. + {
  201. + if(x == 0 || y == 0) start = 0;
  202. + else if(x == size - 1 || y == size - 1) start = -2;
  203. + else start = -1;
  204. + inv_variance[y][x] = 256 + 65536 / (variance(pix,x,y,start) + 180);
  205. + inv_variance[y][x] = 800 - inv_variance[y][x];
  206. + }
  207. + int i;
  208. + int runs = 0;
  209. + int changed[16];
  210. + for( ; runs < 1000; runs++)
  211. + {
  212. + int best_change_i = 0;
  213. + int64_t best_change = 0;
  214. + int best_change_amount = 0;
  215. + uint64_t orig_error = weighted_error4(h,dct,inv_variance,i_qp,idx,i_quant_cat,i_ctxBlockCat);
  216. + for( i = (i_ctxBlockCat == DCT_LUMA_AC); i < 16; i++ )
  217. + {
  218. + //if(changed[i]) continue;
  219. + int change = -1;
  220. + for(; change <= 1; change += 2)
  221. + {
  222. + dct[0][i] += change;
  223. + uint64_t new_error = weighted_error4(h,dct,inv_variance,i_qp,idx,i_quant_cat,i_ctxBlockCat);
  224. + int64_t cur_change = orig_error - new_error;
  225. + if(cur_change > best_change)
  226. + {
  227. + dct[0][i] -= change;
  228. + best_change = cur_change;
  229. + best_change_i = i;
  230. + best_change_amount = change;
  231. + break;
  232. + }
  233. + dct[0][i] -= change;
  234. + }
  235. + }
  236. + if(best_change == 0) break;
  237. + else
  238. + {
  239. + changed[best_change_i] = 1;
  240. + dct[0][best_change_i] += best_change_amount;
  241. + }
  242. + }
  243. +}
  244. +
  245. void x264_quant_4x4_trellis( x264_t *h, int16_t dct[4][4], int i_quant_cat,
  246. int i_qp, int i_ctxBlockCat, int b_intra )
  247. {
  248.  
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!