You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

454 lines
19KB

  1. /*
  2. * DSP utils
  3. * Copyright (c) 2000, 2001 Gerard Lantau.
  4. *
  5. * This program is free software; you can redistribute it and/or modify
  6. * it under the terms of the GNU General Public License as published by
  7. * the Free Software Foundation; either version 2 of the License, or
  8. * (at your option) any later version.
  9. *
  10. * This program is distributed in the hope that it will be useful,
  11. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13. * GNU General Public License for more details.
  14. *
  15. * You should have received a copy of the GNU General Public License
  16. * along with this program; if not, write to the Free Software
  17. * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  18. */
  19. #include <stdlib.h>
  20. #include <stdio.h>
  21. #include "avcodec.h"
  22. #include "dsputil.h"
  23. void (*ff_idct)(DCTELEM *block);
  24. void (*get_pixels)(DCTELEM *block, const UINT8 *pixels, int line_size);
  25. void (*put_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size);
  26. void (*add_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size);
  27. op_pixels_abs_func pix_abs16x16;
  28. op_pixels_abs_func pix_abs16x16_x2;
  29. op_pixels_abs_func pix_abs16x16_y2;
  30. op_pixels_abs_func pix_abs16x16_xy2;
  31. UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
  32. UINT32 squareTbl[512];
  33. extern UINT16 default_intra_matrix[64];
  34. extern UINT16 default_non_intra_matrix[64];
  35. UINT8 zigzag_direct[64] = {
  36. 0, 1, 8, 16, 9, 2, 3, 10,
  37. 17, 24, 32, 25, 18, 11, 4, 5,
  38. 12, 19, 26, 33, 40, 48, 41, 34,
  39. 27, 20, 13, 6, 7, 14, 21, 28,
  40. 35, 42, 49, 56, 57, 50, 43, 36,
  41. 29, 22, 15, 23, 30, 37, 44, 51,
  42. 58, 59, 52, 45, 38, 31, 39, 46,
  43. 53, 60, 61, 54, 47, 55, 62, 63
  44. };
  45. UINT8 ff_alternate_horizontal_scan[64] = {
  46. 0, 1, 2, 3, 8, 9, 16, 17,
  47. 10, 11, 4, 5, 6, 7, 15, 14,
  48. 13, 12, 19, 18, 24, 25, 32, 33,
  49. 26, 27, 20, 21, 22, 23, 28, 29,
  50. 30, 31, 34, 35, 40, 41, 48, 49,
  51. 42, 43, 36, 37, 38, 39, 44, 45,
  52. 46, 47, 50, 51, 56, 57, 58, 59,
  53. 52, 53, 54, 55, 60, 61, 62, 63,
  54. };
  55. UINT8 ff_alternate_vertical_scan[64] = {
  56. 0, 8, 16, 24, 1, 9, 2, 10,
  57. 17, 25, 32, 40, 48, 56, 57, 49,
  58. 41, 33, 26, 18, 3, 11, 4, 12,
  59. 19, 27, 34, 42, 50, 58, 35, 43,
  60. 51, 59, 20, 28, 5, 13, 6, 14,
  61. 21, 29, 36, 44, 52, 60, 37, 45,
  62. 53, 61, 22, 30, 7, 15, 23, 31,
  63. 38, 46, 54, 62, 39, 47, 55, 63,
  64. };
  65. void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size)
  66. {
  67. DCTELEM *p;
  68. const UINT8 *pix;
  69. int i;
  70. /* read the pixels */
  71. p = block;
  72. pix = pixels;
  73. for(i=0;i<8;i++) {
  74. p[0] = pix[0];
  75. p[1] = pix[1];
  76. p[2] = pix[2];
  77. p[3] = pix[3];
  78. p[4] = pix[4];
  79. p[5] = pix[5];
  80. p[6] = pix[6];
  81. p[7] = pix[7];
  82. pix += line_size;
  83. p += 8;
  84. }
  85. }
  86. void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size)
  87. {
  88. const DCTELEM *p;
  89. UINT8 *pix;
  90. int i;
  91. UINT8 *cm = cropTbl + MAX_NEG_CROP;
  92. /* read the pixels */
  93. p = block;
  94. pix = pixels;
  95. for(i=0;i<8;i++) {
  96. pix[0] = cm[p[0]];
  97. pix[1] = cm[p[1]];
  98. pix[2] = cm[p[2]];
  99. pix[3] = cm[p[3]];
  100. pix[4] = cm[p[4]];
  101. pix[5] = cm[p[5]];
  102. pix[6] = cm[p[6]];
  103. pix[7] = cm[p[7]];
  104. pix += line_size;
  105. p += 8;
  106. }
  107. }
  108. void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size)
  109. {
  110. const DCTELEM *p;
  111. UINT8 *pix;
  112. int i;
  113. UINT8 *cm = cropTbl + MAX_NEG_CROP;
  114. /* read the pixels */
  115. p = block;
  116. pix = pixels;
  117. for(i=0;i<8;i++) {
  118. pix[0] = cm[pix[0] + p[0]];
  119. pix[1] = cm[pix[1] + p[1]];
  120. pix[2] = cm[pix[2] + p[2]];
  121. pix[3] = cm[pix[3] + p[3]];
  122. pix[4] = cm[pix[4] + p[4]];
  123. pix[5] = cm[pix[5] + p[5]];
  124. pix[6] = cm[pix[6] + p[6]];
  125. pix[7] = cm[pix[7] + p[7]];
  126. pix += line_size;
  127. p += 8;
  128. }
  129. }
  130. #define PIXOP(BTYPE, OPNAME, OP, INCR) \
  131. \
  132. static void OPNAME ## _pixels(BTYPE *block, const UINT8 *pixels, int line_size, int h) \
  133. { \
  134. BTYPE *p; \
  135. const UINT8 *pix; \
  136. \
  137. p = block; \
  138. pix = pixels; \
  139. do { \
  140. OP(p[0], pix[0]); \
  141. OP(p[1], pix[1]); \
  142. OP(p[2], pix[2]); \
  143. OP(p[3], pix[3]); \
  144. OP(p[4], pix[4]); \
  145. OP(p[5], pix[5]); \
  146. OP(p[6], pix[6]); \
  147. OP(p[7], pix[7]); \
  148. pix += line_size; \
  149. p += INCR; \
  150. } while (--h);; \
  151. } \
  152. \
  153. static void OPNAME ## _pixels_x2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \
  154. { \
  155. BTYPE *p; \
  156. const UINT8 *pix; \
  157. \
  158. p = block; \
  159. pix = pixels; \
  160. do { \
  161. OP(p[0], avg2(pix[0], pix[1])); \
  162. OP(p[1], avg2(pix[1], pix[2])); \
  163. OP(p[2], avg2(pix[2], pix[3])); \
  164. OP(p[3], avg2(pix[3], pix[4])); \
  165. OP(p[4], avg2(pix[4], pix[5])); \
  166. OP(p[5], avg2(pix[5], pix[6])); \
  167. OP(p[6], avg2(pix[6], pix[7])); \
  168. OP(p[7], avg2(pix[7], pix[8])); \
  169. pix += line_size; \
  170. p += INCR; \
  171. } while (--h); \
  172. } \
  173. \
  174. static void OPNAME ## _pixels_y2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \
  175. { \
  176. BTYPE *p; \
  177. const UINT8 *pix; \
  178. const UINT8 *pix1; \
  179. \
  180. p = block; \
  181. pix = pixels; \
  182. pix1 = pixels + line_size; \
  183. do { \
  184. OP(p[0], avg2(pix[0], pix1[0])); \
  185. OP(p[1], avg2(pix[1], pix1[1])); \
  186. OP(p[2], avg2(pix[2], pix1[2])); \
  187. OP(p[3], avg2(pix[3], pix1[3])); \
  188. OP(p[4], avg2(pix[4], pix1[4])); \
  189. OP(p[5], avg2(pix[5], pix1[5])); \
  190. OP(p[6], avg2(pix[6], pix1[6])); \
  191. OP(p[7], avg2(pix[7], pix1[7])); \
  192. pix += line_size; \
  193. pix1 += line_size; \
  194. p += INCR; \
  195. } while(--h); \
  196. } \
  197. \
  198. static void OPNAME ## _pixels_xy2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \
  199. { \
  200. BTYPE *p; \
  201. const UINT8 *pix; \
  202. const UINT8 *pix1; \
  203. \
  204. p = block; \
  205. pix = pixels; \
  206. pix1 = pixels + line_size; \
  207. do { \
  208. OP(p[0], avg4(pix[0], pix[1], pix1[0], pix1[1])); \
  209. OP(p[1], avg4(pix[1], pix[2], pix1[1], pix1[2])); \
  210. OP(p[2], avg4(pix[2], pix[3], pix1[2], pix1[3])); \
  211. OP(p[3], avg4(pix[3], pix[4], pix1[3], pix1[4])); \
  212. OP(p[4], avg4(pix[4], pix[5], pix1[4], pix1[5])); \
  213. OP(p[5], avg4(pix[5], pix[6], pix1[5], pix1[6])); \
  214. OP(p[6], avg4(pix[6], pix[7], pix1[6], pix1[7])); \
  215. OP(p[7], avg4(pix[7], pix[8], pix1[7], pix1[8])); \
  216. pix += line_size; \
  217. pix1 += line_size; \
  218. p += INCR; \
  219. } while(--h); \
  220. } \
  221. \
  222. void (*OPNAME ## _pixels_tab[4])(BTYPE *block, const UINT8 *pixels, int line_size, int h) = { \
  223. OPNAME ## _pixels, \
  224. OPNAME ## _pixels_x2, \
  225. OPNAME ## _pixels_y2, \
  226. OPNAME ## _pixels_xy2, \
  227. };
  228. /* rounding primitives */
  229. #define avg2(a,b) ((a+b+1)>>1)
  230. #define avg4(a,b,c,d) ((a+b+c+d+2)>>2)
  231. #define op_put(a, b) a = b
  232. #define op_avg(a, b) a = avg2(a, b)
  233. #define op_sub(a, b) a -= b
  234. PIXOP(UINT8, put, op_put, line_size)
  235. PIXOP(UINT8, avg, op_avg, line_size)
  236. PIXOP(DCTELEM, sub, op_sub, 8)
  237. /* not rounding primitives */
  238. #undef avg2
  239. #undef avg4
  240. #define avg2(a,b) ((a+b)>>1)
  241. #define avg4(a,b,c,d) ((a+b+c+d+1)>>2)
  242. PIXOP(UINT8, put_no_rnd, op_put, line_size)
  243. PIXOP(UINT8, avg_no_rnd, op_avg, line_size)
  244. /* motion estimation */
  245. #undef avg2
  246. #undef avg4
  247. #define avg2(a,b) ((a+b+1)>>1)
  248. #define avg4(a,b,c,d) ((a+b+c+d+2)>>2)
  249. int pix_abs16x16_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h)
  250. {
  251. int s, i;
  252. s = 0;
  253. for(i=0;i<h;i++) {
  254. s += abs(pix1[0] - pix2[0]);
  255. s += abs(pix1[1] - pix2[1]);
  256. s += abs(pix1[2] - pix2[2]);
  257. s += abs(pix1[3] - pix2[3]);
  258. s += abs(pix1[4] - pix2[4]);
  259. s += abs(pix1[5] - pix2[5]);
  260. s += abs(pix1[6] - pix2[6]);
  261. s += abs(pix1[7] - pix2[7]);
  262. s += abs(pix1[8] - pix2[8]);
  263. s += abs(pix1[9] - pix2[9]);
  264. s += abs(pix1[10] - pix2[10]);
  265. s += abs(pix1[11] - pix2[11]);
  266. s += abs(pix1[12] - pix2[12]);
  267. s += abs(pix1[13] - pix2[13]);
  268. s += abs(pix1[14] - pix2[14]);
  269. s += abs(pix1[15] - pix2[15]);
  270. pix1 += line_size;
  271. pix2 += line_size;
  272. }
  273. return s;
  274. }
  275. int pix_abs16x16_x2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h)
  276. {
  277. int s, i;
  278. s = 0;
  279. for(i=0;i<h;i++) {
  280. s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
  281. s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
  282. s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
  283. s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
  284. s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
  285. s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
  286. s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
  287. s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
  288. s += abs(pix1[8] - avg2(pix2[8], pix2[9]));
  289. s += abs(pix1[9] - avg2(pix2[9], pix2[10]));
  290. s += abs(pix1[10] - avg2(pix2[10], pix2[11]));
  291. s += abs(pix1[11] - avg2(pix2[11], pix2[12]));
  292. s += abs(pix1[12] - avg2(pix2[12], pix2[13]));
  293. s += abs(pix1[13] - avg2(pix2[13], pix2[14]));
  294. s += abs(pix1[14] - avg2(pix2[14], pix2[15]));
  295. s += abs(pix1[15] - avg2(pix2[15], pix2[16]));
  296. pix1 += line_size;
  297. pix2 += line_size;
  298. }
  299. return s;
  300. }
  301. int pix_abs16x16_y2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h)
  302. {
  303. int s, i;
  304. UINT8 *pix3 = pix2 + line_size;
  305. s = 0;
  306. for(i=0;i<h;i++) {
  307. s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
  308. s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
  309. s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
  310. s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
  311. s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
  312. s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
  313. s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
  314. s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
  315. s += abs(pix1[8] - avg2(pix2[8], pix3[8]));
  316. s += abs(pix1[9] - avg2(pix2[9], pix3[9]));
  317. s += abs(pix1[10] - avg2(pix2[10], pix3[10]));
  318. s += abs(pix1[11] - avg2(pix2[11], pix3[11]));
  319. s += abs(pix1[12] - avg2(pix2[12], pix3[12]));
  320. s += abs(pix1[13] - avg2(pix2[13], pix3[13]));
  321. s += abs(pix1[14] - avg2(pix2[14], pix3[14]));
  322. s += abs(pix1[15] - avg2(pix2[15], pix3[15]));
  323. pix1 += line_size;
  324. pix2 += line_size;
  325. pix3 += line_size;
  326. }
  327. return s;
  328. }
  329. int pix_abs16x16_xy2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h)
  330. {
  331. int s, i;
  332. UINT8 *pix3 = pix2 + line_size;
  333. s = 0;
  334. for(i=0;i<h;i++) {
  335. s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
  336. s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
  337. s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
  338. s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
  339. s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
  340. s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
  341. s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
  342. s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
  343. s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9]));
  344. s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10]));
  345. s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));
  346. s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));
  347. s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));
  348. s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));
  349. s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));
  350. s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));
  351. pix1 += line_size;
  352. pix2 += line_size;
  353. pix3 += line_size;
  354. }
  355. return s;
  356. }
  357. /* permute block according so that it corresponds to the MMX idct
  358. order */
  359. void block_permute(INT16 *block)
  360. {
  361. int tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
  362. int i;
  363. for(i=0;i<8;i++) {
  364. tmp1 = block[1];
  365. tmp2 = block[2];
  366. tmp3 = block[3];
  367. tmp4 = block[4];
  368. tmp5 = block[5];
  369. tmp6 = block[6];
  370. block[1] = tmp2;
  371. block[2] = tmp4;
  372. block[3] = tmp6;
  373. block[4] = tmp1;
  374. block[5] = tmp3;
  375. block[6] = tmp5;
  376. block += 8;
  377. }
  378. }
  379. void dsputil_init(void)
  380. {
  381. int i, j;
  382. for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;
  383. for(i=0;i<MAX_NEG_CROP;i++) {
  384. cropTbl[i] = 0;
  385. cropTbl[i + MAX_NEG_CROP + 256] = 255;
  386. }
  387. for(i=0;i<512;i++) {
  388. squareTbl[i] = (i - 256) * (i - 256);
  389. }
  390. ff_idct = j_rev_dct;
  391. get_pixels = get_pixels_c;
  392. put_pixels_clamped = put_pixels_clamped_c;
  393. add_pixels_clamped = add_pixels_clamped_c;
  394. pix_abs16x16 = pix_abs16x16_c;
  395. pix_abs16x16_x2 = pix_abs16x16_x2_c;
  396. pix_abs16x16_y2 = pix_abs16x16_y2_c;
  397. pix_abs16x16_xy2 = pix_abs16x16_xy2_c;
  398. av_fdct = jpeg_fdct_ifast;
  399. /* permute for IDCT */
  400. for(i=0;i<64;i++) {
  401. j = zigzag_direct[i];
  402. zigzag_direct[i] = block_permute_op(j);
  403. j = ff_alternate_horizontal_scan[i];
  404. ff_alternate_horizontal_scan[i] = block_permute_op(j);
  405. j = ff_alternate_vertical_scan[i];
  406. ff_alternate_vertical_scan[i] = block_permute_op(j);
  407. }
  408. block_permute(default_intra_matrix);
  409. block_permute(default_non_intra_matrix);
  410. #ifdef HAVE_MMX
  411. dsputil_init_mmx();
  412. #endif
  413. }