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.

989 lines
29KB

  1. /*
  2. Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
  3. This program is free software; you can redistribute it and/or modify
  4. it under the terms of the GNU General Public License as published by
  5. the Free Software Foundation; either version 2 of the License, or
  6. (at your option) any later version.
  7. This program is distributed in the hope that it will be useful,
  8. but WITHOUT ANY WARRANTY; without even the implied warranty of
  9. MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  10. GNU General Public License for more details.
  11. You should have received a copy of the GNU General Public License
  12. along with this program; if not, write to the Free Software
  13. Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  14. */
  15. /*
  16. C MMX MMX2 3DNow
  17. isVertDC Ec Ec
  18. isVertMinMaxOk Ec Ec
  19. doVertLowPass E e e
  20. doVertDefFilter Ec Ec e e
  21. isHorizDC Ec Ec
  22. isHorizMinMaxOk a E
  23. doHorizLowPass E e e
  24. doHorizDefFilter Ec Ec e e
  25. deRing E e e*
  26. Vertical RKAlgo1 E a a
  27. Horizontal RKAlgo1 a a
  28. Vertical X1# a E E
  29. Horizontal X1# a E E
  30. LinIpolDeinterlace e E E*
  31. CubicIpolDeinterlace a e e*
  32. LinBlendDeinterlace e E E*
  33. MedianDeinterlace# Ec Ec
  34. TempDeNoiser# E e e
  35. * i dont have a 3dnow CPU -> its untested, but noone said it doesnt work so it seems to work
  36. # more or less selfinvented filters so the exactness isnt too meaningfull
  37. E = Exact implementation
  38. e = allmost exact implementation (slightly different rounding,...)
  39. a = alternative / approximate impl
  40. c = checked against the other implementations (-vo md5)
  41. */
  42. /*
  43. TODO:
  44. remove global/static vars
  45. reduce the time wasted on the mem transfer
  46. implement everything in C at least (done at the moment but ...)
  47. unroll stuff if instructions depend too much on the prior one
  48. we use 8x8 blocks for the horizontal filters, opendivx seems to use 8x4?
  49. move YScale thing to the end instead of fixing QP
  50. write a faster and higher quality deblocking filter :)
  51. make the mainloop more flexible (variable number of blocks at once
  52. (the if/else stuff per block is slowing things down)
  53. compare the quality & speed of all filters
  54. split this huge file
  55. border remover
  56. optimize c versions
  57. try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
  58. smart blur
  59. commandline option for the deblock / dering thresholds
  60. put fastmemcpy back
  61. dont use #ifdef ARCH_X86 for the asm stuff ... cross compilers? (note cpudetect uses ARCH_X86)
  62. ...
  63. */
  64. //Changelog: use the CVS log
  65. #include "../config.h"
  66. #include <inttypes.h>
  67. #include <stdio.h>
  68. #include <stdlib.h>
  69. #include <string.h>
  70. #ifdef HAVE_MALLOC_H
  71. #include <malloc.h>
  72. #endif
  73. //#undef HAVE_MMX2
  74. //#define HAVE_3DNOW
  75. //#undef HAVE_MMX
  76. //#undef ARCH_X86
  77. //#define DEBUG_BRIGHTNESS
  78. //#include "../libvo/fastmemcpy.h"
  79. #include "postprocess.h"
  80. #include "../cpudetect.h"
  81. #define MIN(a,b) ((a) > (b) ? (b) : (a))
  82. #define MAX(a,b) ((a) < (b) ? (b) : (a))
  83. #define ABS(a) ((a) > 0 ? (a) : (-(a)))
  84. #define SIGN(a) ((a) > 0 ? 1 : -1)
  85. #define GET_MODE_BUFFER_SIZE 500
  86. #define OPTIONS_ARRAY_SIZE 10
  87. #ifdef ARCH_X86
  88. #define CAN_COMPILE_X86_ASM
  89. #endif
  90. #ifdef CAN_COMPILE_X86_ASM
  91. static volatile uint64_t __attribute__((aligned(8))) packedYOffset= 0x0000000000000000LL;
  92. static volatile uint64_t __attribute__((aligned(8))) packedYScale= 0x0100010001000100LL;
  93. static uint64_t __attribute__((aligned(8))) w05= 0x0005000500050005LL;
  94. static uint64_t __attribute__((aligned(8))) w20= 0x0020002000200020LL;
  95. static uint64_t __attribute__((aligned(8))) w1400= 0x1400140014001400LL;
  96. static uint64_t __attribute__((aligned(8))) bm00000001= 0x00000000000000FFLL;
  97. static uint64_t __attribute__((aligned(8))) bm00010000= 0x000000FF00000000LL;
  98. static uint64_t __attribute__((aligned(8))) bm00001000= 0x00000000FF000000LL;
  99. static uint64_t __attribute__((aligned(8))) bm10000000= 0xFF00000000000000LL;
  100. static uint64_t __attribute__((aligned(8))) bm10000001= 0xFF000000000000FFLL;
  101. static uint64_t __attribute__((aligned(8))) bm11000011= 0xFFFF00000000FFFFLL;
  102. static uint64_t __attribute__((aligned(8))) bm00000011= 0x000000000000FFFFLL;
  103. static uint64_t __attribute__((aligned(8))) bm11111110= 0xFFFFFFFFFFFFFF00LL;
  104. static uint64_t __attribute__((aligned(8))) bm11000000= 0xFFFF000000000000LL;
  105. static uint64_t __attribute__((aligned(8))) bm00011000= 0x000000FFFF000000LL;
  106. static uint64_t __attribute__((aligned(8))) bm00110011= 0x0000FFFF0000FFFFLL;
  107. static uint64_t __attribute__((aligned(8))) bm11001100= 0xFFFF0000FFFF0000LL;
  108. static uint64_t __attribute__((aligned(8))) b00= 0x0000000000000000LL;
  109. static uint64_t __attribute__((aligned(8))) b01= 0x0101010101010101LL;
  110. static uint64_t __attribute__((aligned(8))) b02= 0x0202020202020202LL;
  111. static uint64_t __attribute__((aligned(8))) b0F= 0x0F0F0F0F0F0F0F0FLL;
  112. static uint64_t __attribute__((aligned(8))) b04= 0x0404040404040404LL;
  113. static uint64_t __attribute__((aligned(8))) b08= 0x0808080808080808LL;
  114. static uint64_t __attribute__((aligned(8))) bFF= 0xFFFFFFFFFFFFFFFFLL;
  115. static uint64_t __attribute__((aligned(8))) b20= 0x2020202020202020LL;
  116. static uint64_t __attribute__((aligned(8))) b80= 0x8080808080808080LL;
  117. static uint64_t __attribute__((aligned(8))) mmxDCOffset= 0x7E7E7E7E7E7E7E7ELL;
  118. static uint64_t __attribute__((aligned(8))) mmxDCThreshold= 0x7C7C7C7C7C7C7C7CLL;
  119. static uint64_t __attribute__((aligned(8))) b3F= 0x3F3F3F3F3F3F3F3FLL;
  120. static uint64_t __attribute__((aligned(8))) temp0=0;
  121. static uint64_t __attribute__((aligned(8))) temp1=0;
  122. static uint64_t __attribute__((aligned(8))) temp2=0;
  123. static uint64_t __attribute__((aligned(8))) temp3=0;
  124. static uint64_t __attribute__((aligned(8))) temp4=0;
  125. static uint64_t __attribute__((aligned(8))) temp5=0;
  126. static uint64_t __attribute__((aligned(8))) pQPb=0;
  127. static uint64_t __attribute__((aligned(8))) pQPb2=0;
  128. static uint8_t __attribute__((aligned(8))) tempBlocks[8*16*2]; //used for the horizontal code
  129. static uint32_t __attribute__((aligned(4))) maxTmpNoise[4];
  130. #else
  131. static uint64_t packedYOffset= 0x0000000000000000LL;
  132. static uint64_t packedYScale= 0x0100010001000100LL;
  133. #endif
  134. extern int divx_quality;
  135. int newPPFlag=0; //is set if -npp is used
  136. struct PPMode gPPMode[GET_PP_QUALITY_MAX+1];
  137. extern int verbose;
  138. int hFlatnessThreshold= 56 - 16;
  139. int vFlatnessThreshold= 56 - 16;
  140. int deringThreshold= 20;
  141. static int dcOffset;
  142. static int dcThreshold;
  143. //amount of "black" u r willing to loose to get a brightness corrected picture
  144. double maxClippedThreshold= 0.01;
  145. static struct PPFilter filters[]=
  146. {
  147. {"hb", "hdeblock", 1, 1, 3, H_DEBLOCK},
  148. {"vb", "vdeblock", 1, 2, 4, V_DEBLOCK},
  149. {"hr", "rkhdeblock", 1, 1, 3, H_RK1_FILTER},
  150. {"vr", "rkvdeblock", 1, 2, 4, V_RK1_FILTER},
  151. {"h1", "x1hdeblock", 1, 1, 3, H_X1_FILTER},
  152. {"v1", "x1vdeblock", 1, 2, 4, V_X1_FILTER},
  153. {"dr", "dering", 1, 5, 6, DERING},
  154. {"al", "autolevels", 0, 1, 2, LEVEL_FIX},
  155. {"lb", "linblenddeint", 1, 1, 4, LINEAR_BLEND_DEINT_FILTER},
  156. {"li", "linipoldeint", 1, 1, 4, LINEAR_IPOL_DEINT_FILTER},
  157. {"ci", "cubicipoldeint", 1, 1, 4, CUBIC_IPOL_DEINT_FILTER},
  158. {"md", "mediandeint", 1, 1, 4, MEDIAN_DEINT_FILTER},
  159. {"tn", "tmpnoise", 1, 7, 8, TEMP_NOISE_FILTER},
  160. {"fq", "forcequant", 1, 0, 0, FORCE_QUANT},
  161. {NULL, NULL,0,0,0,0} //End Marker
  162. };
  163. static char *replaceTable[]=
  164. {
  165. "default", "hdeblock:a,vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",
  166. "de", "hdeblock:a,vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",
  167. "fast", "x1hdeblock:a,x1vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",
  168. "fa", "x1hdeblock:a,x1vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",
  169. NULL //End Marker
  170. };
  171. #ifdef CAN_COMPILE_X86_ASM
  172. static inline void unusedVariableWarningFixer()
  173. {
  174. if(
  175. packedYOffset + packedYScale + w05 + w20 + w1400 + bm00000001 + bm00010000
  176. + bm00001000 + bm10000000 + bm10000001 + bm11000011 + bm00000011 + bm11111110
  177. + bm11000000 + bm00011000 + bm00110011 + bm11001100 + b00 + b01 + b02 + b0F
  178. + bFF + b20 + b04+ b08 + pQPb2 + b80 + mmxDCOffset + mmxDCThreshold + b3F + temp0 + temp1 + temp2 + temp3 + temp4
  179. + temp5 + pQPb== 0) b00=0;
  180. }
  181. #endif
  182. #ifdef TIMING
  183. static inline long long rdtsc()
  184. {
  185. long long l;
  186. asm volatile( "rdtsc\n\t"
  187. : "=A" (l)
  188. );
  189. // printf("%d\n", int(l/1000));
  190. return l;
  191. }
  192. #endif
  193. #ifdef CAN_COMPILE_X86_ASM
  194. static inline void prefetchnta(void *p)
  195. {
  196. asm volatile( "prefetchnta (%0)\n\t"
  197. : : "r" (p)
  198. );
  199. }
  200. static inline void prefetcht0(void *p)
  201. {
  202. asm volatile( "prefetcht0 (%0)\n\t"
  203. : : "r" (p)
  204. );
  205. }
  206. static inline void prefetcht1(void *p)
  207. {
  208. asm volatile( "prefetcht1 (%0)\n\t"
  209. : : "r" (p)
  210. );
  211. }
  212. static inline void prefetcht2(void *p)
  213. {
  214. asm volatile( "prefetcht2 (%0)\n\t"
  215. : : "r" (p)
  216. );
  217. }
  218. #endif
  219. // The horizontal Functions exist only in C cuz the MMX code is faster with vertical filters and transposing
  220. /**
  221. * Check if the given 8x8 Block is mostly "flat"
  222. */
  223. static inline int isHorizDC(uint8_t src[], int stride)
  224. {
  225. int numEq= 0;
  226. int y;
  227. for(y=0; y<BLOCK_SIZE; y++)
  228. {
  229. if(((src[0] - src[1] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;
  230. if(((src[1] - src[2] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;
  231. if(((src[2] - src[3] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;
  232. if(((src[3] - src[4] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;
  233. if(((src[4] - src[5] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;
  234. if(((src[5] - src[6] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;
  235. if(((src[6] - src[7] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;
  236. src+= stride;
  237. }
  238. return numEq > hFlatnessThreshold;
  239. }
  240. static inline int isHorizMinMaxOk(uint8_t src[], int stride, int QP)
  241. {
  242. if(abs(src[0] - src[7]) > 2*QP) return 0;
  243. return 1;
  244. }
  245. static inline void doHorizDefFilter(uint8_t dst[], int stride, int QP)
  246. {
  247. int y;
  248. for(y=0; y<BLOCK_SIZE; y++)
  249. {
  250. const int middleEnergy= 5*(dst[4] - dst[5]) + 2*(dst[2] - dst[5]);
  251. if(ABS(middleEnergy) < 8*QP)
  252. {
  253. const int q=(dst[3] - dst[4])/2;
  254. const int leftEnergy= 5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
  255. const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
  256. int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
  257. d= MAX(d, 0);
  258. d= (5*d + 32) >> 6;
  259. d*= SIGN(-middleEnergy);
  260. if(q>0)
  261. {
  262. d= d<0 ? 0 : d;
  263. d= d>q ? q : d;
  264. }
  265. else
  266. {
  267. d= d>0 ? 0 : d;
  268. d= d<q ? q : d;
  269. }
  270. dst[3]-= d;
  271. dst[4]+= d;
  272. }
  273. dst+= stride;
  274. }
  275. }
  276. /**
  277. * Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block)
  278. * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version)
  279. */
  280. static inline void doHorizLowPass(uint8_t dst[], int stride, int QP)
  281. {
  282. int y;
  283. for(y=0; y<BLOCK_SIZE; y++)
  284. {
  285. const int first= ABS(dst[-1] - dst[0]) < QP ? dst[-1] : dst[0];
  286. const int last= ABS(dst[8] - dst[7]) < QP ? dst[8] : dst[7];
  287. int sums[9];
  288. sums[0] = first + dst[0];
  289. sums[1] = dst[0] + dst[1];
  290. sums[2] = dst[1] + dst[2];
  291. sums[3] = dst[2] + dst[3];
  292. sums[4] = dst[3] + dst[4];
  293. sums[5] = dst[4] + dst[5];
  294. sums[6] = dst[5] + dst[6];
  295. sums[7] = dst[6] + dst[7];
  296. sums[8] = dst[7] + last;
  297. dst[0]= ((sums[0]<<2) + ((first + sums[2])<<1) + sums[4] + 8)>>4;
  298. dst[1]= ((dst[1]<<2) + ((first + sums[0] + sums[3])<<1) + sums[5] + 8)>>4;
  299. dst[2]= ((dst[2]<<2) + ((first + sums[1] + sums[4])<<1) + sums[6] + 8)>>4;
  300. dst[3]= ((dst[3]<<2) + ((sums[2] + sums[5])<<1) + sums[0] + sums[7] + 8)>>4;
  301. dst[4]= ((dst[4]<<2) + ((sums[3] + sums[6])<<1) + sums[1] + sums[8] + 8)>>4;
  302. dst[5]= ((dst[5]<<2) + ((last + sums[7] + sums[4])<<1) + sums[2] + 8)>>4;
  303. dst[6]= (((last + dst[6])<<2) + ((dst[7] + sums[5])<<1) + sums[3] + 8)>>4;
  304. dst[7]= ((sums[8]<<2) + ((last + sums[6])<<1) + sums[4] + 8)>>4;
  305. dst+= stride;
  306. }
  307. }
  308. /**
  309. * Experimental Filter 1 (Horizontal)
  310. * will not damage linear gradients
  311. * Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
  312. * can only smooth blocks at the expected locations (it cant smooth them if they did move)
  313. * MMX2 version does correct clipping C version doesnt
  314. * not identical with the vertical one
  315. */
  316. static inline void horizX1Filter(uint8_t *src, int stride, int QP)
  317. {
  318. int y;
  319. static uint64_t *lut= NULL;
  320. if(lut==NULL)
  321. {
  322. int i;
  323. lut= (uint64_t*)memalign(8, 256*8);
  324. for(i=0; i<256; i++)
  325. {
  326. int v= i < 128 ? 2*i : 2*(i-256);
  327. /*
  328. //Simulate 112242211 9-Tap filter
  329. uint64_t a= (v/16) & 0xFF;
  330. uint64_t b= (v/8) & 0xFF;
  331. uint64_t c= (v/4) & 0xFF;
  332. uint64_t d= (3*v/8) & 0xFF;
  333. */
  334. //Simulate piecewise linear interpolation
  335. uint64_t a= (v/16) & 0xFF;
  336. uint64_t b= (v*3/16) & 0xFF;
  337. uint64_t c= (v*5/16) & 0xFF;
  338. uint64_t d= (7*v/16) & 0xFF;
  339. uint64_t A= (0x100 - a)&0xFF;
  340. uint64_t B= (0x100 - b)&0xFF;
  341. uint64_t C= (0x100 - c)&0xFF;
  342. uint64_t D= (0x100 - c)&0xFF;
  343. lut[i] = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
  344. (D<<24) | (C<<16) | (B<<8) | (A);
  345. //lut[i] = (v<<32) | (v<<24);
  346. }
  347. }
  348. for(y=0; y<BLOCK_SIZE; y++)
  349. {
  350. int a= src[1] - src[2];
  351. int b= src[3] - src[4];
  352. int c= src[5] - src[6];
  353. int d= MAX(ABS(b) - (ABS(a) + ABS(c))/2, 0);
  354. if(d < QP)
  355. {
  356. int v = d * SIGN(-b);
  357. src[1] +=v/8;
  358. src[2] +=v/4;
  359. src[3] +=3*v/8;
  360. src[4] -=3*v/8;
  361. src[5] -=v/4;
  362. src[6] -=v/8;
  363. }
  364. src+=stride;
  365. }
  366. }
  367. //Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
  368. //Plain C versions
  369. #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
  370. #define COMPILE_C
  371. #endif
  372. #ifdef CAN_COMPILE_X86_ASM
  373. #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
  374. #define COMPILE_MMX
  375. #endif
  376. #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
  377. #define COMPILE_MMX2
  378. #endif
  379. #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
  380. #define COMPILE_3DNOW
  381. #endif
  382. #endif //CAN_COMPILE_X86_ASM
  383. #undef HAVE_MMX
  384. #undef HAVE_MMX2
  385. #undef HAVE_3DNOW
  386. #undef ARCH_X86
  387. #ifdef COMPILE_C
  388. #undef HAVE_MMX
  389. #undef HAVE_MMX2
  390. #undef HAVE_3DNOW
  391. #undef ARCH_X86
  392. #define RENAME(a) a ## _C
  393. #include "postprocess_template.c"
  394. #endif
  395. //MMX versions
  396. #ifdef COMPILE_MMX
  397. #undef RENAME
  398. #define HAVE_MMX
  399. #undef HAVE_MMX2
  400. #undef HAVE_3DNOW
  401. #define ARCH_X86
  402. #define RENAME(a) a ## _MMX
  403. #include "postprocess_template.c"
  404. #endif
  405. //MMX2 versions
  406. #ifdef COMPILE_MMX2
  407. #undef RENAME
  408. #define HAVE_MMX
  409. #define HAVE_MMX2
  410. #undef HAVE_3DNOW
  411. #define ARCH_X86
  412. #define RENAME(a) a ## _MMX2
  413. #include "postprocess_template.c"
  414. #endif
  415. //3DNOW versions
  416. #ifdef COMPILE_3DNOW
  417. #undef RENAME
  418. #define HAVE_MMX
  419. #undef HAVE_MMX2
  420. #define HAVE_3DNOW
  421. #define ARCH_X86
  422. #define RENAME(a) a ## _3DNow
  423. #include "postprocess_template.c"
  424. #endif
  425. // minor note: the HAVE_xyz is messed up after that line so dont use it
  426. static inline void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
  427. QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode)
  428. {
  429. // useing ifs here as they are faster than function pointers allthough the
  430. // difference wouldnt be messureable here but its much better because
  431. // someone might exchange the cpu whithout restarting mplayer ;)
  432. #ifdef RUNTIME_CPUDETECT
  433. #ifdef CAN_COMPILE_X86_ASM
  434. // ordered per speed fasterst first
  435. if(gCpuCaps.hasMMX2)
  436. postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  437. else if(gCpuCaps.has3DNow)
  438. postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  439. else if(gCpuCaps.hasMMX)
  440. postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  441. else
  442. postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  443. #else
  444. postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  445. #endif
  446. #else //RUNTIME_CPUDETECT
  447. #ifdef HAVE_MMX2
  448. postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  449. #elif defined (HAVE_3DNOW)
  450. postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  451. #elif defined (HAVE_MMX)
  452. postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  453. #else
  454. postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);
  455. #endif
  456. #endif //!RUNTIME_CPUDETECT
  457. }
  458. #ifdef HAVE_ODIVX_POSTPROCESS
  459. #include "../opendivx/postprocess.h"
  460. int use_old_pp=0;
  461. #endif
  462. //static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
  463. // QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode);
  464. /* -pp Command line Help
  465. NOTE/FIXME: put this at an appropriate place (--help, html docs, man mplayer)?
  466. */
  467. char *help=
  468. "-npp <filterName>[:<option>[:<option>...]][,[-]<filterName>[:<option>...]]...\n"
  469. "long form example:\n"
  470. "-npp vdeblock:autoq,hdeblock:autoq,linblenddeint -npp default,-vdeblock\n"
  471. "short form example:\n"
  472. "-npp vb:a,hb:a,lb -npp de,-vb\n"
  473. "more examples:\n"
  474. "-npp tn:64:128:256\n"
  475. "Filters Options\n"
  476. "short long name short long option Description\n"
  477. "* * a autoq cpu power dependant enabler\n"
  478. " c chrom chrominance filtring enabled\n"
  479. " y nochrom chrominance filtring disabled\n"
  480. "hb hdeblock (2 Threshold) horizontal deblocking filter\n"
  481. " 1. Threshold: default=1, higher -> more deblocking\n"
  482. " 2. Threshold: default=40, lower -> more deblocking\n"
  483. " the h & v deblocking filters share these\n"
  484. " so u cant set different thresholds for h / v\n"
  485. "vb vdeblock (2 Threshold) vertical deblocking filter\n"
  486. "hr rkhdeblock\n"
  487. "vr rkvdeblock\n"
  488. "h1 x1hdeblock Experimental h deblock filter 1\n"
  489. "v1 x1vdeblock Experimental v deblock filter 1\n"
  490. "dr dering Deringing filter\n"
  491. "al autolevels automatic brightness / contrast\n"
  492. " f fullyrange stretch luminance to (0..255)\n"
  493. "lb linblenddeint linear blend deinterlacer\n"
  494. "li linipoldeint linear interpolating deinterlace\n"
  495. "ci cubicipoldeint cubic interpolating deinterlacer\n"
  496. "md mediandeint median deinterlacer\n"
  497. "de default hb:a,vb:a,dr:a,al\n"
  498. "fa fast h1:a,v1:a,dr:a,al\n"
  499. "tn tmpnoise (3 Thresholds) Temporal Noise Reducer\n"
  500. " 1. <= 2. <= 3. larger -> stronger filtering\n"
  501. "fq forceQuant <quantizer> Force quantizer\n"
  502. ;
  503. /**
  504. * returns a PPMode struct which will have a non 0 error variable if an error occured
  505. * name is the string after "-pp" on the command line
  506. * quality is a number from 0 to GET_PP_QUALITY_MAX
  507. */
  508. struct PPMode getPPModeByNameAndQuality(char *name, int quality)
  509. {
  510. char temp[GET_MODE_BUFFER_SIZE];
  511. char *p= temp;
  512. char *filterDelimiters= ",";
  513. char *optionDelimiters= ":";
  514. struct PPMode ppMode= {0,0,0,0,0,0,{150,200,400}};
  515. char *filterToken;
  516. strncpy(temp, name, GET_MODE_BUFFER_SIZE);
  517. if(verbose>1) printf("pp: %s\n", name);
  518. for(;;){
  519. char *filterName;
  520. int q= 1000000; //GET_PP_QUALITY_MAX;
  521. int chrom=-1;
  522. char *option;
  523. char *options[OPTIONS_ARRAY_SIZE];
  524. int i;
  525. int filterNameOk=0;
  526. int numOfUnknownOptions=0;
  527. int enable=1; //does the user want us to enabled or disabled the filter
  528. filterToken= strtok(p, filterDelimiters);
  529. if(filterToken == NULL) break;
  530. p+= strlen(filterToken) + 1; // p points to next filterToken
  531. filterName= strtok(filterToken, optionDelimiters);
  532. if(verbose>1) printf("pp: %s::%s\n", filterToken, filterName);
  533. if(*filterName == '-')
  534. {
  535. enable=0;
  536. filterName++;
  537. }
  538. for(;;){ //for all options
  539. option= strtok(NULL, optionDelimiters);
  540. if(option == NULL) break;
  541. if(verbose>1) printf("pp: option: %s\n", option);
  542. if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
  543. else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
  544. else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
  545. else
  546. {
  547. options[numOfUnknownOptions] = option;
  548. numOfUnknownOptions++;
  549. }
  550. if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
  551. }
  552. options[numOfUnknownOptions] = NULL;
  553. /* replace stuff from the replace Table */
  554. for(i=0; replaceTable[2*i]!=NULL; i++)
  555. {
  556. if(!strcmp(replaceTable[2*i], filterName))
  557. {
  558. int newlen= strlen(replaceTable[2*i + 1]);
  559. int plen;
  560. int spaceLeft;
  561. if(p==NULL) p= temp, *p=0; //last filter
  562. else p--, *p=','; //not last filter
  563. plen= strlen(p);
  564. spaceLeft= (int)p - (int)temp + plen;
  565. if(spaceLeft + newlen >= GET_MODE_BUFFER_SIZE)
  566. {
  567. ppMode.error++;
  568. break;
  569. }
  570. memmove(p + newlen, p, plen+1);
  571. memcpy(p, replaceTable[2*i + 1], newlen);
  572. filterNameOk=1;
  573. }
  574. }
  575. for(i=0; filters[i].shortName!=NULL; i++)
  576. {
  577. // printf("Compareing %s, %s, %s\n", filters[i].shortName,filters[i].longName, filterName);
  578. if( !strcmp(filters[i].longName, filterName)
  579. || !strcmp(filters[i].shortName, filterName))
  580. {
  581. ppMode.lumMode &= ~filters[i].mask;
  582. ppMode.chromMode &= ~filters[i].mask;
  583. filterNameOk=1;
  584. if(!enable) break; // user wants to disable it
  585. if(q >= filters[i].minLumQuality)
  586. ppMode.lumMode|= filters[i].mask;
  587. if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
  588. if(q >= filters[i].minChromQuality)
  589. ppMode.chromMode|= filters[i].mask;
  590. if(filters[i].mask == LEVEL_FIX)
  591. {
  592. int o;
  593. ppMode.minAllowedY= 16;
  594. ppMode.maxAllowedY= 234;
  595. for(o=0; options[o]!=NULL; o++)
  596. {
  597. if( !strcmp(options[o],"fullyrange")
  598. ||!strcmp(options[o],"f"))
  599. {
  600. ppMode.minAllowedY= 0;
  601. ppMode.maxAllowedY= 255;
  602. numOfUnknownOptions--;
  603. }
  604. }
  605. }
  606. else if(filters[i].mask == TEMP_NOISE_FILTER)
  607. {
  608. int o;
  609. int numOfNoises=0;
  610. ppMode.maxTmpNoise[0]= 150;
  611. ppMode.maxTmpNoise[1]= 200;
  612. ppMode.maxTmpNoise[2]= 400;
  613. for(o=0; options[o]!=NULL; o++)
  614. {
  615. char *tail;
  616. ppMode.maxTmpNoise[numOfNoises]=
  617. strtol(options[o], &tail, 0);
  618. if(tail!=options[o])
  619. {
  620. numOfNoises++;
  621. numOfUnknownOptions--;
  622. if(numOfNoises >= 3) break;
  623. }
  624. }
  625. }
  626. else if(filters[i].mask == V_DEBLOCK || filters[i].mask == H_DEBLOCK)
  627. {
  628. int o;
  629. ppMode.maxDcDiff=1;
  630. // hFlatnessThreshold= 40;
  631. // vFlatnessThreshold= 40;
  632. for(o=0; options[o]!=NULL && o<2; o++)
  633. {
  634. char *tail;
  635. int val= strtol(options[o], &tail, 0);
  636. if(tail==options[o]) break;
  637. numOfUnknownOptions--;
  638. if(o==0) ppMode.maxDcDiff= val;
  639. else hFlatnessThreshold=
  640. vFlatnessThreshold= val;
  641. }
  642. }
  643. else if(filters[i].mask == FORCE_QUANT)
  644. {
  645. int o;
  646. ppMode.forcedQuant= 15;
  647. for(o=0; options[o]!=NULL && o<1; o++)
  648. {
  649. char *tail;
  650. int val= strtol(options[o], &tail, 0);
  651. if(tail==options[o]) break;
  652. numOfUnknownOptions--;
  653. ppMode.forcedQuant= val;
  654. }
  655. }
  656. }
  657. }
  658. if(!filterNameOk) ppMode.error++;
  659. ppMode.error += numOfUnknownOptions;
  660. }
  661. #ifdef HAVE_ODIVX_POSTPROCESS
  662. if(ppMode.lumMode & H_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_Y_H;
  663. if(ppMode.lumMode & V_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_Y_V;
  664. if(ppMode.chromMode & H_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_C_H;
  665. if(ppMode.chromMode & V_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_C_V;
  666. if(ppMode.lumMode & DERING) ppMode.oldMode |= PP_DERING_Y;
  667. if(ppMode.chromMode & DERING) ppMode.oldMode |= PP_DERING_C;
  668. #endif
  669. if(verbose>1) printf("pp: lumMode=%X, chromMode=%X\n", ppMode.lumMode, ppMode.chromMode);
  670. return ppMode;
  671. }
  672. /**
  673. * Check and load the -npp part of the cmd line
  674. */
  675. int readPPOpt(void *conf, char *arg)
  676. {
  677. int quality;
  678. if(!strcmp("help", arg))
  679. {
  680. printf("%s", help);
  681. exit(1);
  682. }
  683. for(quality=0; quality<GET_PP_QUALITY_MAX+1; quality++)
  684. {
  685. gPPMode[quality]= getPPModeByNameAndQuality(arg, quality);
  686. if(gPPMode[quality].error) return -1;
  687. }
  688. newPPFlag=1;
  689. //divx_quality is passed to postprocess if autoq if off
  690. divx_quality= GET_PP_QUALITY_MAX;
  691. return 1;
  692. }
  693. /**
  694. * Obsolete, dont use it, use postprocess2() instead
  695. * this will check newPPFlag automatically and use postprocess2 if it is set
  696. * mode = quality if newPPFlag
  697. */
  698. void postprocess(unsigned char * src[], int src_stride,
  699. unsigned char * dst[], int dst_stride,
  700. int horizontal_size, int vertical_size,
  701. QP_STORE_T *QP_store, int QP_stride,
  702. int mode)
  703. {
  704. struct PPMode ppMode;
  705. static QP_STORE_T zeroArray[2048/8];
  706. static int firstTime=1;
  707. if(newPPFlag)
  708. {
  709. ppMode= gPPMode[mode];
  710. // printf("%d \n",QP_store[5]);
  711. postprocess2(src, src_stride, dst, dst_stride,
  712. horizontal_size, vertical_size, QP_store, QP_stride, &ppMode);
  713. return;
  714. }
  715. if(firstTime && verbose)
  716. {
  717. printf("using pp filters 0x%X\n", mode);
  718. firstTime=0;
  719. }
  720. if(QP_store==NULL)
  721. {
  722. QP_store= zeroArray;
  723. QP_stride= 0;
  724. }
  725. ppMode.lumMode= mode;
  726. mode= ((mode&0xFF)>>4) | (mode&0xFFFFFF00);
  727. ppMode.chromMode= mode;
  728. ppMode.maxTmpNoise[0]= 700;
  729. ppMode.maxTmpNoise[1]= 1500;
  730. ppMode.maxTmpNoise[2]= 3000;
  731. ppMode.maxAllowedY= 234;
  732. ppMode.minAllowedY= 16;
  733. ppMode.maxDcDiff= 1;
  734. #ifdef HAVE_ODIVX_POSTPROCESS
  735. // Note: I could make this shit outside of this file, but it would mean one
  736. // more function call...
  737. if(use_old_pp){
  738. odivx_postprocess(src,src_stride,dst,dst_stride,horizontal_size,vertical_size,QP_store,QP_stride,mode);
  739. return;
  740. }
  741. #endif
  742. postProcess(src[0], src_stride, dst[0], dst_stride,
  743. horizontal_size, vertical_size, QP_store, QP_stride, 0, &ppMode);
  744. horizontal_size >>= 1;
  745. vertical_size >>= 1;
  746. src_stride >>= 1;
  747. dst_stride >>= 1;
  748. if(ppMode.chromMode)
  749. {
  750. postProcess(src[1], src_stride, dst[1], dst_stride,
  751. horizontal_size, vertical_size, QP_store, QP_stride, 1, &ppMode);
  752. postProcess(src[2], src_stride, dst[2], dst_stride,
  753. horizontal_size, vertical_size, QP_store, QP_stride, 2, &ppMode);
  754. }
  755. else if(src_stride == dst_stride)
  756. {
  757. memcpy(dst[1], src[1], src_stride*vertical_size);
  758. memcpy(dst[2], src[2], src_stride*vertical_size);
  759. }
  760. else
  761. {
  762. int y;
  763. for(y=0; y<vertical_size; y++)
  764. {
  765. memcpy(&(dst[1][y*dst_stride]), &(src[1][y*src_stride]), horizontal_size);
  766. memcpy(&(dst[2][y*dst_stride]), &(src[2][y*src_stride]), horizontal_size);
  767. }
  768. }
  769. #if 0
  770. memset(dst[1], 128, dst_stride*vertical_size);
  771. memset(dst[2], 128, dst_stride*vertical_size);
  772. #endif
  773. }
  774. void postprocess2(unsigned char * src[], int src_stride,
  775. unsigned char * dst[], int dst_stride,
  776. int horizontal_size, int vertical_size,
  777. QP_STORE_T *QP_store, int QP_stride,
  778. struct PPMode *mode)
  779. {
  780. QP_STORE_T quantArray[2048/8];
  781. static int firstTime=1;
  782. if(QP_store==NULL || (mode->lumMode & FORCE_QUANT))
  783. {
  784. int i;
  785. QP_store= quantArray;
  786. QP_stride= 0;
  787. if(mode->lumMode & FORCE_QUANT)
  788. for(i=0; i<2048/8; i++) quantArray[i]= mode->forcedQuant;
  789. else
  790. for(i=0; i<2048/8; i++) quantArray[i]= 1;
  791. }
  792. if(firstTime && verbose)
  793. {
  794. printf("using npp filters 0x%X/0x%X\n", mode->lumMode, mode->chromMode);
  795. firstTime=0;
  796. }
  797. #ifdef HAVE_ODIVX_POSTPROCESS
  798. // Note: I could make this shit outside of this file, but it would mean one
  799. // more function call...
  800. if(use_old_pp){
  801. odivx_postprocess(src,src_stride,dst,dst_stride,horizontal_size,vertical_size,QP_store,QP_stride,
  802. mode->oldMode);
  803. return;
  804. }
  805. #endif
  806. postProcess(src[0], src_stride, dst[0], dst_stride,
  807. horizontal_size, vertical_size, QP_store, QP_stride, 0, mode);
  808. horizontal_size >>= 1;
  809. vertical_size >>= 1;
  810. src_stride >>= 1;
  811. dst_stride >>= 1;
  812. if(mode->chromMode)
  813. {
  814. postProcess(src[1], src_stride, dst[1], dst_stride,
  815. horizontal_size, vertical_size, QP_store, QP_stride, 1, mode);
  816. postProcess(src[2], src_stride, dst[2], dst_stride,
  817. horizontal_size, vertical_size, QP_store, QP_stride, 2, mode);
  818. }
  819. else if(src_stride == dst_stride)
  820. {
  821. memcpy(dst[1], src[1], src_stride*vertical_size);
  822. memcpy(dst[2], src[2], src_stride*vertical_size);
  823. }
  824. else
  825. {
  826. int y;
  827. for(y=0; y<vertical_size; y++)
  828. {
  829. memcpy(&(dst[1][y*dst_stride]), &(src[1][y*src_stride]), horizontal_size);
  830. memcpy(&(dst[2][y*dst_stride]), &(src[2][y*src_stride]), horizontal_size);
  831. }
  832. }
  833. }
  834. /**
  835. * gets the mode flags for a given quality (larger values mean slower but better postprocessing)
  836. * with -npp it simply returns quality
  837. * 0 <= quality <= 6
  838. */
  839. int getPpModeForQuality(int quality){
  840. int modes[1+GET_PP_QUALITY_MAX]= {
  841. 0,
  842. #if 1
  843. // horizontal filters first
  844. LUM_H_DEBLOCK,
  845. LUM_H_DEBLOCK | LUM_V_DEBLOCK,
  846. LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK,
  847. LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK,
  848. LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK | LUM_DERING,
  849. LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK | LUM_DERING | CHROM_DERING
  850. #else
  851. // vertical filters first
  852. LUM_V_DEBLOCK,
  853. LUM_V_DEBLOCK | LUM_H_DEBLOCK,
  854. LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK,
  855. LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK,
  856. LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK | LUM_DERING,
  857. LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK | LUM_DERING | CHROM_DERING
  858. #endif
  859. };
  860. #ifdef HAVE_ODIVX_POSTPROCESS
  861. int odivx_modes[1+GET_PP_QUALITY_MAX]= {
  862. 0,
  863. PP_DEBLOCK_Y_H,
  864. PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V,
  865. PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H,
  866. PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V,
  867. PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V|PP_DERING_Y,
  868. PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V|PP_DERING_Y|PP_DERING_C
  869. };
  870. if(use_old_pp) return odivx_modes[quality];
  871. #endif
  872. if(newPPFlag) return quality;
  873. else return modes[quality];
  874. }