falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
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.
347 Commits
32 Branches
866MB
Tree: 07f8991b60
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '07f8991b60'
${ noResults }
FFmpeg/postproc/postprocess.c

946 lines
28KB
Raw Blame History 

  1. /*

  2.     Copyright (C) 2001 Michael Niedermayer (michaelni@gmx.at)

  3. 

  4.     This program is free software; you can redistribute it and/or modify

  5.     it under the terms of the GNU General Public License as published by

  6.     the Free Software Foundation; either version 2 of the License, or

  7.     (at your option) any later version.

  8. 

  9.     This program is distributed in the hope that it will be useful,

  10.     but WITHOUT ANY WARRANTY; without even the implied warranty of

  11.     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

  12.     GNU General Public License for more details.

  13. 

  14.     You should have received a copy of the GNU General Public License

  15.     along with this program; if not, write to the Free Software

  16.     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  17. */

  18. 

  19. /*

  20. 			C	MMX	MMX2	3DNow

  21. isVertDC		Ec	Ec

  22. isVertMinMaxOk		Ec	Ec

  23. doVertLowPass		E		e	e

  24. doVertDefFilter		Ec	Ec	e	e

  25. isHorizDC		Ec	Ec

  26. isHorizMinMaxOk		a	E

  27. doHorizLowPass		E		e	e

  28. doHorizDefFilter	Ec	Ec	e	e

  29. deRing			E		e	e*

  30. Vertical RKAlgo1	E		a	a

  31. Horizontal RKAlgo1			a	a

  32. Vertical X1#		a		E	E

  33. Horizontal X1#		a		E	E

  34. LinIpolDeinterlace	e		E	E*

  35. CubicIpolDeinterlace	a		e	e*

  36. LinBlendDeinterlace	e		E	E*

  37. MedianDeinterlace#	 	Ec	Ec

  38. TempDeNoiser#		E		e	e

  39. 

  40. * i dont have a 3dnow CPU -> its untested, but noone said it doesnt work so it seems to work

  41. # more or less selfinvented filters so the exactness isnt too meaningfull

  42. E = Exact implementation

  43. e = allmost exact implementation (slightly different rounding,...)

  44. a = alternative / approximate impl

  45. c = checked against the other implementations (-vo md5)

  46. */

  47. 

  48. /*

  49. TODO:

  50. remove global/static vars

  51. reduce the time wasted on the mem transfer

  52. implement everything in C at least (done at the moment but ...)

  53. unroll stuff if instructions depend too much on the prior one

  54. we use 8x8 blocks for the horizontal filters, opendivx seems to use 8x4?

  55. move YScale thing to the end instead of fixing QP

  56. write a faster and higher quality deblocking filter :)

  57. make the mainloop more flexible (variable number of blocks at once

  58. 	(the if/else stuff per block is slowing things down)

  59. compare the quality & speed of all filters

  60. split this huge file

  61. border remover

  62. optimize c versions

  63. try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks

  64. smart blur

  65. commandline option for   the deblock / dering thresholds

  66. put fastmemcpy back

  67. dont use #ifdef ARCH_X86 for the asm stuff ... cross compilers? (note cpudetect uses ARCH_X86)

  68. ...

  69. */

  70. 

  71. //Changelog: use the CVS log

  72. 

  73. #include "../config.h"

  74. #include <inttypes.h>

  75. #include <stdio.h>

  76. #include <stdlib.h>

  77. #include <string.h>

  78. #ifdef HAVE_MALLOC_H

  79. #include <malloc.h>

  80. #endif

  81. //#undef HAVE_MMX2

  82. //#define HAVE_3DNOW

  83. //#undef HAVE_MMX

  84. //#undef ARCH_X86

  85. //#define DEBUG_BRIGHTNESS

  86. //#include "../libvo/fastmemcpy.h"

  87. #include "postprocess.h"

  88. #include "../cpudetect.h"

  89. 

  90. #define MIN(a,b) ((a) > (b) ? (b) : (a))

  91. #define MAX(a,b) ((a) < (b) ? (b) : (a))

  92. #define ABS(a) ((a) > 0 ? (a) : (-(a)))

  93. #define SIGN(a) ((a) > 0 ? 1 : -1)

  94. 

  95. #define GET_MODE_BUFFER_SIZE 500

  96. #define OPTIONS_ARRAY_SIZE 10

  97. 

  98. #ifdef ARCH_X86

  99. #define CAN_COMPILE_X86_ASM

  100. #endif

  101. 

  102. #ifdef CAN_COMPILE_X86_ASM

  103. static volatile uint64_t __attribute__((aligned(8))) packedYOffset=	0x0000000000000000LL;

  104. static volatile uint64_t __attribute__((aligned(8))) packedYScale=	0x0100010001000100LL;

  105. static uint64_t __attribute__((aligned(8))) w05=		0x0005000500050005LL;

  106. static uint64_t __attribute__((aligned(8))) w20=		0x0020002000200020LL;

  107. static uint64_t __attribute__((aligned(8))) w1400=		0x1400140014001400LL;

  108. static uint64_t __attribute__((aligned(8))) bm00000001=		0x00000000000000FFLL;

  109. static uint64_t __attribute__((aligned(8))) bm00010000=		0x000000FF00000000LL;

  110. static uint64_t __attribute__((aligned(8))) bm00001000=		0x00000000FF000000LL;

  111. static uint64_t __attribute__((aligned(8))) bm10000000=		0xFF00000000000000LL;

  112. static uint64_t __attribute__((aligned(8))) bm10000001=		0xFF000000000000FFLL;

  113. static uint64_t __attribute__((aligned(8))) bm11000011=		0xFFFF00000000FFFFLL;

  114. static uint64_t __attribute__((aligned(8))) bm00000011=		0x000000000000FFFFLL;

  115. static uint64_t __attribute__((aligned(8))) bm11111110=		0xFFFFFFFFFFFFFF00LL;

  116. static uint64_t __attribute__((aligned(8))) bm11000000=		0xFFFF000000000000LL;

  117. static uint64_t __attribute__((aligned(8))) bm00011000=		0x000000FFFF000000LL;

  118. static uint64_t __attribute__((aligned(8))) bm00110011=		0x0000FFFF0000FFFFLL;

  119. static uint64_t __attribute__((aligned(8))) bm11001100=		0xFFFF0000FFFF0000LL;

  120. static uint64_t __attribute__((aligned(8))) b00= 		0x0000000000000000LL;

  121. static uint64_t __attribute__((aligned(8))) b01= 		0x0101010101010101LL;

  122. static uint64_t __attribute__((aligned(8))) b02= 		0x0202020202020202LL;

  123. static uint64_t __attribute__((aligned(8))) b0F= 		0x0F0F0F0F0F0F0F0FLL;

  124. static uint64_t __attribute__((aligned(8))) b04= 		0x0404040404040404LL;

  125. static uint64_t __attribute__((aligned(8))) b08= 		0x0808080808080808LL;

  126. static uint64_t __attribute__((aligned(8))) bFF= 		0xFFFFFFFFFFFFFFFFLL;

  127. static uint64_t __attribute__((aligned(8))) b20= 		0x2020202020202020LL;

  128. static uint64_t __attribute__((aligned(8))) b80= 		0x8080808080808080LL;

  129. static uint64_t __attribute__((aligned(8))) mmxDCOffset= 	0x7E7E7E7E7E7E7E7ELL;

  130. static uint64_t __attribute__((aligned(8))) mmxDCThreshold=	0x7C7C7C7C7C7C7C7CLL;

  131. static uint64_t __attribute__((aligned(8))) b3F= 		0x3F3F3F3F3F3F3F3FLL;

  132. static uint64_t __attribute__((aligned(8))) temp0=0;

  133. static uint64_t __attribute__((aligned(8))) temp1=0;

  134. static uint64_t __attribute__((aligned(8))) temp2=0;

  135. static uint64_t __attribute__((aligned(8))) temp3=0;

  136. static uint64_t __attribute__((aligned(8))) temp4=0;

  137. static uint64_t __attribute__((aligned(8))) temp5=0;

  138. static uint64_t __attribute__((aligned(8))) pQPb=0;

  139. static uint64_t __attribute__((aligned(8))) pQPb2=0;

  140. static uint8_t __attribute__((aligned(8))) tempBlocks[8*16*2]; //used for the horizontal code

  141. static uint32_t __attribute__((aligned(4))) maxTmpNoise[4];

  142. #else

  143. static uint64_t packedYOffset=	0x0000000000000000LL;

  144. static uint64_t packedYScale=	0x0100010001000100LL;

  145. #endif

  146. 

  147. extern int divx_quality;

  148. int newPPFlag=0; //is set if -npp is used

  149. struct PPMode gPPMode[GET_PP_QUALITY_MAX+1];

  150. 

  151. extern int verbose;

  152. 

  153. int hFlatnessThreshold= 56 - 16;

  154. int vFlatnessThreshold= 56 - 16;

  155. int deringThreshold= 20;

  156. 

  157. static int dcOffset;

  158. static int dcThreshold;

  159. 

  160. //amount of "black" u r willing to loose to get a brightness corrected picture

  161. double maxClippedThreshold= 0.01;

  162. 

  163. static struct PPFilter filters[]=

  164. {

  165. 	{"hb", "hdeblock", 		1, 1, 3, H_DEBLOCK},

  166. 	{"vb", "vdeblock", 		1, 2, 4, V_DEBLOCK},

  167. 	{"hr", "rkhdeblock", 		1, 1, 3, H_RK1_FILTER},

  168. 	{"vr", "rkvdeblock", 		1, 2, 4, V_RK1_FILTER},

  169. 	{"h1", "x1hdeblock", 		1, 1, 3, H_X1_FILTER},

  170. 	{"v1", "x1vdeblock", 		1, 2, 4, V_X1_FILTER},

  171. 	{"dr", "dering", 		1, 5, 6, DERING},

  172. 	{"al", "autolevels", 		0, 1, 2, LEVEL_FIX},

  173. 	{"lb", "linblenddeint", 	1, 1, 4, LINEAR_BLEND_DEINT_FILTER},

  174. 	{"li", "linipoldeint", 		1, 1, 4, LINEAR_IPOL_DEINT_FILTER},

  175. 	{"ci", "cubicipoldeint",	1, 1, 4, CUBIC_IPOL_DEINT_FILTER},

  176. 	{"md", "mediandeint", 		1, 1, 4, MEDIAN_DEINT_FILTER},

  177. 	{"tn", "tmpnoise", 		1, 7, 8, TEMP_NOISE_FILTER},

  178. 	{NULL, NULL,0,0,0,0} //End Marker

  179. };

  180. 

  181. static char *replaceTable[]=

  182. {

  183. 	"default", 	"hdeblock:a,vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",

  184. 	"de", 		"hdeblock:a,vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",

  185. 	"fast", 	"x1hdeblock:a,x1vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",

  186. 	"fa", 		"x1hdeblock:a,x1vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",

  187. 	NULL //End Marker

  188. };

  189. 

  190. #ifdef CAN_COMPILE_X86_ASM

  191. static inline void unusedVariableWarningFixer()

  192. {

  193. if(

  194.  packedYOffset + packedYScale + w05 + w20 + w1400 + bm00000001 + bm00010000

  195.  + bm00001000 + bm10000000 + bm10000001 + bm11000011 + bm00000011 + bm11111110

  196.  + bm11000000 + bm00011000 + bm00110011 + bm11001100 + b00 + b01 + b02 + b0F

  197.  + bFF + b20 + b04+ b08 + pQPb2 + b80 + mmxDCOffset + mmxDCThreshold + b3F + temp0 + temp1 + temp2 + temp3 + temp4

  198.  + temp5 + pQPb== 0) b00=0;

  199. }

  200. #endif

  201. 

  202. #ifdef TIMING

  203. static inline long long rdtsc()

  204. {

  205. 	long long l;

  206. 	asm volatile(	"rdtsc\n\t"

  207. 		: "=A" (l)

  208. 	);

  209. //	printf("%d\n", int(l/1000));

  210. 	return l;

  211. }

  212. #endif

  213. 

  214. #ifdef CAN_COMPILE_X86_ASM

  215. static inline void prefetchnta(void *p)

  216. {

  217. 	asm volatile(	"prefetchnta (%0)\n\t"

  218. 		: : "r" (p)

  219. 	);

  220. }

  221. 

  222. static inline void prefetcht0(void *p)

  223. {

  224. 	asm volatile(	"prefetcht0 (%0)\n\t"

  225. 		: : "r" (p)

  226. 	);

  227. }

  228. 

  229. static inline void prefetcht1(void *p)

  230. {

  231. 	asm volatile(	"prefetcht1 (%0)\n\t"

  232. 		: : "r" (p)

  233. 	);

  234. }

  235. 

  236. static inline void prefetcht2(void *p)

  237. {

  238. 	asm volatile(	"prefetcht2 (%0)\n\t"

  239. 		: : "r" (p)

  240. 	);

  241. }

  242. #endif

  243. 

  244. // The horizontal Functions exist only in C cuz the MMX code is faster with vertical filters and transposing

  245. 

  246. /**

  247.  * Check if the given 8x8 Block is mostly "flat"

  248.  */

  249. static inline int isHorizDC(uint8_t src[], int stride)

  250. {

  251. 	int numEq= 0;

  252. 	int y;

  253. 	for(y=0; y<BLOCK_SIZE; y++)

  254. 	{

  255. 		if(((src[0] - src[1] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;

  256. 		if(((src[1] - src[2] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;

  257. 		if(((src[2] - src[3] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;

  258. 		if(((src[3] - src[4] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;

  259. 		if(((src[4] - src[5] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;

  260. 		if(((src[5] - src[6] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;

  261. 		if(((src[6] - src[7] + dcOffset) & 0xFFFF) < dcThreshold) numEq++;

  262. 		src+= stride;

  263. 	}

  264. 	return numEq > hFlatnessThreshold;

  265. }

  266. 

  267. static inline int isHorizMinMaxOk(uint8_t src[], int stride, int QP)

  268. {

  269. 	if(abs(src[0] - src[7]) > 2*QP) return 0;

  270. 

  271. 	return 1;

  272. }

  273. 

  274. static inline void doHorizDefFilter(uint8_t dst[], int stride, int QP)

  275. {

  276. 	int y;

  277. 	for(y=0; y<BLOCK_SIZE; y++)

  278. 	{

  279. 		const int middleEnergy= 5*(dst[4] - dst[5]) + 2*(dst[2] - dst[5]);

  280. 

  281. 		if(ABS(middleEnergy) < 8*QP)

  282. 		{

  283. 			const int q=(dst[3] - dst[4])/2;

  284. 			const int leftEnergy=  5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);

  285. 			const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);

  286. 

  287. 			int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );

  288. 			d= MAX(d, 0);

  289. 

  290. 			d= (5*d + 32) >> 6;

  291. 			d*= SIGN(-middleEnergy);

  292. 

  293. 			if(q>0)

  294. 			{

  295. 				d= d<0 ? 0 : d;

  296. 				d= d>q ? q : d;

  297. 			}

  298. 			else

  299. 			{

  300. 				d= d>0 ? 0 : d;

  301. 				d= d<q ? q : d;

  302. 			}

  303. 

  304.         		dst[3]-= d;

  305. 	        	dst[4]+= d;

  306. 		}

  307. 		dst+= stride;

  308. 	}

  309. }

  310. 

  311. /**

  312.  * Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block)

  313.  * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version)

  314.  */

  315. static inline void doHorizLowPass(uint8_t dst[], int stride, int QP)

  316. {

  317. 

  318. 	int y;

  319. 	for(y=0; y<BLOCK_SIZE; y++)

  320. 	{

  321. 		const int first= ABS(dst[-1] - dst[0]) < QP ? dst[-1] : dst[0];

  322. 		const int last= ABS(dst[8] - dst[7]) < QP ? dst[8] : dst[7];

  323. 

  324. 		int sums[9];

  325. 		sums[0] = first + dst[0];

  326. 		sums[1] = dst[0] + dst[1];

  327. 		sums[2] = dst[1] + dst[2];

  328. 		sums[3] = dst[2] + dst[3];

  329. 		sums[4] = dst[3] + dst[4];

  330. 		sums[5] = dst[4] + dst[5];

  331. 		sums[6] = dst[5] + dst[6];

  332. 		sums[7] = dst[6] + dst[7];

  333. 		sums[8] = dst[7] + last;

  334. 

  335. 		dst[0]= ((sums[0]<<2) + ((first + sums[2])<<1) + sums[4] + 8)>>4;

  336. 		dst[1]= ((dst[1]<<2) + ((first + sums[0] + sums[3])<<1) + sums[5] + 8)>>4;

  337. 		dst[2]= ((dst[2]<<2) + ((first + sums[1] + sums[4])<<1) + sums[6] + 8)>>4;

  338. 		dst[3]= ((dst[3]<<2) + ((sums[2] + sums[5])<<1) + sums[0] + sums[7] + 8)>>4;

  339. 		dst[4]= ((dst[4]<<2) + ((sums[3] + sums[6])<<1) + sums[1] + sums[8] + 8)>>4;

  340. 		dst[5]= ((dst[5]<<2) + ((last + sums[7] + sums[4])<<1) + sums[2] + 8)>>4;

  341. 		dst[6]= (((last + dst[6])<<2) + ((dst[7] + sums[5])<<1) + sums[3] + 8)>>4;

  342. 		dst[7]= ((sums[8]<<2) + ((last + sums[6])<<1) + sums[4] + 8)>>4;

  343. 

  344. 		dst+= stride;

  345. 	}

  346. }

  347. 

  348. /**

  349.  * Experimental Filter 1 (Horizontal)

  350.  * will not damage linear gradients

  351.  * Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter

  352.  * can only smooth blocks at the expected locations (it cant smooth them if they did move)

  353.  * MMX2 version does correct clipping C version doesnt

  354.  * not identical with the vertical one

  355.  */

  356. static inline void horizX1Filter(uint8_t *src, int stride, int QP)

  357. {

  358. 	int y;

  359. 	static uint64_t *lut= NULL;

  360. 	if(lut==NULL)

  361. 	{

  362. 		int i;

  363. 		lut= (uint64_t*)memalign(8, 256*8);

  364. 		for(i=0; i<256; i++)

  365. 		{

  366. 			int v= i < 128 ? 2*i : 2*(i-256);

  367. /*

  368. //Simulate 112242211 9-Tap filter

  369. 			uint64_t a= (v/16) & 0xFF;

  370. 			uint64_t b= (v/8) & 0xFF;

  371. 			uint64_t c= (v/4) & 0xFF;

  372. 			uint64_t d= (3*v/8) & 0xFF;

  373. */

  374. //Simulate piecewise linear interpolation

  375. 			uint64_t a= (v/16) & 0xFF;

  376. 			uint64_t b= (v*3/16) & 0xFF;

  377. 			uint64_t c= (v*5/16) & 0xFF;

  378. 			uint64_t d= (7*v/16) & 0xFF;

  379. 			uint64_t A= (0x100 - a)&0xFF;

  380. 			uint64_t B= (0x100 - b)&0xFF;

  381. 			uint64_t C= (0x100 - c)&0xFF;

  382. 			uint64_t D= (0x100 - c)&0xFF;

  383. 

  384. 			lut[i]   = (a<<56) | (b<<48) | (c<<40) | (d<<32) |

  385. 				(D<<24) | (C<<16) | (B<<8) | (A);

  386. 			//lut[i] = (v<<32) | (v<<24);

  387. 		}

  388. 	}

  389. 

  390. 	for(y=0; y<BLOCK_SIZE; y++)

  391. 	{

  392. 		int a= src[1] - src[2];

  393. 		int b= src[3] - src[4];

  394. 		int c= src[5] - src[6];

  395. 

  396. 		int d= MAX(ABS(b) - (ABS(a) + ABS(c))/2, 0);

  397. 

  398. 		if(d < QP)

  399. 		{

  400. 			int v = d * SIGN(-b);

  401. 

  402. 			src[1] +=v/8;

  403. 			src[2] +=v/4;

  404. 			src[3] +=3*v/8;

  405. 			src[4] -=3*v/8;

  406. 			src[5] -=v/4;

  407. 			src[6] -=v/8;

  408. 

  409. 		}

  410. 		src+=stride;

  411. 	}

  412. }

  413. 

  414. 

  415. //Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one

  416. //Plain C versions

  417. #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)

  418. #define COMPILE_C

  419. #endif

  420. 

  421. #ifdef CAN_COMPILE_X86_ASM

  422. 

  423. #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)

  424. #define COMPILE_MMX

  425. #endif

  426. 

  427. #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)

  428. #define COMPILE_MMX2

  429. #endif

  430. 

  431. #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)

  432. #define COMPILE_3DNOW

  433. #endif

  434. #endif //CAN_COMPILE_X86_ASM

  435. 

  436. #undef HAVE_MMX

  437. #undef HAVE_MMX2

  438. #undef HAVE_3DNOW

  439. #undef ARCH_X86

  440. 

  441. #ifdef COMPILE_C

  442. #undef HAVE_MMX

  443. #undef HAVE_MMX2

  444. #undef HAVE_3DNOW

  445. #undef ARCH_X86

  446. #define RENAME(a) a ## _C

  447. #include "postprocess_template.c"

  448. #endif

  449. 

  450. //MMX versions

  451. #ifdef COMPILE_MMX

  452. #undef RENAME

  453. #define HAVE_MMX

  454. #undef HAVE_MMX2

  455. #undef HAVE_3DNOW

  456. #define ARCH_X86

  457. #define RENAME(a) a ## _MMX

  458. #include "postprocess_template.c"

  459. #endif

  460. 

  461. //MMX2 versions

  462. #ifdef COMPILE_MMX2

  463. #undef RENAME

  464. #define HAVE_MMX

  465. #define HAVE_MMX2

  466. #undef HAVE_3DNOW

  467. #define ARCH_X86

  468. #define RENAME(a) a ## _MMX2

  469. #include "postprocess_template.c"

  470. #endif

  471. 

  472. //3DNOW versions

  473. #ifdef COMPILE_3DNOW

  474. #undef RENAME

  475. #define HAVE_MMX

  476. #undef HAVE_MMX2

  477. #define HAVE_3DNOW

  478. #define ARCH_X86

  479. #define RENAME(a) a ## _3DNow

  480. #include "postprocess_template.c"

  481. #endif

  482. 

  483. // minor note: the HAVE_xyz is messed up after that line so dont use it

  484. 

  485. static inline void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  486. 	QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode)

  487. {

  488. 	// useing ifs here as they are faster than function pointers allthough the

  489. 	// difference wouldnt be messureable here but its much better because

  490. 	// someone might exchange the cpu whithout restarting mplayer ;)

  491. #ifdef RUNTIME_CPUDETECT

  492. #ifdef CAN_COMPILE_X86_ASM

  493. 	// ordered per speed fasterst first

  494. 	if(gCpuCaps.hasMMX2)

  495. 		postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  496. 	else if(gCpuCaps.has3DNow)

  497. 		postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  498. 	else if(gCpuCaps.hasMMX)

  499. 		postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  500. 	else

  501. 		postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  502. #else

  503. 		postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  504. #endif

  505. #else //RUNTIME_CPUDETECT

  506. #ifdef HAVE_MMX2

  507. 		postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  508. #elif defined (HAVE_3DNOW)

  509. 		postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  510. #elif defined (HAVE_MMX)

  511. 		postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  512. #else

  513. 		postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, ppMode);

  514. #endif

  515. #endif //!RUNTIME_CPUDETECT

  516. }

  517. 

  518. #ifdef HAVE_ODIVX_POSTPROCESS

  519. #include "../opendivx/postprocess.h"

  520. int use_old_pp=0;

  521. #endif

  522. 

  523. //static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  524. //	QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode);

  525. 

  526. /* -pp Command line Help

  527. NOTE/FIXME: put this at an appropriate place (--help, html docs, man mplayer)?

  528. 

  529. -pp <filterName>[:<option>[:<option>...]][,[-]<filterName>[:<option>...]]...

  530. 

  531. long form example:

  532. -pp vdeblock:autoq,hdeblock:autoq,linblenddeint		-pp default,-vdeblock

  533. short form example:

  534. -pp vb:a,hb:a,lb					-pp de,-vb

  535. more examples:

  536. -pp tn:64:128:256

  537. 

  538. Filters			Options

  539. short	long name	short	long option	Description

  540. *	*		a	autoq		cpu power dependant enabler

  541. 			c	chrom		chrominance filtring enabled

  542. 			y	nochrom		chrominance filtring disabled

  543. hb	hdeblock	(2 Threshold)		horizontal deblocking filter

  544.                          1. Threshold: default is 1, higher means stronger deblocking

  545.                          2. Threshold: default is 40, lower means stronger deblocking

  546. 			the horizontal & vertical deblocking filters share these 

  547. 			so u cant set different thresholds for h / v

  548. vb	vdeblock	(2 Threshold)		vertical deblocking filter

  549. hr	rkhdeblock

  550. vr	rkvdeblock

  551. h1	x1hdeblock				Experimental horizontal deblock filter 1

  552. v1	x1vdeblock				Experimental vertical deblock filter 1

  553. dr	dering					not implemented yet

  554. al	autolevels				automatic brightness / contrast fixer

  555. 			f	fullyrange	stretch luminance range to (0..255)

  556. lb	linblenddeint				linear blend deinterlacer

  557. li	linipoldeint				linear interpolating deinterlacer

  558. ci	cubicipoldeint				cubic interpolating deinterlacer

  559. md	mediandeint				median deinterlacer

  560. de	default					hdeblock:a,vdeblock:a,dering:a,autolevels

  561. fa	fast					x1hdeblock:a,x1vdeblock:a,dering:a,autolevels

  562. tn	tmpnoise	(3 Thresholds)		Temporal Noise Reducer

  563. 			1. <= 2. <= 3. Threshold, larger means stronger filtering

  564. */

  565. 

  566. /**

  567.  * returns a PPMode struct which will have a non 0 error variable if an error occured

  568.  * name is the string after "-pp" on the command line

  569.  * quality is a number from 0 to GET_PP_QUALITY_MAX

  570.  */

  571. struct PPMode getPPModeByNameAndQuality(char *name, int quality)

  572. {

  573. 	char temp[GET_MODE_BUFFER_SIZE];

  574. 	char *p= temp;

  575. 	char *filterDelimiters= ",";

  576. 	char *optionDelimiters= ":";

  577. 	struct PPMode ppMode= {0,0,0,0,0,0,{150,200,400}};

  578. 	char *filterToken;

  579. 

  580. 	strncpy(temp, name, GET_MODE_BUFFER_SIZE);

  581. 

  582. 	if(verbose) printf("%s\n", name);

  583. 

  584. 	for(;;){

  585. 		char *filterName;

  586. 		int q= 1000000; //GET_PP_QUALITY_MAX;

  587. 		int chrom=-1;

  588. 		char *option;

  589. 		char *options[OPTIONS_ARRAY_SIZE];

  590. 		int i;

  591. 		int filterNameOk=0;

  592. 		int numOfUnknownOptions=0;

  593. 		int enable=1; //does the user want us to enabled or disabled the filter

  594. 

  595. 		filterToken= strtok(p, filterDelimiters);

  596. 		if(filterToken == NULL) break;

  597. 		p+= strlen(filterToken) + 1; // p points to next filterToken

  598. 		filterName= strtok(filterToken, optionDelimiters);

  599. 		if(verbose) printf("%s::%s\n", filterToken, filterName);

  600. 

  601. 		if(*filterName == '-')

  602. 		{

  603. 			enable=0;

  604. 			filterName++;

  605. 		}

  606. 

  607. 		for(;;){ //for all options

  608. 			option= strtok(NULL, optionDelimiters);

  609. 			if(option == NULL) break;

  610. 

  611. 			if(verbose) printf(" option: %s\n", option);

  612. 			if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;

  613. 			else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;

  614. 			else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;

  615. 			else

  616. 			{

  617. 				options[numOfUnknownOptions] = option;

  618. 				numOfUnknownOptions++;

  619. 			}

  620. 			if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;

  621. 		}

  622. 		options[numOfUnknownOptions] = NULL;

  623. 

  624. 		/* replace stuff from the replace Table */

  625. 		for(i=0; replaceTable[2*i]!=NULL; i++)

  626. 		{

  627. 			if(!strcmp(replaceTable[2*i], filterName))

  628. 			{

  629. 				int newlen= strlen(replaceTable[2*i + 1]);

  630. 				int plen;

  631. 				int spaceLeft;

  632. 

  633. 				if(p==NULL) p= temp, *p=0; 	//last filter

  634. 				else p--, *p=',';		//not last filter

  635. 

  636. 				plen= strlen(p);

  637. 				spaceLeft= (int)p - (int)temp + plen;

  638. 				if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE)

  639. 				{

  640. 					ppMode.error++;

  641. 					break;

  642. 				}

  643. 				memmove(p + newlen, p, plen+1);

  644. 				memcpy(p, replaceTable[2*i + 1], newlen);

  645. 				filterNameOk=1;

  646. 			}

  647. 		}

  648. 

  649. 		for(i=0; filters[i].shortName!=NULL; i++)

  650. 		{

  651. //			printf("Compareing %s, %s, %s\n", filters[i].shortName,filters[i].longName, filterName);

  652. 			if(   !strcmp(filters[i].longName, filterName)

  653. 			   || !strcmp(filters[i].shortName, filterName))

  654. 			{

  655. 				ppMode.lumMode &= ~filters[i].mask;

  656. 				ppMode.chromMode &= ~filters[i].mask;

  657. 

  658. 				filterNameOk=1;

  659. 				if(!enable) break; // user wants to disable it

  660. 

  661. 				if(q >= filters[i].minLumQuality)

  662. 					ppMode.lumMode|= filters[i].mask;

  663. 				if(chrom==1 || (chrom==-1 && filters[i].chromDefault))

  664. 					if(q >= filters[i].minChromQuality)

  665. 						ppMode.chromMode|= filters[i].mask;

  666. 

  667. 				if(filters[i].mask == LEVEL_FIX)

  668. 				{

  669. 					int o;

  670. 					ppMode.minAllowedY= 16;

  671. 					ppMode.maxAllowedY= 234;

  672. 					for(o=0; options[o]!=NULL; o++)

  673. 					{

  674. 						if(  !strcmp(options[o],"fullyrange")

  675. 						   ||!strcmp(options[o],"f"))

  676. 						{

  677. 							ppMode.minAllowedY= 0;

  678. 							ppMode.maxAllowedY= 255;

  679. 							numOfUnknownOptions--;

  680. 						}

  681. 					}

  682. 				}

  683. 				else if(filters[i].mask == TEMP_NOISE_FILTER)

  684. 				{

  685. 					int o;

  686. 					int numOfNoises=0;

  687. 					ppMode.maxTmpNoise[0]= 150;

  688. 					ppMode.maxTmpNoise[1]= 200;

  689. 					ppMode.maxTmpNoise[2]= 400;

  690. 

  691. 					for(o=0; options[o]!=NULL; o++)

  692. 					{

  693. 						char *tail;

  694. 						ppMode.maxTmpNoise[numOfNoises]=

  695. 							strtol(options[o], &tail, 0);

  696. 						if(tail!=options[o])

  697. 						{

  698. 							numOfNoises++;

  699. 							numOfUnknownOptions--;

  700. 							if(numOfNoises >= 3) break;

  701. 						}

  702. 					}

  703. 				}

  704. 				else if(filters[i].mask == V_DEBLOCK || filters[i].mask == H_DEBLOCK)

  705. 				{

  706. 					int o;

  707. 					ppMode.maxDcDiff=1;

  708. //					hFlatnessThreshold= 40;

  709. //					vFlatnessThreshold= 40;

  710. 

  711. 					for(o=0; options[o]!=NULL && o<2; o++)

  712. 					{

  713. 						char *tail;

  714. 						int val= strtol(options[o], &tail, 0);

  715. 						if(tail==options[o]) break;

  716. 

  717. 						numOfUnknownOptions--;

  718. 						if(o==0) ppMode.maxDcDiff= val;

  719. 						else hFlatnessThreshold=

  720. 						     vFlatnessThreshold= val;

  721. 					}

  722. 				}

  723. 			}

  724. 		}

  725. 		if(!filterNameOk) ppMode.error++;

  726. 		ppMode.error += numOfUnknownOptions;

  727. 	}

  728. 

  729. #ifdef HAVE_ODIVX_POSTPROCESS

  730. 	if(ppMode.lumMode & H_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_Y_H;

  731. 	if(ppMode.lumMode & V_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_Y_V;

  732. 	if(ppMode.chromMode & H_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_C_H;

  733. 	if(ppMode.chromMode & V_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_C_V;

  734. 	if(ppMode.lumMode & DERING) ppMode.oldMode |= PP_DERING_Y;

  735. 	if(ppMode.chromMode & DERING) ppMode.oldMode |= PP_DERING_C;

  736. #endif

  737. 

  738. 	if(verbose) printf("lumMode=%X, chromMode=%X\n", ppMode.lumMode, ppMode.chromMode);

  739. 	return ppMode;

  740. }

  741. 

  742. /**

  743.  * Check and load the -npp part of the cmd line

  744.  */

  745. int readPPOpt(void *conf, char *arg)

  746. {

  747. 	int quality;

  748. 	for(quality=0; quality<GET_PP_QUALITY_MAX+1; quality++)

  749. 	{

  750. 		gPPMode[quality]= getPPModeByNameAndQuality(arg, quality);

  751. 

  752. 		if(gPPMode[quality].error) return -1;

  753. 	}

  754. 	newPPFlag=1;

  755. 

  756. //divx_quality is passed to postprocess if autoq if off

  757. 	divx_quality= GET_PP_QUALITY_MAX; 

  758. 	return 1;

  759. }

  760. 

  761. /**

  762.  * Obsolete, dont use it, use postprocess2() instead

  763.  * this will check newPPFlag automatically and use postprocess2 if it is set

  764.  * mode = quality if newPPFlag

  765.  */

  766. void  postprocess(unsigned char * src[], int src_stride,

  767.                  unsigned char * dst[], int dst_stride,

  768.                  int horizontal_size,   int vertical_size,

  769.                  QP_STORE_T *QP_store,  int QP_stride,

  770. 					  int mode)

  771. {

  772. 	struct PPMode ppMode;

  773. 	static QP_STORE_T zeroArray[2048/8];

  774. 

  775. 	if(newPPFlag)

  776. 	{

  777. 		ppMode= gPPMode[mode];

  778. 		

  779. 		postprocess2(src, src_stride, dst, dst_stride,

  780. 			horizontal_size, vertical_size, QP_store, QP_stride, &ppMode);

  781. 

  782. 		return;

  783. 	}

  784. 	

  785. 	if(QP_store==NULL)

  786. 	{

  787. 		QP_store= zeroArray;

  788. 		QP_stride= 0;

  789. 	}

  790. 

  791. 	ppMode.lumMode= mode;

  792. 	mode= ((mode&0xFF)>>4) | (mode&0xFFFFFF00);

  793. 	ppMode.chromMode= mode;

  794. 	ppMode.maxTmpNoise[0]= 700;

  795. 	ppMode.maxTmpNoise[1]= 1500;

  796. 	ppMode.maxTmpNoise[2]= 3000;

  797. 	ppMode.maxAllowedY= 234;

  798. 	ppMode.minAllowedY= 16;

  799. 	ppMode.maxDcDiff= 1;

  800. 

  801. #ifdef HAVE_ODIVX_POSTPROCESS

  802. // Note: I could make this shit outside of this file, but it would mean one

  803. // more function call...

  804. 	if(use_old_pp){

  805. 	    odivx_postprocess(src,src_stride,dst,dst_stride,horizontal_size,vertical_size,QP_store,QP_stride,mode);

  806. 	    return;

  807. 	}

  808. #endif

  809. 

  810. 	postProcess(src[0], src_stride, dst[0], dst_stride,

  811. 		horizontal_size, vertical_size, QP_store, QP_stride, 0, &ppMode);

  812. 

  813. 	horizontal_size >>= 1;

  814. 	vertical_size   >>= 1;

  815. 	src_stride      >>= 1;

  816. 	dst_stride      >>= 1;

  817. 

  818. 	if(ppMode.chromMode)

  819. 	{

  820. 		postProcess(src[1], src_stride, dst[1], dst_stride,

  821. 			horizontal_size, vertical_size, QP_store, QP_stride, 1, &ppMode);

  822. 		postProcess(src[2], src_stride, dst[2], dst_stride,

  823. 			horizontal_size, vertical_size, QP_store, QP_stride, 2, &ppMode);

  824. 	}

  825. 	else if(src_stride == dst_stride)

  826. 	{

  827. 		memcpy(dst[1], src[1], src_stride*vertical_size);

  828. 		memcpy(dst[2], src[2], src_stride*vertical_size);

  829. 	}

  830. 	else

  831. 	{

  832. 		int y;

  833. 		for(y=0; y<vertical_size; y++)

  834. 		{

  835. 			memcpy(&(dst[1][y*dst_stride]), &(src[1][y*src_stride]), horizontal_size);

  836. 			memcpy(&(dst[2][y*dst_stride]), &(src[2][y*src_stride]), horizontal_size);

  837. 		}

  838. 	}

  839. 

  840. #if 0

  841. 		memset(dst[1], 128, dst_stride*vertical_size);

  842. 		memset(dst[2], 128, dst_stride*vertical_size);

  843. #endif

  844. }

  845. 

  846. void  postprocess2(unsigned char * src[], int src_stride,

  847.                  unsigned char * dst[], int dst_stride,

  848.                  int horizontal_size,   int vertical_size,

  849.                  QP_STORE_T *QP_store,  int QP_stride,

  850. 		 struct PPMode *mode)

  851. {

  852. 

  853. 	static QP_STORE_T zeroArray[2048/8];

  854. 	if(QP_store==NULL)

  855. 	{

  856. 		QP_store= zeroArray;

  857. 		QP_stride= 0;

  858. 	}

  859. 

  860. #ifdef HAVE_ODIVX_POSTPROCESS

  861. // Note: I could make this shit outside of this file, but it would mean one

  862. // more function call...

  863. 	if(use_old_pp){

  864. 	    odivx_postprocess(src,src_stride,dst,dst_stride,horizontal_size,vertical_size,QP_store,QP_stride,

  865. 	    mode->oldMode);

  866. 	    return;

  867. 	}

  868. #endif

  869. 

  870. 	postProcess(src[0], src_stride, dst[0], dst_stride,

  871. 		horizontal_size, vertical_size, QP_store, QP_stride, 0, mode);

  872. 

  873. 	horizontal_size >>= 1;

  874. 	vertical_size   >>= 1;

  875. 	src_stride      >>= 1;

  876. 	dst_stride      >>= 1;

  877. 

  878. 	if(mode->chromMode)

  879. 	{

  880. 		postProcess(src[1], src_stride, dst[1], dst_stride,

  881. 			horizontal_size, vertical_size, QP_store, QP_stride, 1, mode);

  882. 		postProcess(src[2], src_stride, dst[2], dst_stride,

  883. 			horizontal_size, vertical_size, QP_store, QP_stride, 2, mode);

  884. 	}

  885. 	else if(src_stride == dst_stride)

  886. 	{

  887. 		memcpy(dst[1], src[1], src_stride*vertical_size);

  888. 		memcpy(dst[2], src[2], src_stride*vertical_size);

  889. 	}

  890. 	else

  891. 	{

  892. 		int y;

  893. 		for(y=0; y<vertical_size; y++)

  894. 		{

  895. 			memcpy(&(dst[1][y*dst_stride]), &(src[1][y*src_stride]), horizontal_size);

  896. 			memcpy(&(dst[2][y*dst_stride]), &(src[2][y*src_stride]), horizontal_size);

  897. 		}

  898. 	}

  899. }

  900. 

  901. 

  902. /**

  903.  * gets the mode flags for a given quality (larger values mean slower but better postprocessing)

  904.  * with -npp it simply returns quality 

  905.  * 0 <= quality <= 6

  906.  */

  907. int getPpModeForQuality(int quality){

  908. 	int modes[1+GET_PP_QUALITY_MAX]= {

  909. 		0,

  910. #if 1

  911. 		// horizontal filters first

  912. 		LUM_H_DEBLOCK,

  913. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK,

  914. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK,

  915. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK,

  916. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK | LUM_DERING,

  917. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK | LUM_DERING | CHROM_DERING

  918. #else

  919. 		// vertical filters first

  920. 		LUM_V_DEBLOCK,

  921. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK,

  922. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK,

  923. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK,

  924. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK | LUM_DERING,

  925. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK | LUM_DERING | CHROM_DERING

  926. #endif

  927. 	};

  928. 

  929. #ifdef HAVE_ODIVX_POSTPROCESS

  930. 	int odivx_modes[1+GET_PP_QUALITY_MAX]= {

  931. 		0,

  932. 		PP_DEBLOCK_Y_H,

  933. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V,

  934. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H,

  935. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V,

  936. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V|PP_DERING_Y,

  937. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V|PP_DERING_Y|PP_DERING_C

  938. 	};

  939. 	if(use_old_pp) return odivx_modes[quality];

  940. #endif

  941. 	if(newPPFlag)	return quality;

  942. 	else		return modes[quality];

  943. }

  944.