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.
376 Commits
32 Branches
5.3GB
Tree: 9b4644281d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '9b4644281d'
${ noResults }
FFmpeg/postproc/postprocess.c

990 lines
29KB
Raw Blame History 

  1. /*

  2.     Copyright (C) 2001-2002 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. #include "../mangle.h"

  90. 

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

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

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

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

  95. 

  96. #define GET_MODE_BUFFER_SIZE 500

  97. #define OPTIONS_ARRAY_SIZE 10

  98. 

  99. #ifdef ARCH_X86

  100. #define CAN_COMPILE_X86_ASM

  101. #endif

  102. 

  103. #ifdef CAN_COMPILE_X86_ASM

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  143. #else

  144. static uint64_t packedYOffset=	0x0000000000000000LL;

  145. static uint64_t packedYScale=	0x0100010001000100LL;

  146. #endif

  147. 

  148. extern int divx_quality;

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

  150. struct PPMode gPPMode[GET_PP_QUALITY_MAX+1];

  151. 

  152. extern int verbose;

  153. 

  154. int hFlatnessThreshold= 56 - 16;

  155. int vFlatnessThreshold= 56 - 16;

  156. int deringThreshold= 20;

  157. 

  158. static int dcOffset;

  159. static int dcThreshold;

  160. 

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

  162. double maxClippedThreshold= 0.01;

  163. 

  164. static struct PPFilter filters[]=

  165. {

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

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

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

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

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

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

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

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

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

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

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

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

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

  179. 	{"fq", "forcequant", 		1, 0, 0, FORCE_QUANT},

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

  181. };

  182. 

  183. static char *replaceTable[]=

  184. {

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

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

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

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

  189. 	NULL //End Marker

  190. };

  191. 

  192. #ifdef CAN_COMPILE_X86_ASM

  193. static inline void unusedVariableWarningFixer()

  194. {

  195. if(

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

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

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

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

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

  201. }

  202. #endif

  203. 

  204. #ifdef TIMING

  205. static inline long long rdtsc()

  206. {

  207. 	long long l;

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

  209. 		: "=A" (l)

  210. 	);

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

  212. 	return l;

  213. }

  214. #endif

  215. 

  216. #ifdef CAN_COMPILE_X86_ASM

  217. static inline void prefetchnta(void *p)

  218. {

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

  220. 		: : "r" (p)

  221. 	);

  222. }

  223. 

  224. static inline void prefetcht0(void *p)

  225. {

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

  227. 		: : "r" (p)

  228. 	);

  229. }

  230. 

  231. static inline void prefetcht1(void *p)

  232. {

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

  234. 		: : "r" (p)

  235. 	);

  236. }

  237. 

  238. static inline void prefetcht2(void *p)

  239. {

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

  241. 		: : "r" (p)

  242. 	);

  243. }

  244. #endif

  245. 

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

  247. 

  248. /**

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

  250.  */

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

  252. {

  253. 	int numEq= 0;

  254. 	int y;

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

  256. 	{

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

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

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

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

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

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

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

  264. 		src+= stride;

  265. 	}

  266. 	return numEq > hFlatnessThreshold;

  267. }

  268. 

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

  270. {

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

  272. 

  273. 	return 1;

  274. }

  275. 

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

  277. {

  278. 	int y;

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

  280. 	{

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

  282. 

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

  284. 		{

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

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

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

  288. 

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

  290. 			d= MAX(d, 0);

  291. 

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

  293. 			d*= SIGN(-middleEnergy);

  294. 

  295. 			if(q>0)

  296. 			{

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

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

  299. 			}

  300. 			else

  301. 			{

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

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

  304. 			}

  305. 

  306.         		dst[3]-= d;

  307. 	        	dst[4]+= d;

  308. 		}

  309. 		dst+= stride;

  310. 	}

  311. }

  312. 

  313. /**

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

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

  316.  */

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

  318. {

  319. 

  320. 	int y;

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

  322. 	{

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

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

  325. 

  326. 		int sums[9];

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

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

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

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

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

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

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

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

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

  336. 

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

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

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

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

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

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

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

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

  345. 

  346. 		dst+= stride;

  347. 	}

  348. }

  349. 

  350. /**

  351.  * Experimental Filter 1 (Horizontal)

  352.  * will not damage linear gradients

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

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

  355.  * MMX2 version does correct clipping C version doesnt

  356.  * not identical with the vertical one

  357.  */

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

  359. {

  360. 	int y;

  361. 	static uint64_t *lut= NULL;

  362. 	if(lut==NULL)

  363. 	{

  364. 		int i;

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

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

  367. 		{

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

  369. /*

  370. //Simulate 112242211 9-Tap filter

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

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

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

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

  375. */

  376. //Simulate piecewise linear interpolation

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

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

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

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

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

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

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

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

  385. 

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

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

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

  389. 		}

  390. 	}

  391. 

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

  393. 	{

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

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

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

  397. 

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

  399. 

  400. 		if(d < QP)

  401. 		{

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

  403. 

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

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

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

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

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

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

  410. 

  411. 		}

  412. 		src+=stride;

  413. 	}

  414. }

  415. 

  416. 

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

  418. //Plain C versions

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

  420. #define COMPILE_C

  421. #endif

  422. 

  423. #ifdef CAN_COMPILE_X86_ASM

  424. 

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

  426. #define COMPILE_MMX

  427. #endif

  428. 

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

  430. #define COMPILE_MMX2

  431. #endif

  432. 

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

  434. #define COMPILE_3DNOW

  435. #endif

  436. #endif //CAN_COMPILE_X86_ASM

  437. 

  438. #undef HAVE_MMX

  439. #undef HAVE_MMX2

  440. #undef HAVE_3DNOW

  441. #undef ARCH_X86

  442. 

  443. #ifdef COMPILE_C

  444. #undef HAVE_MMX

  445. #undef HAVE_MMX2

  446. #undef HAVE_3DNOW

  447. #undef ARCH_X86

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

  449. #include "postprocess_template.c"

  450. #endif

  451. 

  452. //MMX versions

  453. #ifdef COMPILE_MMX

  454. #undef RENAME

  455. #define HAVE_MMX

  456. #undef HAVE_MMX2

  457. #undef HAVE_3DNOW

  458. #define ARCH_X86

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

  460. #include "postprocess_template.c"

  461. #endif

  462. 

  463. //MMX2 versions

  464. #ifdef COMPILE_MMX2

  465. #undef RENAME

  466. #define HAVE_MMX

  467. #define HAVE_MMX2

  468. #undef HAVE_3DNOW

  469. #define ARCH_X86

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

  471. #include "postprocess_template.c"

  472. #endif

  473. 

  474. //3DNOW versions

  475. #ifdef COMPILE_3DNOW

  476. #undef RENAME

  477. #define HAVE_MMX

  478. #undef HAVE_MMX2

  479. #define HAVE_3DNOW

  480. #define ARCH_X86

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

  482. #include "postprocess_template.c"

  483. #endif

  484. 

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

  486. 

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

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

  489. {

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

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

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

  493. #ifdef RUNTIME_CPUDETECT

  494. #ifdef CAN_COMPILE_X86_ASM

  495. 	// ordered per speed fasterst first

  496. 	if(gCpuCaps.hasMMX2)

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

  498. 	else if(gCpuCaps.has3DNow)

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

  500. 	else if(gCpuCaps.hasMMX)

  501. 		postProcess_MMX(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. #else

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

  506. #endif

  507. #else //RUNTIME_CPUDETECT

  508. #ifdef HAVE_MMX2

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

  510. #elif defined (HAVE_3DNOW)

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

  512. #elif defined (HAVE_MMX)

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

  514. #else

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

  516. #endif

  517. #endif //!RUNTIME_CPUDETECT

  518. }

  519. 

  520. #ifdef HAVE_ODIVX_POSTPROCESS

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

  522. int use_old_pp=0;

  523. #endif

  524. 

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

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

  527. 

  528. /* -pp Command line Help

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

  530. */

  531. char *help=

  532. "-npp <filterName>[:<option>[:<option>...]][,[-]<filterName>[:<option>...]]...\n"

  533. "long form example:\n"

  534. "-npp vdeblock:autoq,hdeblock:autoq,linblenddeint	-npp default,-vdeblock\n"

  535. "short form example:\n"

  536. "-npp vb:a,hb:a,lb					-npp de,-vb\n"

  537. "more examples:\n"

  538. "-npp tn:64:128:256\n"

  539. "Filters			Options\n"

  540. "short	long name	short	long option	Description\n"

  541. "*	*		a	autoq		cpu power dependant enabler\n"

  542. "			c	chrom		chrominance filtring enabled\n"

  543. "			y	nochrom		chrominance filtring disabled\n"

  544. "hb	hdeblock	(2 Threshold)		horizontal deblocking filter\n"

  545. "                        1. Threshold: default=1, higher -> more deblocking\n"

  546. "                        2. Threshold: default=40, lower -> more deblocking\n"

  547. "			the h & v deblocking filters share these\n"

  548. "			so u cant set different thresholds for h / v\n"

  549. "vb	vdeblock	(2 Threshold)		vertical deblocking filter\n"

  550. "hr	rkhdeblock\n"

  551. "vr	rkvdeblock\n"

  552. "h1	x1hdeblock				Experimental h deblock filter 1\n"

  553. "v1	x1vdeblock				Experimental v deblock filter 1\n"

  554. "dr	dering					Deringing filter\n"

  555. "al	autolevels				automatic brightness / contrast\n"

  556. "			f	fullyrange	stretch luminance to (0..255)\n"

  557. "lb	linblenddeint				linear blend deinterlacer\n"

  558. "li	linipoldeint				linear interpolating deinterlace\n"

  559. "ci	cubicipoldeint				cubic interpolating deinterlacer\n"

  560. "md	mediandeint				median deinterlacer\n"

  561. "de	default					hb:a,vb:a,dr:a,al\n"

  562. "fa	fast					h1:a,v1:a,dr:a,al\n"

  563. "tn	tmpnoise	(3 Thresholds)		Temporal Noise Reducer\n"

  564. "			1. <= 2. <= 3.		larger -> stronger filtering\n"

  565. "fq	forceQuant	<quantizer>		Force quantizer\n"

  566. ;

  567. 

  568. /**

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

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

  571.  * quality is a number from 0 to GET_PP_QUALITY_MAX

  572.  */

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

  574. {

  575. 	char temp[GET_MODE_BUFFER_SIZE];

  576. 	char *p= temp;

  577. 	char *filterDelimiters= ",";

  578. 	char *optionDelimiters= ":";

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

  580. 	char *filterToken;

  581. 

  582. 	strncpy(temp, name, GET_MODE_BUFFER_SIZE);

  583. 

  584. 	if(verbose>1) printf("pp: %s\n", name);

  585. 

  586. 	for(;;){

  587. 		char *filterName;

  588. 		int q= 1000000; //GET_PP_QUALITY_MAX;

  589. 		int chrom=-1;

  590. 		char *option;

  591. 		char *options[OPTIONS_ARRAY_SIZE];

  592. 		int i;

  593. 		int filterNameOk=0;

  594. 		int numOfUnknownOptions=0;

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

  596. 

  597. 		filterToken= strtok(p, filterDelimiters);

  598. 		if(filterToken == NULL) break;

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

  600. 		filterName= strtok(filterToken, optionDelimiters);

  601. 		if(verbose>1) printf("pp: %s::%s\n", filterToken, filterName);

  602. 

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

  604. 		{

  605. 			enable=0;

  606. 			filterName++;

  607. 		}

  608. 

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

  610. 			option= strtok(NULL, optionDelimiters);

  611. 			if(option == NULL) break;

  612. 

  613. 			if(verbose>1) printf("pp: option: %s\n", option);

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

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

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

  617. 			else

  618. 			{

  619. 				options[numOfUnknownOptions] = option;

  620. 				numOfUnknownOptions++;

  621. 			}

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

  623. 		}

  624. 		options[numOfUnknownOptions] = NULL;

  625. 

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

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

  628. 		{

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

  630. 			{

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

  632. 				int plen;

  633. 				int spaceLeft;

  634. 

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

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

  637. 

  638. 				plen= strlen(p);

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

  640. 				if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE)

  641. 				{

  642. 					ppMode.error++;

  643. 					break;

  644. 				}

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

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

  647. 				filterNameOk=1;

  648. 			}

  649. 		}

  650. 

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

  652. 		{

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

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

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

  656. 			{

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

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

  659. 

  660. 				filterNameOk=1;

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

  662. 

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

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

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

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

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

  668. 

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

  670. 				{

  671. 					int o;

  672. 					ppMode.minAllowedY= 16;

  673. 					ppMode.maxAllowedY= 234;

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

  675. 					{

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

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

  678. 						{

  679. 							ppMode.minAllowedY= 0;

  680. 							ppMode.maxAllowedY= 255;

  681. 							numOfUnknownOptions--;

  682. 						}

  683. 					}

  684. 				}

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

  686. 				{

  687. 					int o;

  688. 					int numOfNoises=0;

  689. 					ppMode.maxTmpNoise[0]= 150;

  690. 					ppMode.maxTmpNoise[1]= 200;

  691. 					ppMode.maxTmpNoise[2]= 400;

  692. 

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

  694. 					{

  695. 						char *tail;

  696. 						ppMode.maxTmpNoise[numOfNoises]=

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

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

  699. 						{

  700. 							numOfNoises++;

  701. 							numOfUnknownOptions--;

  702. 							if(numOfNoises >= 3) break;

  703. 						}

  704. 					}

  705. 				}

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

  707. 				{

  708. 					int o;

  709. 					ppMode.maxDcDiff=1;

  710. //					hFlatnessThreshold= 40;

  711. //					vFlatnessThreshold= 40;

  712. 

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

  714. 					{

  715. 						char *tail;

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

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

  718. 

  719. 						numOfUnknownOptions--;

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

  721. 						else hFlatnessThreshold=

  722. 						     vFlatnessThreshold= val;

  723. 					}

  724. 				}

  725. 				else if(filters[i].mask == FORCE_QUANT)

  726. 				{

  727. 					int o;

  728. 					ppMode.forcedQuant= 15;

  729. 

  730. 					for(o=0; options[o]!=NULL && o<1; o++)

  731. 					{

  732. 						char *tail;

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

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

  735. 

  736. 						numOfUnknownOptions--;

  737. 						ppMode.forcedQuant= val;

  738. 					}

  739. 				}

  740. 			}

  741. 		}

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

  743. 		ppMode.error += numOfUnknownOptions;

  744. 	}

  745. 

  746. #ifdef HAVE_ODIVX_POSTPROCESS

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

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

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

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

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

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

  753. #endif

  754. 

  755. 	if(verbose>1) printf("pp: lumMode=%X, chromMode=%X\n", ppMode.lumMode, ppMode.chromMode);

  756. 	return ppMode;

  757. }

  758. 

  759. /**

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

  761.  */

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

  763. {

  764. 	int quality;

  765. 	

  766. 	if(!strcmp("help", arg))

  767. 	{

  768. 		printf("%s", help);

  769. 		exit(1);

  770. 	}

  771. 	

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

  773. 	{

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

  775. 

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

  777. 	}

  778. 	newPPFlag=1;

  779. 

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

  781. 	divx_quality= GET_PP_QUALITY_MAX; 

  782. 	return 1;

  783. }

  784. 

  785. /**

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

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

  788.  * mode = quality if newPPFlag

  789.  */

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

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

  792.                  int horizontal_size,   int vertical_size,

  793.                  QP_STORE_T *QP_store,  int QP_stride,

  794. 					  int mode)

  795. {

  796. 	struct PPMode ppMode;

  797. 	static QP_STORE_T zeroArray[2048/8];

  798. 	static int firstTime=1;

  799. 

  800. 	if(newPPFlag)

  801. 	{

  802. 		ppMode= gPPMode[mode];

  803. //		printf("%d \n",QP_store[5]);

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

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

  806. 

  807. 		return;

  808. 	}

  809. 

  810. 	if(firstTime && verbose)

  811. 	{

  812. 		printf("using pp filters 0x%X\n", mode);

  813. 		firstTime=0;

  814. 	}

  815. 

  816. 	if(QP_store==NULL)

  817. 	{

  818. 		QP_store= zeroArray;

  819. 		QP_stride= 0;

  820. 	}

  821. 

  822. 	ppMode.lumMode= mode;

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

  824. 	ppMode.chromMode= mode;

  825. 	ppMode.maxTmpNoise[0]= 700;

  826. 	ppMode.maxTmpNoise[1]= 1500;

  827. 	ppMode.maxTmpNoise[2]= 3000;

  828. 	ppMode.maxAllowedY= 234;

  829. 	ppMode.minAllowedY= 16;

  830. 	ppMode.maxDcDiff= 1;

  831. 

  832. #ifdef HAVE_ODIVX_POSTPROCESS

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

  834. // more function call...

  835. 	if(use_old_pp){

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

  837. 	    return;

  838. 	}

  839. #endif

  840. 

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

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

  843. 

  844. 	horizontal_size >>= 1;

  845. 	vertical_size   >>= 1;

  846. 	src_stride      >>= 1;

  847. 	dst_stride      >>= 1;

  848. 

  849. 	if(ppMode.chromMode)

  850. 	{

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

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

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

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

  855. 	}

  856. 	else if(src_stride == dst_stride)

  857. 	{

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

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

  860. 	}

  861. 	else

  862. 	{

  863. 		int y;

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

  865. 		{

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

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

  868. 		}

  869. 	}

  870. 

  871. #if 0

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

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

  874. #endif

  875. }

  876. 

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

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

  879.                  int horizontal_size,   int vertical_size,

  880.                  QP_STORE_T *QP_store,  int QP_stride,

  881. 		 struct PPMode *mode)

  882. {

  883. 

  884. 	QP_STORE_T quantArray[2048/8];

  885. 	static int firstTime=1;

  886. 	

  887. 	if(QP_store==NULL || (mode->lumMode & FORCE_QUANT)) 

  888. 	{

  889. 		int i;

  890. 		QP_store= quantArray;

  891. 		QP_stride= 0;

  892. 		if(mode->lumMode & FORCE_QUANT)

  893. 			for(i=0; i<2048/8; i++) quantArray[i]= mode->forcedQuant;

  894. 		else

  895. 			for(i=0; i<2048/8; i++) quantArray[i]= 1;

  896. 	}

  897. 

  898. 	if(firstTime && verbose)

  899. 	{

  900. 		printf("using npp filters 0x%X/0x%X\n", mode->lumMode, mode->chromMode);

  901. 		firstTime=0;

  902. 	}

  903. 

  904. #ifdef HAVE_ODIVX_POSTPROCESS

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

  906. // more function call...

  907. 	if(use_old_pp){

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

  909. 	    mode->oldMode);

  910. 	    return;

  911. 	}

  912. #endif

  913. 

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

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

  916. 

  917. 	horizontal_size >>= 1;

  918. 	vertical_size   >>= 1;

  919. 	src_stride      >>= 1;

  920. 	dst_stride      >>= 1;

  921. 

  922. 	if(mode->chromMode)

  923. 	{

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

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

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

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

  928. 	}

  929. 	else if(src_stride == dst_stride)

  930. 	{

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

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

  933. 	}

  934. 	else

  935. 	{

  936. 		int y;

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

  938. 		{

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

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

  941. 		}

  942. 	}

  943. }

  944. 

  945. 

  946. /**

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

  948.  * with -npp it simply returns quality 

  949.  * 0 <= quality <= 6

  950.  */

  951. int getPpModeForQuality(int quality){

  952. 	int modes[1+GET_PP_QUALITY_MAX]= {

  953. 		0,

  954. #if 1

  955. 		// horizontal filters first

  956. 		LUM_H_DEBLOCK,

  957. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK,

  958. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK,

  959. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK,

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

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

  962. #else

  963. 		// vertical filters first

  964. 		LUM_V_DEBLOCK,

  965. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK,

  966. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK,

  967. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK,

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

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

  970. #endif

  971. 	};

  972. 

  973. #ifdef HAVE_ODIVX_POSTPROCESS

  974. 	int odivx_modes[1+GET_PP_QUALITY_MAX]= {

  975. 		0,

  976. 		PP_DEBLOCK_Y_H,

  977. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V,

  978. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H,

  979. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V,

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

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

  982. 	};

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

  984. #endif

  985. 	if(newPPFlag)	return quality;

  986. 	else		return modes[quality];

  987. }

  988.