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.
349 Commits
32 Branches
837MB
Tree: 8aaac43576
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '8aaac43576'
${ noResults }
FFmpeg/postproc/postprocess.c

968 lines
28KB
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. 

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

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

  180. };

  181. 

  182. static char *replaceTable[]=

  183. {

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

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

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

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

  188. 	NULL //End Marker

  189. };

  190. 

  191. #ifdef CAN_COMPILE_X86_ASM

  192. static inline void unusedVariableWarningFixer()

  193. {

  194. if(

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

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

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

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

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

  200. }

  201. #endif

  202. 

  203. #ifdef TIMING

  204. static inline long long rdtsc()

  205. {

  206. 	long long l;

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

  208. 		: "=A" (l)

  209. 	);

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

  211. 	return l;

  212. }

  213. #endif

  214. 

  215. #ifdef CAN_COMPILE_X86_ASM

  216. static inline void prefetchnta(void *p)

  217. {

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

  219. 		: : "r" (p)

  220. 	);

  221. }

  222. 

  223. static inline void prefetcht0(void *p)

  224. {

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

  226. 		: : "r" (p)

  227. 	);

  228. }

  229. 

  230. static inline void prefetcht1(void *p)

  231. {

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

  233. 		: : "r" (p)

  234. 	);

  235. }

  236. 

  237. static inline void prefetcht2(void *p)

  238. {

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

  240. 		: : "r" (p)

  241. 	);

  242. }

  243. #endif

  244. 

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

  246. 

  247. /**

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

  249.  */

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

  251. {

  252. 	int numEq= 0;

  253. 	int y;

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

  255. 	{

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

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

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

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

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

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

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

  263. 		src+= stride;

  264. 	}

  265. 	return numEq > hFlatnessThreshold;

  266. }

  267. 

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

  269. {

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

  271. 

  272. 	return 1;

  273. }

  274. 

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

  276. {

  277. 	int y;

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

  279. 	{

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

  281. 

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

  283. 		{

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

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

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

  287. 

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

  289. 			d= MAX(d, 0);

  290. 

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

  292. 			d*= SIGN(-middleEnergy);

  293. 

  294. 			if(q>0)

  295. 			{

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

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

  298. 			}

  299. 			else

  300. 			{

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

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

  303. 			}

  304. 

  305.         		dst[3]-= d;

  306. 	        	dst[4]+= d;

  307. 		}

  308. 		dst+= stride;

  309. 	}

  310. }

  311. 

  312. /**

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

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

  315.  */

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

  317. {

  318. 

  319. 	int y;

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

  321. 	{

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

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

  324. 

  325. 		int sums[9];

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

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

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

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

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

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

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

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

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

  335. 

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

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

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

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

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

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

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

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

  344. 

  345. 		dst+= stride;

  346. 	}

  347. }

  348. 

  349. /**

  350.  * Experimental Filter 1 (Horizontal)

  351.  * will not damage linear gradients

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

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

  354.  * MMX2 version does correct clipping C version doesnt

  355.  * not identical with the vertical one

  356.  */

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

  358. {

  359. 	int y;

  360. 	static uint64_t *lut= NULL;

  361. 	if(lut==NULL)

  362. 	{

  363. 		int i;

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

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

  366. 		{

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

  368. /*

  369. //Simulate 112242211 9-Tap filter

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

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

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

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

  374. */

  375. //Simulate piecewise linear interpolation

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

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

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

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

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

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

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

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

  384. 

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

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

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

  388. 		}

  389. 	}

  390. 

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

  392. 	{

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

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

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

  396. 

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

  398. 

  399. 		if(d < QP)

  400. 		{

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

  402. 

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

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

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

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

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

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

  409. 

  410. 		}

  411. 		src+=stride;

  412. 	}

  413. }

  414. 

  415. 

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

  417. //Plain C versions

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

  419. #define COMPILE_C

  420. #endif

  421. 

  422. #ifdef CAN_COMPILE_X86_ASM

  423. 

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

  425. #define COMPILE_MMX

  426. #endif

  427. 

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

  429. #define COMPILE_MMX2

  430. #endif

  431. 

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

  433. #define COMPILE_3DNOW

  434. #endif

  435. #endif //CAN_COMPILE_X86_ASM

  436. 

  437. #undef HAVE_MMX

  438. #undef HAVE_MMX2

  439. #undef HAVE_3DNOW

  440. #undef ARCH_X86

  441. 

  442. #ifdef COMPILE_C

  443. #undef HAVE_MMX

  444. #undef HAVE_MMX2

  445. #undef HAVE_3DNOW

  446. #undef ARCH_X86

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

  448. #include "postprocess_template.c"

  449. #endif

  450. 

  451. //MMX versions

  452. #ifdef COMPILE_MMX

  453. #undef RENAME

  454. #define HAVE_MMX

  455. #undef HAVE_MMX2

  456. #undef HAVE_3DNOW

  457. #define ARCH_X86

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

  459. #include "postprocess_template.c"

  460. #endif

  461. 

  462. //MMX2 versions

  463. #ifdef COMPILE_MMX2

  464. #undef RENAME

  465. #define HAVE_MMX

  466. #define HAVE_MMX2

  467. #undef HAVE_3DNOW

  468. #define ARCH_X86

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

  470. #include "postprocess_template.c"

  471. #endif

  472. 

  473. //3DNOW versions

  474. #ifdef COMPILE_3DNOW

  475. #undef RENAME

  476. #define HAVE_MMX

  477. #undef HAVE_MMX2

  478. #define HAVE_3DNOW

  479. #define ARCH_X86

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

  481. #include "postprocess_template.c"

  482. #endif

  483. 

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

  485. 

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

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

  488. {

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

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

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

  492. #ifdef RUNTIME_CPUDETECT

  493. #ifdef CAN_COMPILE_X86_ASM

  494. 	// ordered per speed fasterst first

  495. 	if(gCpuCaps.hasMMX2)

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

  497. 	else if(gCpuCaps.has3DNow)

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

  499. 	else if(gCpuCaps.hasMMX)

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

  501. 	else

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

  503. #else

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

  505. #endif

  506. #else //RUNTIME_CPUDETECT

  507. #ifdef HAVE_MMX2

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

  509. #elif defined (HAVE_3DNOW)

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

  511. #elif defined (HAVE_MMX)

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

  513. #else

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

  515. #endif

  516. #endif //!RUNTIME_CPUDETECT

  517. }

  518. 

  519. #ifdef HAVE_ODIVX_POSTPROCESS

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

  521. int use_old_pp=0;

  522. #endif

  523. 

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

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

  526. 

  527. /* -pp Command line Help

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

  529. 

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

  531. 

  532. long form example:

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

  534. short form example:

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

  536. more examples:

  537. -pp tn:64:128:256

  538. 

  539. Filters			Options

  540. short	long name	short	long option	Description

  541. *	*		a	autoq		cpu power dependant enabler

  542. 			c	chrom		chrominance filtring enabled

  543. 			y	nochrom		chrominance filtring disabled

  544. hb	hdeblock	(2 Threshold)		horizontal deblocking filter

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

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

  547. 			the horizontal & vertical deblocking filters share these 

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

  549. vb	vdeblock	(2 Threshold)		vertical deblocking filter

  550. hr	rkhdeblock

  551. vr	rkvdeblock

  552. h1	x1hdeblock				Experimental horizontal deblock filter 1

  553. v1	x1vdeblock				Experimental vertical deblock filter 1

  554. dr	dering					not implemented yet

  555. al	autolevels				automatic brightness / contrast fixer

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

  557. lb	linblenddeint				linear blend deinterlacer

  558. li	linipoldeint				linear interpolating deinterlacer

  559. ci	cubicipoldeint				cubic interpolating deinterlacer

  560. md	mediandeint				median deinterlacer

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

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

  563. tn	tmpnoise	(3 Thresholds)		Temporal Noise Reducer

  564. 			1. <= 2. <= 3. Threshold, larger means stronger filtering

  565. fq	forceQuant	<quantizer>		Force quantizer (for reencoded lq stuff)

  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) printf("%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) printf("%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) printf(" 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) printf("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. 	for(quality=0; quality<GET_PP_QUALITY_MAX+1; quality++)

  766. 	{

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

  768. 

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

  770. 	}

  771. 	newPPFlag=1;

  772. 

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

  774. 	divx_quality= GET_PP_QUALITY_MAX; 

  775. 	return 1;

  776. }

  777. 

  778. /**

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

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

  781.  * mode = quality if newPPFlag

  782.  */

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

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

  785.                  int horizontal_size,   int vertical_size,

  786.                  QP_STORE_T *QP_store,  int QP_stride,

  787. 					  int mode)

  788. {

  789. 	struct PPMode ppMode;

  790. 	static QP_STORE_T zeroArray[2048/8];

  791. 

  792. 	if(newPPFlag)

  793. 	{

  794. 		ppMode= gPPMode[mode];

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

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

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

  798. 

  799. 		return;

  800. 	}

  801. 	

  802. 	if(QP_store==NULL)

  803. 	{

  804. 		QP_store= zeroArray;

  805. 		QP_stride= 0;

  806. 	}

  807. 

  808. 	ppMode.lumMode= mode;

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

  810. 	ppMode.chromMode= mode;

  811. 	ppMode.maxTmpNoise[0]= 700;

  812. 	ppMode.maxTmpNoise[1]= 1500;

  813. 	ppMode.maxTmpNoise[2]= 3000;

  814. 	ppMode.maxAllowedY= 234;

  815. 	ppMode.minAllowedY= 16;

  816. 	ppMode.maxDcDiff= 1;

  817. 

  818. #ifdef HAVE_ODIVX_POSTPROCESS

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

  820. // more function call...

  821. 	if(use_old_pp){

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

  823. 	    return;

  824. 	}

  825. #endif

  826. 

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

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

  829. 

  830. 	horizontal_size >>= 1;

  831. 	vertical_size   >>= 1;

  832. 	src_stride      >>= 1;

  833. 	dst_stride      >>= 1;

  834. 

  835. 	if(ppMode.chromMode)

  836. 	{

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

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

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

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

  841. 	}

  842. 	else if(src_stride == dst_stride)

  843. 	{

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

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

  846. 	}

  847. 	else

  848. 	{

  849. 		int y;

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

  851. 		{

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

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

  854. 		}

  855. 	}

  856. 

  857. #if 0

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

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

  860. #endif

  861. }

  862. 

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

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

  865.                  int horizontal_size,   int vertical_size,

  866.                  QP_STORE_T *QP_store,  int QP_stride,

  867. 		 struct PPMode *mode)

  868. {

  869. 

  870. 	QP_STORE_T quantArray[2048/8];

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

  872. 	{

  873. 		int i;

  874. 		QP_store= quantArray;

  875. 		QP_stride= 0;

  876. 		if(mode->lumMode & FORCE_QUANT)

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

  878. 		else

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

  880. 	}

  881. 

  882. #ifdef HAVE_ODIVX_POSTPROCESS

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

  884. // more function call...

  885. 	if(use_old_pp){

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

  887. 	    mode->oldMode);

  888. 	    return;

  889. 	}

  890. #endif

  891. 

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

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

  894. 

  895. 	horizontal_size >>= 1;

  896. 	vertical_size   >>= 1;

  897. 	src_stride      >>= 1;

  898. 	dst_stride      >>= 1;

  899. 

  900. 	if(mode->chromMode)

  901. 	{

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

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

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

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

  906. 	}

  907. 	else if(src_stride == dst_stride)

  908. 	{

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

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

  911. 	}

  912. 	else

  913. 	{

  914. 		int y;

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

  916. 		{

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

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

  919. 		}

  920. 	}

  921. }

  922. 

  923. 

  924. /**

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

  926.  * with -npp it simply returns quality 

  927.  * 0 <= quality <= 6

  928.  */

  929. int getPpModeForQuality(int quality){

  930. 	int modes[1+GET_PP_QUALITY_MAX]= {

  931. 		0,

  932. #if 1

  933. 		// horizontal filters first

  934. 		LUM_H_DEBLOCK,

  935. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK,

  936. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK,

  937. 		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK,

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

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

  940. #else

  941. 		// vertical filters first

  942. 		LUM_V_DEBLOCK,

  943. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK,

  944. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK,

  945. 		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK,

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

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

  948. #endif

  949. 	};

  950. 

  951. #ifdef HAVE_ODIVX_POSTPROCESS

  952. 	int odivx_modes[1+GET_PP_QUALITY_MAX]= {

  953. 		0,

  954. 		PP_DEBLOCK_Y_H,

  955. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V,

  956. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H,

  957. 		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V,

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

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

  960. 	};

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

  962. #endif

  963. 	if(newPPFlag)	return quality;

  964. 	else		return modes[quality];

  965. }

  966.