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.
3421 Commits
32 Branches
444MB
Tree: 12eebd26ae
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '12eebd26ae'
${ noResults }
FFmpeg/libavcodec/libpostproc/postprocess.c

1143 lines
32KB
Raw Blame History 

  1. /*

  2.     Copyright (C) 2001-2003 Michael Niedermayer (michaelni@gmx.at)

  3. 

  4.     AltiVec optimizations (C) 2004 Romain Dolbeau <romain@dolbeau.org>

  5. 

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

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

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

  9.     (at your option) any later version.

  10. 

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

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

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

  14.     GNU General Public License for more details.

  15. 

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

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

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

  19. */

  20. 

  21. /**

  22.  * @file postprocess.c

  23.  * postprocessing.

  24.  */

  25.  

  26. /*

  27. 			C	MMX	MMX2	3DNow	AltiVec

  28. isVertDC		Ec	Ec			Ec

  29. isVertMinMaxOk		Ec	Ec			Ec

  30. doVertLowPass		E		e	e	Ec

  31. doVertDefFilter		Ec	Ec	e	e	Ec

  32. isHorizDC		Ec	Ec

  33. isHorizMinMaxOk		a	E

  34. doHorizLowPass		E		e	e

  35. doHorizDefFilter	Ec	Ec	e	e

  36. deRing			E		e	e*	Ecp

  37. Vertical RKAlgo1	E		a	a

  38. Horizontal RKAlgo1			a	a

  39. Vertical X1#		a		E	E

  40. Horizontal X1#		a		E	E

  41. LinIpolDeinterlace	e		E	E*

  42. CubicIpolDeinterlace	a		e	e*

  43. LinBlendDeinterlace	e		E	E*

  44. MedianDeinterlace#	E	Ec	Ec

  45. TempDeNoiser#		E		e	e

  46. 

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

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

  49. E = Exact implementation

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

  51. a = alternative / approximate impl

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

  53. p = partially optimized, still some work to do

  54. */

  55. 

  56. /*

  57. TODO:

  58. reduce the time wasted on the mem transfer

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

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

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

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

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

  64. compare the quality & speed of all filters

  65. split this huge file

  66. optimize c versions

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

  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. #ifdef USE_FASTMEMCPY

  87. #include "fastmemcpy.h"

  88. #endif

  89. #include "postprocess.h"

  90. #include "postprocess_internal.h"

  91. 

  92. #include "mangle.h" //FIXME should be supressed

  93. 

  94. #ifndef HAVE_MEMALIGN

  95. #define memalign(a,b) malloc(b)

  96. #endif

  97. 

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

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

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

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

  102. 

  103. #define GET_MODE_BUFFER_SIZE 500

  104. #define OPTIONS_ARRAY_SIZE 10

  105. #define BLOCK_SIZE 8

  106. #define TEMP_STRIDE 8

  107. //#define NUM_BLOCKS_AT_ONCE 16 //not used yet

  108. 

  109. #if defined(__GNUC__) && (__GNUC__ > 3 || __GNUC__ == 3 && __GNUC_MINOR__ > 0)

  110. #    define attribute_used __attribute__((used))

  111. #    define always_inline __attribute__((always_inline)) inline

  112. #else

  113. #    define attribute_used

  114. #    define always_inline inline

  115. #endif

  116. 

  117. #ifdef ARCH_X86

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

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

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

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

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

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

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

  125. #endif

  126. 

  127. static uint8_t clip_table[3*256];

  128. static uint8_t * const clip_tab= clip_table + 256;

  129. 

  130. static const int verbose= 0;

  131. 

  132. static const int attribute_used deringThreshold= 20;

  133. 

  134. 

  135. static struct PPFilter filters[]=

  136. {

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

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

  139. /*	{"hr", "rkhdeblock", 		1, 1, 3, H_RK1_FILTER},

  140. 	{"vr", "rkvdeblock", 		1, 2, 4, V_RK1_FILTER},*/

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

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

  143. 	{"ha", "ahdeblock", 		1, 1, 3, H_A_DEBLOCK},

  144. 	{"va", "avdeblock", 		1, 2, 4, V_A_DEBLOCK},

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

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

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

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

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

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

  151. 	{"fd", "ffmpegdeint", 		1, 1, 4, FFMPEG_DEINT_FILTER},

  152. 	{"l5", "lowpass5", 		1, 1, 4, LOWPASS5_DEINT_FILTER},

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

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

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

  156. };

  157. 

  158. static char *replaceTable[]=

  159. {

  160. 	"default", 	"hdeblock:a,vdeblock:a,dering:a",

  161. 	"de", 		"hdeblock:a,vdeblock:a,dering:a",

  162. 	"fast", 	"x1hdeblock:a,x1vdeblock:a,dering:a",

  163. 	"fa", 		"x1hdeblock:a,x1vdeblock:a,dering:a",

  164. 	"ac", 		"ha:a:128:7,va:a,dering:a",

  165. 	NULL //End Marker

  166. };

  167. 

  168. 

  169. #ifdef ARCH_X86

  170. static inline void prefetchnta(void *p)

  171. {

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

  173. 		: : "r" (p)

  174. 	);

  175. }

  176. 

  177. static inline void prefetcht0(void *p)

  178. {

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

  180. 		: : "r" (p)

  181. 	);

  182. }

  183. 

  184. static inline void prefetcht1(void *p)

  185. {

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

  187. 		: : "r" (p)

  188. 	);

  189. }

  190. 

  191. static inline void prefetcht2(void *p)

  192. {

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

  194. 		: : "r" (p)

  195. 	);

  196. }

  197. #endif

  198. 

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

  200. 

  201. /**

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

  203.  */

  204. static inline int isHorizDC_C(uint8_t src[], int stride, PPContext *c)

  205. {

  206. 	int numEq= 0;

  207. 	int y;

  208. 	const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;

  209. 	const int dcThreshold= dcOffset*2 + 1;

  210. 

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

  212. 	{

  213. 		if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++;

  214. 		if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++;

  215. 		if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++;

  216. 		if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++;

  217. 		if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++;

  218. 		if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++;

  219. 		if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++;

  220. 		src+= stride;

  221. 	}

  222. 	return numEq > c->ppMode.flatnessThreshold;

  223. }

  224. 

  225. /**

  226.  * Check if the middle 8x8 Block in the given 8x16 block is flat

  227.  */

  228. static inline int isVertDC_C(uint8_t src[], int stride, PPContext *c){

  229. 	int numEq= 0;

  230. 	int y;

  231. 	const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;

  232. 	const int dcThreshold= dcOffset*2 + 1;

  233. 

  234. 	src+= stride*4; // src points to begin of the 8x8 Block

  235. 	for(y=0; y<BLOCK_SIZE-1; y++)

  236. 	{

  237. 		if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++;

  238. 		if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++;

  239. 		if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++;

  240. 		if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++;

  241. 		if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++;

  242. 		if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++;

  243. 		if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++;

  244. 		if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++;

  245. 		src+= stride;

  246. 	}

  247. 	return numEq > c->ppMode.flatnessThreshold;

  248. }

  249. 

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

  251. {

  252. 	int i;

  253. #if 1

  254. 	for(i=0; i<2; i++){

  255. 		if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;

  256. 		src += stride;

  257. 		if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;

  258. 		src += stride;

  259. 		if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;

  260. 		src += stride;

  261. 		if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;

  262. 		src += stride;

  263. 	}

  264. #else        

  265. 	for(i=0; i<8; i++){

  266. 		if((unsigned)(src[0] - src[7] + 2*QP) > 4*QP) return 0;

  267. 		src += stride;

  268. 	}

  269. #endif

  270. 	return 1;

  271. }

  272. 

  273. static inline int isVertMinMaxOk_C(uint8_t src[], int stride, int QP)

  274. {

  275. #if 1

  276. #if 1

  277. 	int x;

  278. 	src+= stride*4;

  279. 	for(x=0; x<BLOCK_SIZE; x+=4)

  280. 	{

  281. 		if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;

  282. 		if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;

  283. 		if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;

  284. 		if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;

  285. 	}

  286. #else

  287. 	int x;

  288. 	src+= stride*3;

  289. 	for(x=0; x<BLOCK_SIZE; x++)

  290. 	{

  291. 		if((unsigned)(src[x + stride] - src[x + (stride<<3)] + 2*QP) > 4*QP) return 0;

  292. 	}

  293. #endif

  294. 	return 1;

  295. #else

  296. 	int x;

  297. 	src+= stride*4;

  298. 	for(x=0; x<BLOCK_SIZE; x++)

  299. 	{

  300. 		int min=255;

  301. 		int max=0;

  302. 		int y;

  303. 		for(y=0; y<8; y++){

  304. 			int v= src[x + y*stride];

  305. 			if(v>max) max=v;

  306. 			if(v<min) min=v;

  307. 		}

  308. 		if(max-min > 2*QP) return 0;

  309. 	}

  310. 	return 1;

  311. #endif

  312. }

  313. 

  314. static inline int horizClassify_C(uint8_t src[], int stride, PPContext *c){

  315. 	if( isHorizDC_C(src, stride, c) ){

  316. 		if( isHorizMinMaxOk_C(src, stride, c->QP) )

  317. 			return 1;

  318. 		else

  319. 			return 0;

  320. 	}else{

  321. 		return 2;

  322. 	}

  323. }

  324. 

  325. static inline int vertClassify_C(uint8_t src[], int stride, PPContext *c){

  326. 	if( isVertDC_C(src, stride, c) ){

  327. 		if( isVertMinMaxOk_C(src, stride, c->QP) )

  328. 			return 1;

  329. 		else

  330. 			return 0;

  331. 	}else{

  332. 		return 2;

  333. 	}

  334. }

  335. 

  336. static inline void doHorizDefFilter_C(uint8_t dst[], int stride, PPContext *c)

  337. {

  338. 	int y;

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

  340. 	{

  341. 		const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);

  342. 

  343. 		if(ABS(middleEnergy) < 8*c->QP)

  344. 		{

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

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

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

  348. 

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

  350. 			d= MAX(d, 0);

  351. 

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

  353. 			d*= SIGN(-middleEnergy);

  354. 

  355. 			if(q>0)

  356. 			{

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

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

  359. 			}

  360. 			else

  361. 			{

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

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

  364. 			}

  365. 

  366.         		dst[3]-= d;

  367. 	        	dst[4]+= d;

  368. 		}

  369. 		dst+= stride;

  370. 	}

  371. }

  372. 

  373. /**

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

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

  376.  */

  377. static inline void doHorizLowPass_C(uint8_t dst[], int stride, PPContext *c)

  378. {

  379. 

  380. 	int y;

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

  382. 	{

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

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

  385. 

  386. 		int sums[9];

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

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

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

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

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

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

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

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

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

  396. 

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

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

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

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

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

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

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

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

  405. 

  406. 		dst+= stride;

  407. 	}

  408. }

  409. 

  410. /**

  411.  * Experimental Filter 1 (Horizontal)

  412.  * will not damage linear gradients

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

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

  415.  * MMX2 version does correct clipping C version doesnt

  416.  * not identical with the vertical one

  417.  */

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

  419. {

  420. 	int y;

  421. 	static uint64_t *lut= NULL;

  422. 	if(lut==NULL)

  423. 	{

  424. 		int i;

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

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

  427. 		{

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

  429. /*

  430. //Simulate 112242211 9-Tap filter

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

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

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

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

  435. */

  436. //Simulate piecewise linear interpolation

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

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

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

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

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

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

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

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

  445. 

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

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

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

  449. 		}

  450. 	}

  451. 

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

  453. 	{

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

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

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

  457. 

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

  459. 

  460. 		if(d < QP)

  461. 		{

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

  463. 

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

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

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

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

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

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

  470. 

  471. 		}

  472. 		src+=stride;

  473. 	}

  474. }

  475. 

  476. /**

  477.  * accurate deblock filter

  478.  */

  479. static always_inline void do_a_deblock(uint8_t *src, int step, int stride, PPContext *c){

  480. 	int y;

  481. 	const int QP= c->QP;

  482. 	const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;

  483. 	const int dcThreshold= dcOffset*2 + 1;

  484. //START_TIMER

  485. 	src+= step*4; // src points to begin of the 8x8 Block

  486. 	for(y=0; y<8; y++){

  487. 		int numEq= 0;

  488. 

  489. 		if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++;

  490. 		if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++;

  491. 		if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++;

  492. 		if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++;

  493. 		if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++;

  494. 		if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++;

  495. 		if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++;

  496. 		if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++;

  497. 		if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++;

  498. 		if(numEq > c->ppMode.flatnessThreshold){

  499. 			int min, max, x;

  500. 			

  501. 			if(src[0] > src[step]){

  502. 			    max= src[0];

  503. 			    min= src[step];

  504. 			}else{

  505. 			    max= src[step];

  506. 			    min= src[0];

  507. 			}

  508. 			for(x=2; x<8; x+=2){

  509. 				if(src[x*step] > src[(x+1)*step]){

  510. 					if(src[x    *step] > max) max= src[ x   *step];

  511. 					if(src[(x+1)*step] < min) min= src[(x+1)*step];

  512. 				}else{

  513. 					if(src[(x+1)*step] > max) max= src[(x+1)*step];

  514. 					if(src[ x   *step] < min) min= src[ x   *step];

  515. 				}

  516. 			}

  517. 			if(max-min < 2*QP){

  518. 				const int first= ABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];

  519. 				const int last= ABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];

  520. 				

  521. 				int sums[10];

  522. 				sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;

  523. 				sums[1] = sums[0] - first       + src[3*step];

  524. 				sums[2] = sums[1] - first       + src[4*step];

  525. 				sums[3] = sums[2] - first       + src[5*step];

  526. 				sums[4] = sums[3] - first       + src[6*step];

  527. 				sums[5] = sums[4] - src[0*step] + src[7*step];

  528. 				sums[6] = sums[5] - src[1*step] + last;

  529. 				sums[7] = sums[6] - src[2*step] + last;

  530. 				sums[8] = sums[7] - src[3*step] + last;

  531. 				sums[9] = sums[8] - src[4*step] + last;

  532. 

  533. 				src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;

  534. 				src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;

  535. 				src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;

  536. 				src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;

  537. 				src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;

  538. 				src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;

  539. 				src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;

  540. 				src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;

  541. 			}

  542. 		}else{

  543. 			const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);

  544. 

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

  546. 			{

  547. 				const int q=(src[3*step] - src[4*step])/2;

  548. 				const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);

  549. 				const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);

  550. 

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

  552. 				d= MAX(d, 0);

  553. 	

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

  555. 				d*= SIGN(-middleEnergy);

  556. 	

  557. 				if(q>0)

  558. 				{

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

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

  561. 				}

  562. 				else

  563. 				{

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

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

  566. 				}

  567. 	

  568. 				src[3*step]-= d;

  569. 				src[4*step]+= d;

  570. 			}

  571. 		}

  572. 

  573. 		src += stride;

  574. 	}

  575. /*if(step==16){

  576.     STOP_TIMER("step16")

  577. }else{

  578.     STOP_TIMER("stepX")

  579. }*/

  580. }

  581. 

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

  583. //Plain C versions

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

  585. #define COMPILE_C

  586. #endif

  587. 

  588. #ifdef ARCH_POWERPC

  589. #ifdef HAVE_ALTIVEC

  590. #define COMPILE_ALTIVEC

  591. #ifndef CONFIG_DARWIN

  592. #warning "################################################################################"

  593. #warning  "WARNING: No gcc available as of today (2004-05-25) seems to be able to compile properly some of the code under non-Darwin PPC OSes. Some functions result in wrong results, while others simply won't compile (gcc explodes after allocating 1GiB+)."

  594. #warning "################################################################################"

  595. #endif //CONFIG_DARWIN

  596. #endif //HAVE_ALTIVEC

  597. #endif //ARCH_POWERPC

  598. 

  599. #ifdef ARCH_X86

  600. 

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

  602. #define COMPILE_MMX

  603. #endif

  604. 

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

  606. #define COMPILE_MMX2

  607. #endif

  608. 

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

  610. #define COMPILE_3DNOW

  611. #endif

  612. #endif //ARCH_X86

  613. 

  614. #undef HAVE_MMX

  615. #undef HAVE_MMX2

  616. #undef HAVE_3DNOW

  617. #undef HAVE_ALTIVEC

  618. #undef ARCH_X86

  619. 

  620. #ifdef COMPILE_C

  621. #undef HAVE_MMX

  622. #undef HAVE_MMX2

  623. #undef HAVE_3DNOW

  624. #undef ARCH_X86

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

  626. #include "postprocess_template.c"

  627. #endif

  628. 

  629. #ifdef ARCH_POWERPC

  630. #ifdef COMPILE_ALTIVEC

  631. #undef RENAME

  632. #define HAVE_ALTIVEC

  633. #define RENAME(a) a ## _altivec

  634. #include "postprocess_altivec_template.c"

  635. #include "postprocess_template.c"

  636. #endif

  637. #endif //ARCH_POWERPC

  638. 

  639. //MMX versions

  640. #ifdef COMPILE_MMX

  641. #undef RENAME

  642. #define HAVE_MMX

  643. #undef HAVE_MMX2

  644. #undef HAVE_3DNOW

  645. #define ARCH_X86

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

  647. #include "postprocess_template.c"

  648. #endif

  649. 

  650. //MMX2 versions

  651. #ifdef COMPILE_MMX2

  652. #undef RENAME

  653. #define HAVE_MMX

  654. #define HAVE_MMX2

  655. #undef HAVE_3DNOW

  656. #define ARCH_X86

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

  658. #include "postprocess_template.c"

  659. #endif

  660. 

  661. //3DNOW versions

  662. #ifdef COMPILE_3DNOW

  663. #undef RENAME

  664. #define HAVE_MMX

  665. #undef HAVE_MMX2

  666. #define HAVE_3DNOW

  667. #define ARCH_X86

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

  669. #include "postprocess_template.c"

  670. #endif

  671. 

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

  673. 

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

  675. 	QP_STORE_T QPs[], int QPStride, int isColor, pp_mode_t *vm, pp_context_t *vc)

  676. {

  677. 	PPContext *c= (PPContext *)vc;

  678. 	PPMode *ppMode= (PPMode *)vm;

  679. 	c->ppMode= *ppMode; //FIXME

  680. 

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

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

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

  684. #ifdef RUNTIME_CPUDETECT

  685. #ifdef ARCH_X86

  686. 	// ordered per speed fasterst first

  687. 	if(c->cpuCaps & PP_CPU_CAPS_MMX2)

  688. 		postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  689. 	else if(c->cpuCaps & PP_CPU_CAPS_3DNOW)

  690. 		postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  691. 	else if(c->cpuCaps & PP_CPU_CAPS_MMX)

  692. 		postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  693. 	else

  694. 		postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  695. #else

  696. #ifdef ARCH_POWERPC

  697. #ifdef HAVE_ALTIVEC

  698.         else if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC)

  699. 		postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  700.         else

  701. #endif

  702. #endif

  703. 		postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  704. #endif

  705. #else //RUNTIME_CPUDETECT

  706. #ifdef HAVE_MMX2

  707. 		postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  708. #elif defined (HAVE_3DNOW)

  709. 		postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  710. #elif defined (HAVE_MMX)

  711. 		postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  712. #elif defined (HAVE_ALTIVEC)

  713. 		postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  714. #else

  715. 		postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  716. #endif

  717. #endif //!RUNTIME_CPUDETECT

  718. }

  719. 

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

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

  722. 

  723. /* -pp Command line Help

  724. */

  725. char *pp_help=

  726. "<filterName>[:<option>[:<option>...]][[,|/][-]<filterName>[:<option>...]]...\n"

  727. "long form example:\n"

  728. "vdeblock:autoq/hdeblock:autoq/linblenddeint	default,-vdeblock\n"

  729. "short form example:\n"

  730. "vb:a/hb:a/lb					de,-vb\n"

  731. "more examples:\n"

  732. "tn:64:128:256\n"

  733. "Filters			Options\n"

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

  735. "*	*		a	autoq		CPU power dependent enabler\n"

  736. "			c	chrom		chrominance filtering enabled\n"

  737. "			y	nochrom		chrominance filtering disabled\n"

  738. "hb	hdeblock	(2 threshold)		horizontal deblocking filter\n"

  739. "	1. difference factor: default=32, higher -> more deblocking\n"

  740. "	2. flatness threshold: default=39, lower -> more deblocking\n"

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

  742. "			so you can't set different thresholds for h / v\n"

  743. "vb	vdeblock	(2 threshold)		vertical deblocking filter\n"

  744. "ha	hadeblock	(2 threshold)		horizontal deblocking filter\n"

  745. "va	vadeblock	(2 threshold)		vertical deblocking filter\n"

  746. "h1	x1hdeblock				experimental h deblock filter 1\n"

  747. "v1	x1vdeblock				experimental v deblock filter 1\n"

  748. "dr	dering					deringing filter\n"

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

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

  751. "lb	linblenddeint				linear blend deinterlacer\n"

  752. "li	linipoldeint				linear interpolating deinterlace\n"

  753. "ci	cubicipoldeint				cubic interpolating deinterlacer\n"

  754. "md	mediandeint				median deinterlacer\n"

  755. "fd	ffmpegdeint				ffmpeg deinterlacer\n"

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

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

  758. "tn	tmpnoise	(3 threshold)		temporal noise reducer\n"

  759. "			1. <= 2. <= 3.		larger -> stronger filtering\n"

  760. "fq	forceQuant	<quantizer>		force quantizer\n"

  761. ;

  762. 

  763. pp_mode_t *pp_get_mode_by_name_and_quality(char *name, int quality)

  764. {

  765. 	char temp[GET_MODE_BUFFER_SIZE];

  766. 	char *p= temp;

  767. 	char *filterDelimiters= ",/";

  768. 	char *optionDelimiters= ":";

  769. 	struct PPMode *ppMode;

  770. 	char *filterToken;

  771. 

  772. 	ppMode= memalign(8, sizeof(PPMode));

  773. 	

  774. 	ppMode->lumMode= 0;

  775. 	ppMode->chromMode= 0;

  776. 	ppMode->maxTmpNoise[0]= 700;

  777. 	ppMode->maxTmpNoise[1]= 1500;

  778. 	ppMode->maxTmpNoise[2]= 3000;

  779. 	ppMode->maxAllowedY= 234;

  780. 	ppMode->minAllowedY= 16;

  781. 	ppMode->baseDcDiff= 256/8;

  782. 	ppMode->flatnessThreshold= 56-16-1;

  783. 	ppMode->maxClippedThreshold= 0.01;

  784. 	ppMode->error=0;

  785. 

  786. 	strncpy(temp, name, GET_MODE_BUFFER_SIZE);

  787. 

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

  789. 

  790. 	for(;;){

  791. 		char *filterName;

  792. 		int q= 1000000; //PP_QUALITY_MAX;

  793. 		int chrom=-1;

  794. 		char *option;

  795. 		char *options[OPTIONS_ARRAY_SIZE];

  796. 		int i;

  797. 		int filterNameOk=0;

  798. 		int numOfUnknownOptions=0;

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

  800. 

  801. 		filterToken= strtok(p, filterDelimiters);

  802. 		if(filterToken == NULL) break;

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

  804. 		filterName= strtok(filterToken, optionDelimiters);

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

  806. 

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

  808. 		{

  809. 			enable=0;

  810. 			filterName++;

  811. 		}

  812. 

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

  814. 			option= strtok(NULL, optionDelimiters);

  815. 			if(option == NULL) break;

  816. 

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

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

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

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

  821. 			else

  822. 			{

  823. 				options[numOfUnknownOptions] = option;

  824. 				numOfUnknownOptions++;

  825. 			}

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

  827. 		}

  828. 		options[numOfUnknownOptions] = NULL;

  829. 

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

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

  832. 		{

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

  834. 			{

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

  836. 				int plen;

  837. 				int spaceLeft;

  838. 

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

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

  841. 

  842. 				plen= strlen(p);

  843. 				spaceLeft= p - temp + plen;

  844. 				if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE)

  845. 				{

  846. 					ppMode->error++;

  847. 					break;

  848. 				}

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

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

  851. 				filterNameOk=1;

  852. 			}

  853. 		}

  854. 

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

  856. 		{

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

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

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

  860. 			{

  861. 				ppMode->lumMode &= ~filters[i].mask;

  862. 				ppMode->chromMode &= ~filters[i].mask;

  863. 

  864. 				filterNameOk=1;

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

  866. 

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

  868. 					ppMode->lumMode|= filters[i].mask;

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

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

  871. 						ppMode->chromMode|= filters[i].mask;

  872. 

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

  874. 				{

  875. 					int o;

  876. 					ppMode->minAllowedY= 16;

  877. 					ppMode->maxAllowedY= 234;

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

  879. 					{

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

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

  882. 						{

  883. 							ppMode->minAllowedY= 0;

  884. 							ppMode->maxAllowedY= 255;

  885. 							numOfUnknownOptions--;

  886. 						}

  887. 					}

  888. 				}

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

  890. 				{

  891. 					int o;

  892. 					int numOfNoises=0;

  893. 

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

  895. 					{

  896. 						char *tail;

  897. 						ppMode->maxTmpNoise[numOfNoises]=

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

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

  900. 						{

  901. 							numOfNoises++;

  902. 							numOfUnknownOptions--;

  903. 							if(numOfNoises >= 3) break;

  904. 						}

  905. 					}

  906. 				}

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

  908. 				     || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK)

  909. 				{

  910. 					int o;

  911. 

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

  913. 					{

  914. 						char *tail;

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

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

  917. 

  918. 						numOfUnknownOptions--;

  919. 						if(o==0) ppMode->baseDcDiff= val;

  920. 						else ppMode->flatnessThreshold= val;

  921. 					}

  922. 				}

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

  924. 				{

  925. 					int o;

  926. 					ppMode->forcedQuant= 15;

  927. 

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

  929. 					{

  930. 						char *tail;

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

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

  933. 

  934. 						numOfUnknownOptions--;

  935. 						ppMode->forcedQuant= val;

  936. 					}

  937. 				}

  938. 			}

  939. 		}

  940. 		if(!filterNameOk) ppMode->error++;

  941. 		ppMode->error += numOfUnknownOptions;

  942. 	}

  943. 

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

  945. 	if(ppMode->error)

  946. 	{

  947. 		fprintf(stderr, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);

  948. 		free(ppMode);

  949. 		return NULL;

  950. 	}

  951. 	return ppMode;

  952. }

  953. 

  954. void pp_free_mode(pp_mode_t *mode){

  955.     if(mode) free(mode);

  956. }

  957. 

  958. static void reallocAlign(void **p, int alignment, int size){

  959. 	if(*p) free(*p);

  960. 	*p= memalign(alignment, size);

  961. 	memset(*p, 0, size);

  962. }

  963. 

  964. static void reallocBuffers(PPContext *c, int width, int height, int stride, int qpStride){

  965. 	int mbWidth = (width+15)>>4;

  966. 	int mbHeight= (height+15)>>4;

  967. 	int i;

  968. 

  969. 	c->stride= stride;

  970. 	c->qpStride= qpStride;

  971. 

  972. 	reallocAlign((void **)&c->tempDst, 8, stride*24);

  973. 	reallocAlign((void **)&c->tempSrc, 8, stride*24);

  974. 	reallocAlign((void **)&c->tempBlocks, 8, 2*16*8);

  975. 	reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t));

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

  977. 		c->yHistogram[i]= width*height/64*15/256;

  978. 

  979. 	for(i=0; i<3; i++)

  980. 	{

  981. 		//Note:the +17*1024 is just there so i dont have to worry about r/w over te end

  982. 		reallocAlign((void **)&c->tempBlured[i], 8, stride*mbHeight*16 + 17*1024);

  983. 		reallocAlign((void **)&c->tempBluredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size

  984. 	}

  985. 

  986. 	reallocAlign((void **)&c->deintTemp, 8, 2*width+32);

  987. 	reallocAlign((void **)&c->nonBQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));

  988. 	reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));

  989. 	reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));

  990. }

  991. 

  992. static void global_init(void){

  993. 	int i;

  994. 	memset(clip_table, 0, 256);

  995. 	for(i=256; i<512; i++)

  996. 		clip_table[i]= i;

  997. 	memset(clip_table+512, 0, 256);

  998. }

  999. 

  1000. pp_context_t *pp_get_context(int width, int height, int cpuCaps){

  1001. 	PPContext *c= memalign(32, sizeof(PPContext));

  1002. 	int stride= (width+15)&(~15); //assumed / will realloc if needed

  1003. 	int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed

  1004.         

  1005. 	global_init();

  1006. 

  1007. 	memset(c, 0, sizeof(PPContext));

  1008. 	c->cpuCaps= cpuCaps;

  1009. 	if(cpuCaps&PP_FORMAT){

  1010. 		c->hChromaSubSample= cpuCaps&0x3;

  1011. 		c->vChromaSubSample= (cpuCaps>>4)&0x3;

  1012. 	}else{

  1013. 		c->hChromaSubSample= 1;

  1014. 		c->vChromaSubSample= 1;

  1015. 	}

  1016. 

  1017. 	reallocBuffers(c, width, height, stride, qpStride);

  1018.         

  1019. 	c->frameNum=-1;

  1020. 

  1021. 	return c;

  1022. }

  1023. 

  1024. void pp_free_context(void *vc){

  1025. 	PPContext *c = (PPContext*)vc;

  1026. 	int i;

  1027. 	

  1028. 	for(i=0; i<3; i++) free(c->tempBlured[i]);

  1029. 	for(i=0; i<3; i++) free(c->tempBluredPast[i]);

  1030. 	

  1031. 	free(c->tempBlocks);

  1032. 	free(c->yHistogram);

  1033. 	free(c->tempDst);

  1034. 	free(c->tempSrc);

  1035. 	free(c->deintTemp);

  1036. 	free(c->stdQPTable);

  1037. 	free(c->nonBQPTable);

  1038. 	free(c->forcedQPTable);

  1039.         

  1040. 	memset(c, 0, sizeof(PPContext));

  1041. 

  1042. 	free(c);

  1043. }

  1044. 

  1045. void  pp_postprocess(uint8_t * src[3], int srcStride[3],

  1046.                  uint8_t * dst[3], int dstStride[3],

  1047.                  int width, int height,

  1048.                  QP_STORE_T *QP_store,  int QPStride,

  1049. 		 pp_mode_t *vm,  void *vc, int pict_type)

  1050. {

  1051. 	int mbWidth = (width+15)>>4;

  1052. 	int mbHeight= (height+15)>>4;

  1053. 	PPMode *mode = (PPMode*)vm;

  1054. 	PPContext *c = (PPContext*)vc;

  1055.         int minStride= MAX(srcStride[0], dstStride[0]);

  1056. 

  1057. 	if(c->stride < minStride || c->qpStride < QPStride)

  1058. 		reallocBuffers(c, width, height, 

  1059. 				MAX(minStride, c->stride), 

  1060. 				MAX(c->qpStride, QPStride));

  1061. 

  1062. 	if(QP_store==NULL || (mode->lumMode & FORCE_QUANT)) 

  1063. 	{

  1064. 		int i;

  1065. 		QP_store= c->forcedQPTable;

  1066. 		QPStride= 0;

  1067. 		if(mode->lumMode & FORCE_QUANT)

  1068. 			for(i=0; i<mbWidth; i++) QP_store[i]= mode->forcedQuant;

  1069. 		else

  1070. 			for(i=0; i<mbWidth; i++) QP_store[i]= 1;

  1071. 	}

  1072. //printf("pict_type:%d\n", pict_type);

  1073. 

  1074. 	if(pict_type & PP_PICT_TYPE_QP2){

  1075. 		int i;

  1076. 		const int count= mbHeight * QPStride;

  1077. 		for(i=0; i<(count>>2); i++){

  1078. 			((uint32_t*)c->stdQPTable)[i] = (((uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F;

  1079. 		}

  1080. 		for(i<<=2; i<count; i++){

  1081. 			c->stdQPTable[i] = QP_store[i]>>1;

  1082. 		}

  1083.                 QP_store= c->stdQPTable;

  1084. 	}

  1085. 

  1086. if(0){

  1087. int x,y;

  1088. for(y=0; y<mbHeight; y++){

  1089. 	for(x=0; x<mbWidth; x++){

  1090. 		printf("%2d ", QP_store[x + y*QPStride]);

  1091. 	}

  1092. 	printf("\n");

  1093. }

  1094. 	printf("\n");

  1095. }

  1096. 

  1097. 	if((pict_type&7)!=3)

  1098. 	{

  1099. 		int i;

  1100. 		const int count= mbHeight * QPStride;

  1101. 		for(i=0; i<(count>>2); i++){

  1102. 			((uint32_t*)c->nonBQPTable)[i] = ((uint32_t*)QP_store)[i] & 0x3F3F3F3F;

  1103. 		}

  1104. 		for(i<<=2; i<count; i++){

  1105. 			c->nonBQPTable[i] = QP_store[i] & 0x3F;

  1106. 		}

  1107. 	}

  1108. 

  1109. 	if(verbose>2)

  1110. 	{

  1111. 		printf("using npp filters 0x%X/0x%X\n", mode->lumMode, mode->chromMode);

  1112. 	}

  1113. 

  1114. 	postProcess(src[0], srcStride[0], dst[0], dstStride[0],

  1115. 		width, height, QP_store, QPStride, 0, mode, c);

  1116. 

  1117. 	width  = (width )>>c->hChromaSubSample;

  1118. 	height = (height)>>c->vChromaSubSample;

  1119. 

  1120. 	if(mode->chromMode)

  1121. 	{

  1122. 		postProcess(src[1], srcStride[1], dst[1], dstStride[1],

  1123. 			width, height, QP_store, QPStride, 1, mode, c);

  1124. 		postProcess(src[2], srcStride[2], dst[2], dstStride[2],

  1125. 			width, height, QP_store, QPStride, 2, mode, c);

  1126. 	}

  1127. 	else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2])

  1128. 	{

  1129. 		memcpy(dst[1], src[1], srcStride[1]*height);

  1130. 		memcpy(dst[2], src[2], srcStride[2]*height);

  1131. 	}

  1132. 	else

  1133. 	{

  1134. 		int y;

  1135. 		for(y=0; y<height; y++)

  1136. 		{

  1137. 			memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);

  1138. 			memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);

  1139. 		}

  1140. 	}

  1141. }