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.
3422 Commits
32 Branches
303MB
Tree: 8c8bbd10e0
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '8c8bbd10e0'
${ noResults }
FFmpeg/libavcodec/libpostproc/postprocess_template.c

3259 lines
96KB
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.  * @file postprocess_template.c

  21.  * mmx/mmx2/3dnow postprocess code.

  22.  */

  23. 

  24. 

  25. #undef PAVGB

  26. #undef PMINUB

  27. #undef PMAXUB

  28. 

  29. #ifdef HAVE_MMX2

  30. #define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"

  31. #elif defined (HAVE_3DNOW)

  32. #define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"

  33. #endif

  34. 

  35. #ifdef HAVE_MMX2

  36. #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"

  37. #elif defined (HAVE_MMX)

  38. #define PMINUB(b,a,t) \

  39. 	"movq " #a ", " #t " \n\t"\

  40. 	"psubusb " #b ", " #t " \n\t"\

  41. 	"psubb " #t ", " #a " \n\t"

  42. #endif

  43. 

  44. #ifdef HAVE_MMX2

  45. #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"

  46. #elif defined (HAVE_MMX)

  47. #define PMAXUB(a,b) \

  48. 	"psubusb " #a ", " #b " \n\t"\

  49. 	"paddb " #a ", " #b " \n\t"

  50. #endif

  51. 

  52. //FIXME? |255-0| = 1 (shouldnt be a problem ...)

  53. #ifdef HAVE_MMX

  54. /**

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

  56.  */

  57. static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){

  58. 	int numEq= 0, dcOk;

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

  60. asm volatile(

  61. 		"movq %0, %%mm7					\n\t" 

  62. 		"movq %1, %%mm6					\n\t" 

  63.                 : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])

  64.                 );

  65.                 

  66. asm volatile(

  67. 		"leal (%2, %3), %%eax				\n\t"

  68. //	0	1	2	3	4	5	6	7	8	9

  69. //	%1	eax	eax+%2	eax+2%2	%1+4%2	ecx	ecx+%2	ecx+2%2	%1+8%2	ecx+4%2

  70. 

  71. 		"movq (%2), %%mm0				\n\t"

  72. 		"movq (%%eax), %%mm1				\n\t"

  73.                 "movq %%mm0, %%mm3				\n\t"

  74.                 "movq %%mm0, %%mm4				\n\t"

  75.                 PMAXUB(%%mm1, %%mm4)

  76.                 PMINUB(%%mm1, %%mm3, %%mm5)

  77. 		"psubb %%mm1, %%mm0				\n\t" // mm0 = differnece

  78. 		"paddb %%mm7, %%mm0				\n\t"

  79. 		"pcmpgtb %%mm6, %%mm0				\n\t"

  80. 

  81. 		"movq (%%eax,%3), %%mm2				\n\t"

  82.                 PMAXUB(%%mm2, %%mm4)

  83.                 PMINUB(%%mm2, %%mm3, %%mm5)

  84. 		"psubb %%mm2, %%mm1				\n\t"

  85. 		"paddb %%mm7, %%mm1				\n\t"

  86. 		"pcmpgtb %%mm6, %%mm1				\n\t"

  87. 		"paddb %%mm1, %%mm0				\n\t"

  88. 

  89. 		"movq (%%eax, %3, 2), %%mm1			\n\t"

  90.                 PMAXUB(%%mm1, %%mm4)

  91.                 PMINUB(%%mm1, %%mm3, %%mm5)

  92. 		"psubb %%mm1, %%mm2				\n\t"

  93. 		"paddb %%mm7, %%mm2				\n\t"

  94. 		"pcmpgtb %%mm6, %%mm2				\n\t"

  95. 		"paddb %%mm2, %%mm0				\n\t"

  96. 		

  97. 		"leal (%%eax, %3, 4), %%eax			\n\t"

  98. 

  99. 		"movq (%2, %3, 4), %%mm2			\n\t"

  100.                 PMAXUB(%%mm2, %%mm4)

  101.                 PMINUB(%%mm2, %%mm3, %%mm5)

  102. 		"psubb %%mm2, %%mm1				\n\t"

  103. 		"paddb %%mm7, %%mm1				\n\t"

  104. 		"pcmpgtb %%mm6, %%mm1				\n\t"

  105. 		"paddb %%mm1, %%mm0				\n\t"

  106. 

  107. 		"movq (%%eax), %%mm1				\n\t"

  108.                 PMAXUB(%%mm1, %%mm4)

  109.                 PMINUB(%%mm1, %%mm3, %%mm5)

  110. 		"psubb %%mm1, %%mm2				\n\t"

  111. 		"paddb %%mm7, %%mm2				\n\t"

  112. 		"pcmpgtb %%mm6, %%mm2				\n\t"

  113. 		"paddb %%mm2, %%mm0				\n\t"

  114. 

  115. 		"movq (%%eax, %3), %%mm2			\n\t"

  116.                 PMAXUB(%%mm2, %%mm4)

  117.                 PMINUB(%%mm2, %%mm3, %%mm5)

  118. 		"psubb %%mm2, %%mm1				\n\t"

  119. 		"paddb %%mm7, %%mm1				\n\t"

  120. 		"pcmpgtb %%mm6, %%mm1				\n\t"

  121. 		"paddb %%mm1, %%mm0				\n\t"

  122. 

  123. 		"movq (%%eax, %3, 2), %%mm1			\n\t"

  124.                 PMAXUB(%%mm1, %%mm4)

  125.                 PMINUB(%%mm1, %%mm3, %%mm5)

  126. 		"psubb %%mm1, %%mm2				\n\t"

  127. 		"paddb %%mm7, %%mm2				\n\t"

  128. 		"pcmpgtb %%mm6, %%mm2				\n\t"

  129. 		"paddb %%mm2, %%mm0				\n\t"

  130. 		"psubusb %%mm3, %%mm4				\n\t"

  131. 

  132. 		"						\n\t"

  133. #ifdef HAVE_MMX2

  134. 		"pxor %%mm7, %%mm7				\n\t"

  135. 		"psadbw %%mm7, %%mm0				\n\t"

  136. #else

  137. 		"movq %%mm0, %%mm1				\n\t"

  138. 		"psrlw $8, %%mm0				\n\t"

  139. 		"paddb %%mm1, %%mm0				\n\t"

  140. 		"movq %%mm0, %%mm1				\n\t"

  141. 		"psrlq $16, %%mm0				\n\t"

  142. 		"paddb %%mm1, %%mm0				\n\t"

  143. 		"movq %%mm0, %%mm1				\n\t"

  144. 		"psrlq $32, %%mm0				\n\t"

  145. 		"paddb %%mm1, %%mm0				\n\t"

  146. #endif

  147.                 "movq %4, %%mm7					\n\t" // QP,..., QP

  148. 		"paddusb %%mm7, %%mm7				\n\t" // 2QP ... 2QP

  149. 		"psubusb %%mm7, %%mm4				\n\t" // Diff <= 2QP -> 0

  150. 		"packssdw %%mm4, %%mm4				\n\t"

  151. 		"movd %%mm0, %0					\n\t"

  152. 		"movd %%mm4, %1					\n\t"

  153. 

  154. 		: "=r" (numEq), "=r" (dcOk)

  155. 		: "r" (src), "r" (stride), "m" (c->pQPb)

  156. 		: "%eax"

  157. 		);

  158. 

  159. 	numEq= (-numEq) &0xFF;

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

  161.             if(dcOk) return 0;

  162.             else     return 1;

  163.         }else{

  164.             return 2;

  165.         }

  166. }

  167. #endif

  168. 

  169. /**

  170.  * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)

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

  172.  */

  173. #ifndef HAVE_ALTIVEC

  174. static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)

  175. {

  176. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  177. 	src+= stride*3;

  178. 	asm volatile(	//"movv %0 %1 %2\n\t"

  179. 		"movq %2, %%mm0			\n\t"  // QP,..., QP

  180. 		"pxor %%mm4, %%mm4				\n\t"

  181. 

  182. 		"movq (%0), %%mm6				\n\t"

  183. 		"movq (%0, %1), %%mm5				\n\t"

  184. 		"movq %%mm5, %%mm1				\n\t"

  185. 		"movq %%mm6, %%mm2				\n\t"

  186. 		"psubusb %%mm6, %%mm5				\n\t"

  187. 		"psubusb %%mm1, %%mm2				\n\t"

  188. 		"por %%mm5, %%mm2				\n\t" // ABS Diff of lines

  189. 		"psubusb %%mm0, %%mm2				\n\t" // diff <= QP -> 0

  190. 		"pcmpeqb %%mm4, %%mm2			\n\t" // diff <= QP -> FF

  191. 

  192. 		"pand %%mm2, %%mm6				\n\t"

  193. 		"pandn %%mm1, %%mm2				\n\t"

  194. 		"por %%mm2, %%mm6				\n\t"// First Line to Filter

  195. 

  196. 		"movq (%0, %1, 8), %%mm5			\n\t"

  197. 		"leal (%0, %1, 4), %%eax			\n\t"

  198. 		"leal (%0, %1, 8), %%ecx			\n\t"

  199. 		"subl %1, %%ecx					\n\t"

  200. 		"addl %1, %0					\n\t" // %0 points to line 1 not 0

  201. 		"movq (%0, %1, 8), %%mm7			\n\t"

  202. 		"movq %%mm5, %%mm1				\n\t"

  203. 		"movq %%mm7, %%mm2				\n\t"

  204. 		"psubusb %%mm7, %%mm5				\n\t"

  205. 		"psubusb %%mm1, %%mm2				\n\t"

  206. 		"por %%mm5, %%mm2				\n\t" // ABS Diff of lines

  207. 		"psubusb %%mm0, %%mm2				\n\t" // diff <= QP -> 0

  208. 		"pcmpeqb %%mm4, %%mm2			\n\t" // diff <= QP -> FF

  209. 

  210. 		"pand %%mm2, %%mm7				\n\t"

  211. 		"pandn %%mm1, %%mm2				\n\t"

  212. 		"por %%mm2, %%mm7				\n\t" // First Line to Filter

  213. 

  214. 

  215. 		// 	1	2	3	4	5	6	7	8

  216. 		//	%0	%0+%1	%0+2%1	eax	%0+4%1	eax+2%1	ecx	eax+4%1

  217. 		// 6 4 2 2 1 1

  218. 		// 6 4 4 2

  219. 		// 6 8 2

  220. 

  221. 		"movq (%0, %1), %%mm0				\n\t" //  1

  222. 		"movq %%mm0, %%mm1				\n\t" //  1

  223. 		PAVGB(%%mm6, %%mm0)				      //1 1	/2

  224. 		PAVGB(%%mm6, %%mm0)				      //3 1	/4

  225. 

  226. 		"movq (%0, %1, 4), %%mm2			\n\t" //     1

  227. 		"movq %%mm2, %%mm5				\n\t" //     1

  228. 		PAVGB((%%eax), %%mm2)				      //    11	/2

  229. 		PAVGB((%0, %1, 2), %%mm2)			      //   211	/4

  230. 		"movq %%mm2, %%mm3				\n\t" //   211	/4

  231. 		"movq (%0), %%mm4				\n\t" // 1

  232. 		PAVGB(%%mm4, %%mm3)				      // 4 211	/8

  233. 		PAVGB(%%mm0, %%mm3)				      //642211	/16

  234. 		"movq %%mm3, (%0)				\n\t" // X

  235. 		// mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9

  236. 		"movq %%mm1, %%mm0				\n\t" //  1

  237. 		PAVGB(%%mm6, %%mm0)				      //1 1	/2

  238. 		"movq %%mm4, %%mm3				\n\t" // 1

  239. 		PAVGB((%0,%1,2), %%mm3)				      // 1 1	/2

  240. 		PAVGB((%%eax,%1,2), %%mm5)			      //     11	/2

  241. 		PAVGB((%%eax), %%mm5)				      //    211 /4

  242. 		PAVGB(%%mm5, %%mm3)				      // 2 2211 /8

  243. 		PAVGB(%%mm0, %%mm3)				      //4242211 /16

  244. 		"movq %%mm3, (%0,%1)				\n\t" //  X

  245. 		// mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9

  246. 		PAVGB(%%mm4, %%mm6)				      //11	/2

  247. 		"movq (%%ecx), %%mm0				\n\t" //       1

  248. 		PAVGB((%%eax, %1, 2), %%mm0)			      //      11/2

  249. 		"movq %%mm0, %%mm3				\n\t" //      11/2

  250. 		PAVGB(%%mm1, %%mm0)				      //  2   11/4

  251. 		PAVGB(%%mm6, %%mm0)				      //222   11/8

  252. 		PAVGB(%%mm2, %%mm0)				      //22242211/16

  253. 		"movq (%0, %1, 2), %%mm2			\n\t" //   1

  254. 		"movq %%mm0, (%0, %1, 2)			\n\t" //   X

  255. 		// mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9

  256. 		"movq (%%eax, %1, 4), %%mm0			\n\t" //        1

  257. 		PAVGB((%%ecx), %%mm0)				      //       11	/2

  258. 		PAVGB(%%mm0, %%mm6)				      //11     11	/4

  259. 		PAVGB(%%mm1, %%mm4)				      // 11		/2

  260. 		PAVGB(%%mm2, %%mm1)				      //  11		/2

  261. 		PAVGB(%%mm1, %%mm6)				      //1122   11	/8

  262. 		PAVGB(%%mm5, %%mm6)				      //112242211	/16

  263. 		"movq (%%eax), %%mm5				\n\t" //    1

  264. 		"movq %%mm6, (%%eax)				\n\t" //    X

  265. 		// mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9

  266. 		"movq (%%eax, %1, 4), %%mm6			\n\t" //        1

  267. 		PAVGB(%%mm7, %%mm6)				      //        11	/2

  268. 		PAVGB(%%mm4, %%mm6)				      // 11     11	/4

  269. 		PAVGB(%%mm3, %%mm6)				      // 11   2211	/8

  270. 		PAVGB(%%mm5, %%mm2)				      //   11		/2

  271. 		"movq (%0, %1, 4), %%mm4			\n\t" //     1

  272. 		PAVGB(%%mm4, %%mm2)				      //   112		/4

  273. 		PAVGB(%%mm2, %%mm6)				      // 112242211	/16

  274. 		"movq %%mm6, (%0, %1, 4)			\n\t" //     X

  275. 		// mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9

  276. 		PAVGB(%%mm7, %%mm1)				      //  11     2	/4

  277. 		PAVGB(%%mm4, %%mm5)				      //    11		/2

  278. 		PAVGB(%%mm5, %%mm0)				      //    11 11	/4

  279. 		"movq (%%eax, %1, 2), %%mm6			\n\t" //      1

  280. 		PAVGB(%%mm6, %%mm1)				      //  11  4  2	/8

  281. 		PAVGB(%%mm0, %%mm1)				      //  11224222	/16

  282. 		"movq %%mm1, (%%eax, %1, 2)			\n\t" //      X

  283. 		// mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9

  284. 		PAVGB((%%ecx), %%mm2)				      //   112 4	/8

  285. 		"movq (%%eax, %1, 4), %%mm0			\n\t" //        1

  286. 		PAVGB(%%mm0, %%mm6)				      //      1 1	/2

  287. 		PAVGB(%%mm7, %%mm6)				      //      1 12	/4

  288. 		PAVGB(%%mm2, %%mm6)				      //   1122424	/4

  289. 		"movq %%mm6, (%%ecx)				\n\t" //       X

  290. 		// mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9

  291. 		PAVGB(%%mm7, %%mm5)				      //    11   2	/4

  292. 		PAVGB(%%mm7, %%mm5)				      //    11   6	/8

  293. 

  294. 		PAVGB(%%mm3, %%mm0)				      //      112	/4

  295. 		PAVGB(%%mm0, %%mm5)				      //    112246	/16

  296. 		"movq %%mm5, (%%eax, %1, 4)			\n\t" //        X

  297. 		"subl %1, %0					\n\t"

  298. 

  299. 		:

  300. 		: "r" (src), "r" (stride), "m" (c->pQPb)

  301. 		: "%eax", "%ecx"

  302. 	);

  303. #else

  304. 	const int l1= stride;

  305. 	const int l2= stride + l1;

  306. 	const int l3= stride + l2;

  307. 	const int l4= stride + l3;

  308. 	const int l5= stride + l4;

  309. 	const int l6= stride + l5;

  310. 	const int l7= stride + l6;

  311. 	const int l8= stride + l7;

  312. 	const int l9= stride + l8;

  313. 	int x;

  314. 	src+= stride*3;

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

  316. 	{

  317. 		const int first= ABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];

  318. 		const int last= ABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];

  319. 

  320. 		int sums[10];

  321. 		sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;

  322. 		sums[1] = sums[0] - first  + src[l4];

  323. 		sums[2] = sums[1] - first  + src[l5];

  324. 		sums[3] = sums[2] - first  + src[l6];

  325. 		sums[4] = sums[3] - first  + src[l7];

  326. 		sums[5] = sums[4] - src[l1] + src[l8];

  327. 		sums[6] = sums[5] - src[l2] + last;

  328. 		sums[7] = sums[6] - src[l3] + last;

  329. 		sums[8] = sums[7] - src[l4] + last;

  330. 		sums[9] = sums[8] - src[l5] + last;

  331. 

  332. 		src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;

  333. 		src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;

  334. 		src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;

  335. 		src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;

  336. 		src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;

  337. 		src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;

  338. 		src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;

  339. 		src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;

  340. 

  341. 		src++;

  342. 	}

  343. #endif

  344. }

  345. #endif //HAVE_ALTIVEC

  346. 

  347. #if 0

  348. /**

  349.  * Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar

  350.  * values are correctly clipped (MMX2)

  351.  * values are wraparound (C)

  352.  * conclusion: its fast, but introduces ugly horizontal patterns if there is a continious gradient

  353. 	0 8 16 24

  354. 	x = 8

  355. 	x/2 = 4

  356. 	x/8 = 1

  357. 	1 12 12 23

  358.  */

  359. static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)

  360. {

  361. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  362. 	src+= stride*3;

  363. // FIXME rounding

  364. 	asm volatile(

  365. 		"pxor %%mm7, %%mm7				\n\t" // 0

  366. 		"movq "MANGLE(b80)", %%mm6			\n\t" // MIN_SIGNED_BYTE

  367. 		"leal (%0, %1), %%eax				\n\t"

  368. 		"leal (%%eax, %1, 4), %%ecx			\n\t"

  369. //	0	1	2	3	4	5	6	7	8	9

  370. //	%0	eax	eax+%1	eax+2%1	%0+4%1	ecx	ecx+%1	ecx+2%1	%0+8%1	ecx+4%1

  371. 		"movq "MANGLE(pQPb)", %%mm0			\n\t" // QP,..., QP

  372. 		"movq %%mm0, %%mm1				\n\t" // QP,..., QP

  373. 		"paddusb "MANGLE(b02)", %%mm0			\n\t"

  374. 		"psrlw $2, %%mm0				\n\t"

  375. 		"pand "MANGLE(b3F)", %%mm0			\n\t" // QP/4,..., QP/4

  376. 		"paddusb %%mm1, %%mm0				\n\t" // QP*1.25 ...

  377. 		"movq (%0, %1, 4), %%mm2			\n\t" // line 4

  378. 		"movq (%%ecx), %%mm3				\n\t" // line 5

  379. 		"movq %%mm2, %%mm4				\n\t" // line 4

  380. 		"pcmpeqb %%mm5, %%mm5				\n\t" // -1

  381. 		"pxor %%mm2, %%mm5				\n\t" // -line 4 - 1

  382. 		PAVGB(%%mm3, %%mm5)

  383. 		"paddb %%mm6, %%mm5				\n\t" // (l5-l4)/2

  384. 		"psubusb %%mm3, %%mm4				\n\t"

  385. 		"psubusb %%mm2, %%mm3				\n\t"

  386. 		"por %%mm3, %%mm4				\n\t" // |l4 - l5|

  387. 		"psubusb %%mm0, %%mm4				\n\t"

  388. 		"pcmpeqb %%mm7, %%mm4				\n\t"

  389. 		"pand %%mm4, %%mm5				\n\t" // d/2

  390. 

  391. //		"paddb %%mm6, %%mm2				\n\t" // line 4 + 0x80

  392. 		"paddb %%mm5, %%mm2				\n\t"

  393. //		"psubb %%mm6, %%mm2				\n\t"

  394. 		"movq %%mm2, (%0,%1, 4)				\n\t"

  395. 

  396. 		"movq (%%ecx), %%mm2				\n\t"

  397. //		"paddb %%mm6, %%mm2				\n\t" // line 5 + 0x80

  398. 		"psubb %%mm5, %%mm2				\n\t"

  399. //		"psubb %%mm6, %%mm2				\n\t"

  400. 		"movq %%mm2, (%%ecx)				\n\t"

  401. 

  402. 		"paddb %%mm6, %%mm5				\n\t"

  403. 		"psrlw $2, %%mm5				\n\t"

  404. 		"pand "MANGLE(b3F)", %%mm5			\n\t"

  405. 		"psubb "MANGLE(b20)", %%mm5			\n\t" // (l5-l4)/8

  406. 

  407. 		"movq (%%eax, %1, 2), %%mm2			\n\t"

  408. 		"paddb %%mm6, %%mm2				\n\t" // line 3 + 0x80

  409. 		"paddsb %%mm5, %%mm2				\n\t"

  410. 		"psubb %%mm6, %%mm2				\n\t"

  411. 		"movq %%mm2, (%%eax, %1, 2)			\n\t"

  412. 

  413. 		"movq (%%ecx, %1), %%mm2			\n\t"

  414. 		"paddb %%mm6, %%mm2				\n\t" // line 6 + 0x80

  415. 		"psubsb %%mm5, %%mm2				\n\t"

  416. 		"psubb %%mm6, %%mm2				\n\t"

  417. 		"movq %%mm2, (%%ecx, %1)			\n\t"

  418. 

  419. 		:

  420. 		: "r" (src), "r" (stride)

  421. 		: "%eax", "%ecx"

  422. 	);

  423. #else

  424.  	const int l1= stride;

  425. 	const int l2= stride + l1;

  426. 	const int l3= stride + l2;

  427. 	const int l4= stride + l3;

  428. 	const int l5= stride + l4;

  429. 	const int l6= stride + l5;

  430. //	const int l7= stride + l6;

  431. //	const int l8= stride + l7;

  432. //	const int l9= stride + l8;

  433. 	int x;

  434. 	const int QP15= QP + (QP>>2);

  435. 	src+= stride*3;

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

  437. 	{

  438. 		const int v = (src[x+l5] - src[x+l4]);

  439. 		if(ABS(v) < QP15)

  440. 		{

  441. 			src[x+l3] +=v>>3;

  442. 			src[x+l4] +=v>>1;

  443. 			src[x+l5] -=v>>1;

  444. 			src[x+l6] -=v>>3;

  445. 

  446. 		}

  447. 	}

  448. 

  449. #endif

  450. }

  451. #endif

  452. 

  453. /**

  454.  * Experimental Filter 1

  455.  * will not damage linear gradients

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

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

  458.  * MMX2 version does correct clipping C version doesnt

  459.  */

  460. static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)

  461. {

  462. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  463. 	src+= stride*3;

  464. 

  465. 	asm volatile(

  466. 		"pxor %%mm7, %%mm7				\n\t" // 0

  467. 		"leal (%0, %1), %%eax				\n\t"

  468. 		"leal (%%eax, %1, 4), %%ecx			\n\t"

  469. //	0	1	2	3	4	5	6	7	8	9

  470. //	%0	eax	eax+%1	eax+2%1	%0+4%1	ecx	ecx+%1	ecx+2%1	%0+8%1	ecx+4%1

  471. 		"movq (%%eax, %1, 2), %%mm0			\n\t" // line 3

  472. 		"movq (%0, %1, 4), %%mm1			\n\t" // line 4

  473. 		"movq %%mm1, %%mm2				\n\t" // line 4

  474. 		"psubusb %%mm0, %%mm1				\n\t"

  475. 		"psubusb %%mm2, %%mm0				\n\t"

  476. 		"por %%mm1, %%mm0				\n\t" // |l2 - l3|

  477. 		"movq (%%ecx), %%mm3				\n\t" // line 5

  478. 		"movq (%%ecx, %1), %%mm4			\n\t" // line 6

  479. 		"movq %%mm3, %%mm5				\n\t" // line 5

  480. 		"psubusb %%mm4, %%mm3				\n\t"

  481. 		"psubusb %%mm5, %%mm4				\n\t"

  482. 		"por %%mm4, %%mm3				\n\t" // |l5 - l6|

  483. 		PAVGB(%%mm3, %%mm0)				      // (|l2 - l3| + |l5 - l6|)/2

  484. 		"movq %%mm2, %%mm1				\n\t" // line 4

  485. 		"psubusb %%mm5, %%mm2				\n\t"

  486. 		"movq %%mm2, %%mm4				\n\t"

  487. 		"pcmpeqb %%mm7, %%mm2				\n\t" // (l4 - l5) <= 0 ? -1 : 0

  488. 		"psubusb %%mm1, %%mm5				\n\t"

  489. 		"por %%mm5, %%mm4				\n\t" // |l4 - l5|

  490. 		"psubusb %%mm0, %%mm4		\n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)

  491. 		"movq %%mm4, %%mm3				\n\t" // d

  492. 		"movq %2, %%mm0			\n\t"

  493.                 "paddusb %%mm0, %%mm0				\n\t"

  494. 		"psubusb %%mm0, %%mm4				\n\t"

  495. 		"pcmpeqb %%mm7, %%mm4				\n\t" // d <= QP ? -1 : 0

  496. 		"psubusb "MANGLE(b01)", %%mm3			\n\t"

  497. 		"pand %%mm4, %%mm3				\n\t" // d <= QP ? d : 0

  498. 

  499. 		PAVGB(%%mm7, %%mm3)				      // d/2

  500. 		"movq %%mm3, %%mm1				\n\t" // d/2

  501. 		PAVGB(%%mm7, %%mm3)				      // d/4

  502. 		PAVGB(%%mm1, %%mm3)				      // 3*d/8

  503. 

  504. 		"movq (%0, %1, 4), %%mm0			\n\t" // line 4

  505. 		"pxor %%mm2, %%mm0				\n\t" //(l4 - l5) <= 0 ? -l4-1 : l4

  506. 		"psubusb %%mm3, %%mm0				\n\t"

  507. 		"pxor %%mm2, %%mm0				\n\t"

  508. 		"movq %%mm0, (%0, %1, 4)			\n\t" // line 4

  509. 

  510. 		"movq (%%ecx), %%mm0				\n\t" // line 5

  511. 		"pxor %%mm2, %%mm0				\n\t" //(l4 - l5) <= 0 ? -l5-1 : l5

  512. 		"paddusb %%mm3, %%mm0				\n\t"

  513. 		"pxor %%mm2, %%mm0				\n\t"

  514. 		"movq %%mm0, (%%ecx)				\n\t" // line 5

  515. 

  516. 		PAVGB(%%mm7, %%mm1)				      // d/4

  517. 

  518. 		"movq (%%eax, %1, 2), %%mm0			\n\t" // line 3

  519. 		"pxor %%mm2, %%mm0				\n\t" //(l4 - l5) <= 0 ? -l4-1 : l4

  520. 		"psubusb %%mm1, %%mm0				\n\t"

  521. 		"pxor %%mm2, %%mm0				\n\t"

  522. 		"movq %%mm0, (%%eax, %1, 2)			\n\t" // line 3

  523. 

  524. 		"movq (%%ecx, %1), %%mm0			\n\t" // line 6

  525. 		"pxor %%mm2, %%mm0				\n\t" //(l4 - l5) <= 0 ? -l5-1 : l5

  526. 		"paddusb %%mm1, %%mm0				\n\t"

  527. 		"pxor %%mm2, %%mm0				\n\t"

  528. 		"movq %%mm0, (%%ecx, %1)			\n\t" // line 6

  529. 

  530. 		PAVGB(%%mm7, %%mm1)				      // d/8

  531. 

  532. 		"movq (%%eax, %1), %%mm0			\n\t" // line 2

  533. 		"pxor %%mm2, %%mm0				\n\t" //(l4 - l5) <= 0 ? -l2-1 : l2

  534. 		"psubusb %%mm1, %%mm0				\n\t"

  535. 		"pxor %%mm2, %%mm0				\n\t"

  536. 		"movq %%mm0, (%%eax, %1)			\n\t" // line 2

  537. 

  538. 		"movq (%%ecx, %1, 2), %%mm0			\n\t" // line 7

  539. 		"pxor %%mm2, %%mm0				\n\t" //(l4 - l5) <= 0 ? -l7-1 : l7

  540. 		"paddusb %%mm1, %%mm0				\n\t"

  541. 		"pxor %%mm2, %%mm0				\n\t"

  542. 		"movq %%mm0, (%%ecx, %1, 2)			\n\t" // line 7

  543. 

  544. 		:

  545. 		: "r" (src), "r" (stride), "m" (co->pQPb)

  546. 		: "%eax", "%ecx"

  547. 	);

  548. #else

  549. 

  550.  	const int l1= stride;

  551. 	const int l2= stride + l1;

  552. 	const int l3= stride + l2;

  553. 	const int l4= stride + l3;

  554. 	const int l5= stride + l4;

  555. 	const int l6= stride + l5;

  556. 	const int l7= stride + l6;

  557. //	const int l8= stride + l7;

  558. //	const int l9= stride + l8;

  559. 	int x;

  560. 

  561. 	src+= stride*3;

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

  563. 	{

  564. 		int a= src[l3] - src[l4];

  565. 		int b= src[l4] - src[l5];

  566. 		int c= src[l5] - src[l6];

  567. 

  568. 		int d= ABS(b) - ((ABS(a) + ABS(c))>>1);

  569. 		d= MAX(d, 0);

  570. 

  571. 		if(d < co->QP*2)

  572. 		{

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

  574. 

  575. 			src[l2] +=v>>3;

  576. 			src[l3] +=v>>2;

  577. 			src[l4] +=(3*v)>>3;

  578. 			src[l5] -=(3*v)>>3;

  579. 			src[l6] -=v>>2;

  580. 			src[l7] -=v>>3;

  581. 

  582. 		}

  583. 		src++;

  584. 	}

  585. #endif

  586. }

  587. 

  588. #ifndef HAVE_ALTIVEC

  589. static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)

  590. {

  591. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  592. /*

  593. 	uint8_t tmp[16];

  594. 	const int l1= stride;

  595. 	const int l2= stride + l1;

  596. 	const int l3= stride + l2;

  597. 	const int l4= (int)tmp - (int)src - stride*3;

  598. 	const int l5= (int)tmp - (int)src - stride*3 + 8;

  599. 	const int l6= stride*3 + l3;

  600. 	const int l7= stride + l6;

  601. 	const int l8= stride + l7;

  602. 

  603. 	memcpy(tmp, src+stride*7, 8);

  604. 	memcpy(tmp+8, src+stride*8, 8);

  605. */

  606. 	src+= stride*4;

  607. 	asm volatile(

  608. 

  609. #if 0 //sligtly more accurate and slightly slower

  610. 		"pxor %%mm7, %%mm7				\n\t" // 0

  611. 		"leal (%0, %1), %%eax				\n\t"

  612. 		"leal (%%eax, %1, 4), %%ecx			\n\t"

  613. //	0	1	2	3	4	5	6	7

  614. //	%0	%0+%1	%0+2%1	eax+2%1	%0+4%1	eax+4%1	ecx+%1	ecx+2%1

  615. //	%0	eax	eax+%1	eax+2%1	%0+4%1	ecx	ecx+%1	ecx+2%1

  616. 

  617. 

  618. 		"movq (%0, %1, 2), %%mm0			\n\t" // l2

  619. 		"movq (%0), %%mm1				\n\t" // l0

  620. 		"movq %%mm0, %%mm2				\n\t" // l2

  621. 		PAVGB(%%mm7, %%mm0)				      // ~l2/2

  622. 		PAVGB(%%mm1, %%mm0)				      // ~(l2 + 2l0)/4

  623. 		PAVGB(%%mm2, %%mm0)				      // ~(5l2 + 2l0)/8

  624. 

  625. 		"movq (%%eax), %%mm1				\n\t" // l1

  626. 		"movq (%%eax, %1, 2), %%mm3			\n\t" // l3

  627. 		"movq %%mm1, %%mm4				\n\t" // l1

  628. 		PAVGB(%%mm7, %%mm1)				      // ~l1/2

  629. 		PAVGB(%%mm3, %%mm1)				      // ~(l1 + 2l3)/4

  630. 		PAVGB(%%mm4, %%mm1)				      // ~(5l1 + 2l3)/8

  631. 

  632. 		"movq %%mm0, %%mm4				\n\t" // ~(5l2 + 2l0)/8

  633. 		"psubusb %%mm1, %%mm0				\n\t"

  634. 		"psubusb %%mm4, %%mm1				\n\t"

  635. 		"por %%mm0, %%mm1				\n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8

  636. // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0

  637. 

  638. 		"movq (%0, %1, 4), %%mm0			\n\t" // l4

  639. 		"movq %%mm0, %%mm4				\n\t" // l4

  640. 		PAVGB(%%mm7, %%mm0)				      // ~l4/2

  641. 		PAVGB(%%mm2, %%mm0)				      // ~(l4 + 2l2)/4

  642. 		PAVGB(%%mm4, %%mm0)				      // ~(5l4 + 2l2)/8

  643. 

  644. 		"movq (%%ecx), %%mm2				\n\t" // l5

  645. 		"movq %%mm3, %%mm5				\n\t" // l3

  646. 		PAVGB(%%mm7, %%mm3)				      // ~l3/2

  647. 		PAVGB(%%mm2, %%mm3)				      // ~(l3 + 2l5)/4

  648. 		PAVGB(%%mm5, %%mm3)				      // ~(5l3 + 2l5)/8

  649. 

  650. 		"movq %%mm0, %%mm6				\n\t" // ~(5l4 + 2l2)/8

  651. 		"psubusb %%mm3, %%mm0				\n\t"

  652. 		"psubusb %%mm6, %%mm3				\n\t"

  653. 		"por %%mm0, %%mm3				\n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8

  654. 		"pcmpeqb %%mm7, %%mm0				\n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)

  655. // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0

  656. 

  657. 		"movq (%%ecx, %1), %%mm6			\n\t" // l6

  658. 		"movq %%mm6, %%mm5				\n\t" // l6

  659. 		PAVGB(%%mm7, %%mm6)				      // ~l6/2

  660. 		PAVGB(%%mm4, %%mm6)				      // ~(l6 + 2l4)/4

  661. 		PAVGB(%%mm5, %%mm6)				      // ~(5l6 + 2l4)/8

  662. 

  663. 		"movq (%%ecx, %1, 2), %%mm5			\n\t" // l7

  664. 		"movq %%mm2, %%mm4				\n\t" // l5

  665. 		PAVGB(%%mm7, %%mm2)				      // ~l5/2

  666. 		PAVGB(%%mm5, %%mm2)				      // ~(l5 + 2l7)/4

  667. 		PAVGB(%%mm4, %%mm2)				      // ~(5l5 + 2l7)/8

  668. 

  669. 		"movq %%mm6, %%mm4				\n\t" // ~(5l6 + 2l4)/8

  670. 		"psubusb %%mm2, %%mm6				\n\t"

  671. 		"psubusb %%mm4, %%mm2				\n\t"

  672. 		"por %%mm6, %%mm2				\n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8

  673. // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0

  674. 

  675. 

  676. 		PMINUB(%%mm2, %%mm1, %%mm4)			      // MIN(|lenergy|,|renergy|)/8

  677. 		"movq %2, %%mm4					\n\t" // QP //FIXME QP+1 ?

  678. 		"paddusb "MANGLE(b01)", %%mm4			\n\t"

  679. 		"pcmpgtb %%mm3, %%mm4				\n\t" // |menergy|/8 < QP

  680. 		"psubusb %%mm1, %%mm3				\n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8

  681. 		"pand %%mm4, %%mm3				\n\t"

  682. 

  683. 		"movq %%mm3, %%mm1				\n\t"

  684. //		"psubusb "MANGLE(b01)", %%mm3			\n\t"

  685. 		PAVGB(%%mm7, %%mm3)

  686. 		PAVGB(%%mm7, %%mm3)

  687. 		"paddusb %%mm1, %%mm3				\n\t"

  688. //		"paddusb "MANGLE(b01)", %%mm3			\n\t"

  689. 

  690. 		"movq (%%eax, %1, 2), %%mm6			\n\t" //l3

  691. 		"movq (%0, %1, 4), %%mm5			\n\t" //l4

  692. 		"movq (%0, %1, 4), %%mm4			\n\t" //l4

  693. 		"psubusb %%mm6, %%mm5				\n\t"

  694. 		"psubusb %%mm4, %%mm6				\n\t"

  695. 		"por %%mm6, %%mm5				\n\t" // |l3-l4|

  696. 		"pcmpeqb %%mm7, %%mm6				\n\t" // SIGN(l3-l4)

  697. 		"pxor %%mm6, %%mm0				\n\t"

  698. 		"pand %%mm0, %%mm3				\n\t"

  699. 		PMINUB(%%mm5, %%mm3, %%mm0)

  700. 

  701. 		"psubusb "MANGLE(b01)", %%mm3			\n\t"

  702. 		PAVGB(%%mm7, %%mm3)

  703. 

  704. 		"movq (%%eax, %1, 2), %%mm0			\n\t"

  705. 		"movq (%0, %1, 4), %%mm2			\n\t"

  706. 		"pxor %%mm6, %%mm0				\n\t"

  707. 		"pxor %%mm6, %%mm2				\n\t"

  708. 		"psubb %%mm3, %%mm0				\n\t"

  709. 		"paddb %%mm3, %%mm2				\n\t"

  710. 		"pxor %%mm6, %%mm0				\n\t"

  711. 		"pxor %%mm6, %%mm2				\n\t"

  712. 		"movq %%mm0, (%%eax, %1, 2)			\n\t"

  713. 		"movq %%mm2, (%0, %1, 4)			\n\t"

  714. #endif

  715. 

  716. 		"leal (%0, %1), %%eax				\n\t"

  717. 		"pcmpeqb %%mm6, %%mm6				\n\t" // -1

  718. //	0	1	2	3	4	5	6	7

  719. //	%0	%0+%1	%0+2%1	eax+2%1	%0+4%1	eax+4%1	ecx+%1	ecx+2%1

  720. //	%0	eax	eax+%1	eax+2%1	%0+4%1	ecx	ecx+%1	ecx+2%1

  721. 

  722. 

  723. 		"movq (%%eax, %1, 2), %%mm1			\n\t" // l3

  724. 		"movq (%0, %1, 4), %%mm0			\n\t" // l4

  725. 		"pxor %%mm6, %%mm1				\n\t" // -l3-1

  726. 		PAVGB(%%mm1, %%mm0)				      // -q+128 = (l4-l3+256)/2

  727. // mm1=-l3-1, mm0=128-q

  728. 

  729. 		"movq (%%eax, %1, 4), %%mm2			\n\t" // l5

  730. 		"movq (%%eax, %1), %%mm3			\n\t" // l2

  731. 		"pxor %%mm6, %%mm2				\n\t" // -l5-1

  732. 		"movq %%mm2, %%mm5				\n\t" // -l5-1

  733. 		"movq "MANGLE(b80)", %%mm4			\n\t" // 128

  734. 		"leal (%%eax, %1, 4), %%ecx			\n\t"

  735. 		PAVGB(%%mm3, %%mm2)				      // (l2-l5+256)/2

  736. 		PAVGB(%%mm0, %%mm4)				      // ~(l4-l3)/4 + 128

  737. 		PAVGB(%%mm2, %%mm4)				      // ~(l2-l5)/4 +(l4-l3)/8 + 128

  738. 		PAVGB(%%mm0, %%mm4)				      // ~(l2-l5)/8 +5(l4-l3)/16 + 128

  739. // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1

  740. 

  741. 		"movq (%%eax), %%mm2				\n\t" // l1

  742. 		"pxor %%mm6, %%mm2				\n\t" // -l1-1

  743. 		PAVGB(%%mm3, %%mm2)				      // (l2-l1+256)/2

  744. 		PAVGB((%0), %%mm1)				      // (l0-l3+256)/2

  745. 		"movq "MANGLE(b80)", %%mm3			\n\t" // 128

  746. 		PAVGB(%%mm2, %%mm3)				      // ~(l2-l1)/4 + 128

  747. 		PAVGB(%%mm1, %%mm3)				      // ~(l0-l3)/4 +(l2-l1)/8 + 128

  748. 		PAVGB(%%mm2, %%mm3)				      // ~(l0-l3)/8 +5(l2-l1)/16 + 128

  749. // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1

  750. 

  751. 		PAVGB((%%ecx, %1), %%mm5)			      // (l6-l5+256)/2

  752. 		"movq (%%ecx, %1, 2), %%mm1			\n\t" // l7

  753. 		"pxor %%mm6, %%mm1				\n\t" // -l7-1

  754. 		PAVGB((%0, %1, 4), %%mm1)			      // (l4-l7+256)/2

  755. 		"movq "MANGLE(b80)", %%mm2			\n\t" // 128

  756. 		PAVGB(%%mm5, %%mm2)				      // ~(l6-l5)/4 + 128

  757. 		PAVGB(%%mm1, %%mm2)				      // ~(l4-l7)/4 +(l6-l5)/8 + 128

  758. 		PAVGB(%%mm5, %%mm2)				      // ~(l4-l7)/8 +5(l6-l5)/16 + 128

  759. // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128

  760. 

  761. 		"movq "MANGLE(b00)", %%mm1			\n\t" // 0

  762. 		"movq "MANGLE(b00)", %%mm5			\n\t" // 0

  763. 		"psubb %%mm2, %%mm1				\n\t" // 128 - renergy/16

  764. 		"psubb %%mm3, %%mm5				\n\t" // 128 - lenergy/16

  765. 		PMAXUB(%%mm1, %%mm2)				      // 128 + |renergy/16|

  766.  		PMAXUB(%%mm5, %%mm3)				      // 128 + |lenergy/16|

  767. 		PMINUB(%%mm2, %%mm3, %%mm1)			      // 128 + MIN(|lenergy|,|renergy|)/16

  768. 

  769. // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128

  770. 

  771. 		"movq "MANGLE(b00)", %%mm7			\n\t" // 0

  772. 		"movq %2, %%mm2					\n\t" // QP

  773. 		PAVGB(%%mm6, %%mm2)				      // 128 + QP/2

  774. 		"psubb %%mm6, %%mm2				\n\t"

  775. 

  776. 		"movq %%mm4, %%mm1				\n\t"

  777. 		"pcmpgtb %%mm7, %%mm1				\n\t" // SIGN(menergy)

  778. 		"pxor %%mm1, %%mm4				\n\t"

  779. 		"psubb %%mm1, %%mm4				\n\t" // 128 + |menergy|/16

  780. 		"pcmpgtb %%mm4, %%mm2				\n\t" // |menergy|/16 < QP/2

  781. 		"psubusb %%mm3, %%mm4				\n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16

  782. // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16

  783. 

  784. 		"movq %%mm4, %%mm3				\n\t" // d

  785. 		"psubusb "MANGLE(b01)", %%mm4			\n\t"

  786. 		PAVGB(%%mm7, %%mm4)				      // d/32

  787. 		PAVGB(%%mm7, %%mm4)				      // (d + 32)/64

  788. 		"paddb %%mm3, %%mm4				\n\t" // 5d/64

  789. 		"pand %%mm2, %%mm4				\n\t"

  790. 

  791. 		"movq "MANGLE(b80)", %%mm5			\n\t" // 128

  792. 		"psubb %%mm0, %%mm5				\n\t" // q

  793. 		"paddsb %%mm6, %%mm5				\n\t" // fix bad rounding

  794. 		"pcmpgtb %%mm5, %%mm7				\n\t" // SIGN(q)

  795. 		"pxor %%mm7, %%mm5				\n\t"

  796. 

  797. 		PMINUB(%%mm5, %%mm4, %%mm3)			      // MIN(|q|, 5d/64)

  798. 		"pxor %%mm1, %%mm7				\n\t" // SIGN(d*q)

  799. 

  800. 		"pand %%mm7, %%mm4				\n\t"

  801. 		"movq (%%eax, %1, 2), %%mm0			\n\t"

  802. 		"movq (%0, %1, 4), %%mm2			\n\t"

  803. 		"pxor %%mm1, %%mm0				\n\t"

  804. 		"pxor %%mm1, %%mm2				\n\t"

  805. 		"paddb %%mm4, %%mm0				\n\t"

  806. 		"psubb %%mm4, %%mm2				\n\t"

  807. 		"pxor %%mm1, %%mm0				\n\t"

  808. 		"pxor %%mm1, %%mm2				\n\t"

  809. 		"movq %%mm0, (%%eax, %1, 2)			\n\t"

  810. 		"movq %%mm2, (%0, %1, 4)			\n\t"

  811. 

  812. 		:

  813. 		: "r" (src), "r" (stride), "m" (c->pQPb)

  814. 		: "%eax", "%ecx"

  815. 	);

  816. 

  817. /*

  818. 	{

  819. 	int x;

  820. 	src-= stride;

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

  822. 	{

  823. 		const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);

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

  825. 		{

  826. 			const int q=(src[l4] - src[l5])/2;

  827. 			const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);

  828. 			const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

  829. 

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

  831. 			d= MAX(d, 0);

  832. 

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

  834. 			d*= SIGN(-middleEnergy);

  835. 

  836. 			if(q>0)

  837. 			{

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

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

  840. 			}

  841. 			else

  842. 			{

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

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

  845. 			}

  846. 

  847.         		src[l4]-= d;

  848. 	        	src[l5]+= d;

  849. 		}

  850. 		src++;

  851. 	}

  852. src-=8;

  853. 	for(x=0; x<8; x++)

  854. 	{

  855. 		int y;

  856. 		for(y=4; y<6; y++)

  857. 		{

  858. 			int d= src[x+y*stride] - tmp[x+(y-4)*8];

  859. 			int ad= ABS(d);

  860. 			static int max=0;

  861. 			static int sum=0;

  862. 			static int num=0;

  863. 			static int bias=0;

  864. 

  865. 			if(max<ad) max=ad;

  866. 			sum+= ad>3 ? 1 : 0;

  867. 			if(ad>3)

  868. 			{

  869. 				src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;

  870. 			}

  871. 			if(y==4) bias+=d;

  872. 			num++;

  873. 			if(num%1000000 == 0)

  874. 			{

  875. 				printf(" %d %d %d %d\n", num, sum, max, bias);

  876. 			}

  877. 		}

  878. 	}

  879. }

  880. */

  881. #elif defined (HAVE_MMX)

  882. 	src+= stride*4;

  883. 	asm volatile(

  884. 		"pxor %%mm7, %%mm7				\n\t"

  885. 		"leal -40(%%esp), %%ecx				\n\t" // make space for 4 8-byte vars

  886. 		"andl $0xFFFFFFF8, %%ecx			\n\t" // align

  887. //	0	1	2	3	4	5	6	7

  888. //	%0	%0+%1	%0+2%1	eax+2%1	%0+4%1	eax+4%1	edx+%1	edx+2%1

  889. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1

  890. 

  891. 		"movq (%0), %%mm0				\n\t"

  892. 		"movq %%mm0, %%mm1				\n\t"

  893. 		"punpcklbw %%mm7, %%mm0				\n\t" // low part of line 0

  894. 		"punpckhbw %%mm7, %%mm1				\n\t" // high part of line 0

  895. 

  896. 		"movq (%0, %1), %%mm2				\n\t"

  897. 		"leal (%0, %1, 2), %%eax			\n\t"

  898. 		"movq %%mm2, %%mm3				\n\t"

  899. 		"punpcklbw %%mm7, %%mm2				\n\t" // low part of line 1

  900. 		"punpckhbw %%mm7, %%mm3				\n\t" // high part of line 1

  901. 

  902. 		"movq (%%eax), %%mm4				\n\t"

  903. 		"movq %%mm4, %%mm5				\n\t"

  904. 		"punpcklbw %%mm7, %%mm4				\n\t" // low part of line 2

  905. 		"punpckhbw %%mm7, %%mm5				\n\t" // high part of line 2

  906. 

  907. 		"paddw %%mm0, %%mm0				\n\t" // 2L0

  908. 		"paddw %%mm1, %%mm1				\n\t" // 2H0

  909. 		"psubw %%mm4, %%mm2				\n\t" // L1 - L2

  910. 		"psubw %%mm5, %%mm3				\n\t" // H1 - H2

  911. 		"psubw %%mm2, %%mm0				\n\t" // 2L0 - L1 + L2

  912. 		"psubw %%mm3, %%mm1				\n\t" // 2H0 - H1 + H2

  913. 

  914. 		"psllw $2, %%mm2				\n\t" // 4L1 - 4L2

  915. 		"psllw $2, %%mm3				\n\t" // 4H1 - 4H2

  916. 		"psubw %%mm2, %%mm0				\n\t" // 2L0 - 5L1 + 5L2

  917. 		"psubw %%mm3, %%mm1				\n\t" // 2H0 - 5H1 + 5H2

  918. 

  919. 		"movq (%%eax, %1), %%mm2			\n\t"

  920. 		"movq %%mm2, %%mm3				\n\t"

  921. 		"punpcklbw %%mm7, %%mm2				\n\t" // L3

  922. 		"punpckhbw %%mm7, %%mm3				\n\t" // H3

  923. 

  924. 		"psubw %%mm2, %%mm0				\n\t" // 2L0 - 5L1 + 5L2 - L3

  925. 		"psubw %%mm3, %%mm1				\n\t" // 2H0 - 5H1 + 5H2 - H3

  926. 		"psubw %%mm2, %%mm0				\n\t" // 2L0 - 5L1 + 5L2 - 2L3

  927. 		"psubw %%mm3, %%mm1				\n\t" // 2H0 - 5H1 + 5H2 - 2H3

  928. 		"movq %%mm0, (%%ecx)				\n\t" // 2L0 - 5L1 + 5L2 - 2L3

  929. 		"movq %%mm1, 8(%%ecx)				\n\t" // 2H0 - 5H1 + 5H2 - 2H3

  930. 

  931. 		"movq (%%eax, %1, 2), %%mm0			\n\t"

  932. 		"movq %%mm0, %%mm1				\n\t"

  933. 		"punpcklbw %%mm7, %%mm0				\n\t" // L4

  934. 		"punpckhbw %%mm7, %%mm1				\n\t" // H4

  935. 

  936. 		"psubw %%mm0, %%mm2				\n\t" // L3 - L4

  937. 		"psubw %%mm1, %%mm3				\n\t" // H3 - H4

  938. 		"movq %%mm2, 16(%%ecx)				\n\t" // L3 - L4

  939. 		"movq %%mm3, 24(%%ecx)				\n\t" // H3 - H4

  940. 		"paddw %%mm4, %%mm4				\n\t" // 2L2

  941. 		"paddw %%mm5, %%mm5				\n\t" // 2H2

  942. 		"psubw %%mm2, %%mm4				\n\t" // 2L2 - L3 + L4

  943. 		"psubw %%mm3, %%mm5				\n\t" // 2H2 - H3 + H4

  944. 

  945. 		"leal (%%eax, %1), %0				\n\t"

  946. 		"psllw $2, %%mm2				\n\t" // 4L3 - 4L4

  947. 		"psllw $2, %%mm3				\n\t" // 4H3 - 4H4

  948. 		"psubw %%mm2, %%mm4				\n\t" // 2L2 - 5L3 + 5L4

  949. 		"psubw %%mm3, %%mm5				\n\t" // 2H2 - 5H3 + 5H4

  950. //50 opcodes so far

  951. 		"movq (%0, %1, 2), %%mm2			\n\t"

  952. 		"movq %%mm2, %%mm3				\n\t"

  953. 		"punpcklbw %%mm7, %%mm2				\n\t" // L5

  954. 		"punpckhbw %%mm7, %%mm3				\n\t" // H5

  955. 		"psubw %%mm2, %%mm4				\n\t" // 2L2 - 5L3 + 5L4 - L5

  956. 		"psubw %%mm3, %%mm5				\n\t" // 2H2 - 5H3 + 5H4 - H5

  957. 		"psubw %%mm2, %%mm4				\n\t" // 2L2 - 5L3 + 5L4 - 2L5

  958. 		"psubw %%mm3, %%mm5				\n\t" // 2H2 - 5H3 + 5H4 - 2H5

  959. 

  960. 		"movq (%%eax, %1, 4), %%mm6			\n\t"

  961. 		"punpcklbw %%mm7, %%mm6				\n\t" // L6

  962. 		"psubw %%mm6, %%mm2				\n\t" // L5 - L6

  963. 		"movq (%%eax, %1, 4), %%mm6			\n\t"

  964. 		"punpckhbw %%mm7, %%mm6				\n\t" // H6

  965. 		"psubw %%mm6, %%mm3				\n\t" // H5 - H6

  966. 

  967. 		"paddw %%mm0, %%mm0				\n\t" // 2L4

  968. 		"paddw %%mm1, %%mm1				\n\t" // 2H4

  969. 		"psubw %%mm2, %%mm0				\n\t" // 2L4 - L5 + L6

  970. 		"psubw %%mm3, %%mm1				\n\t" // 2H4 - H5 + H6

  971. 

  972. 		"psllw $2, %%mm2				\n\t" // 4L5 - 4L6

  973. 		"psllw $2, %%mm3				\n\t" // 4H5 - 4H6

  974. 		"psubw %%mm2, %%mm0				\n\t" // 2L4 - 5L5 + 5L6

  975. 		"psubw %%mm3, %%mm1				\n\t" // 2H4 - 5H5 + 5H6

  976. 

  977. 		"movq (%0, %1, 4), %%mm2			\n\t"

  978. 		"movq %%mm2, %%mm3				\n\t"

  979. 		"punpcklbw %%mm7, %%mm2				\n\t" // L7

  980. 		"punpckhbw %%mm7, %%mm3				\n\t" // H7

  981. 

  982. 		"paddw %%mm2, %%mm2				\n\t" // 2L7

  983. 		"paddw %%mm3, %%mm3				\n\t" // 2H7

  984. 		"psubw %%mm2, %%mm0				\n\t" // 2L4 - 5L5 + 5L6 - 2L7

  985. 		"psubw %%mm3, %%mm1				\n\t" // 2H4 - 5H5 + 5H6 - 2H7

  986. 

  987. 		"movq (%%ecx), %%mm2				\n\t" // 2L0 - 5L1 + 5L2 - 2L3

  988. 		"movq 8(%%ecx), %%mm3				\n\t" // 2H0 - 5H1 + 5H2 - 2H3

  989. 

  990. #ifdef HAVE_MMX2

  991. 		"movq %%mm7, %%mm6				\n\t" // 0

  992. 		"psubw %%mm0, %%mm6				\n\t"

  993. 		"pmaxsw %%mm6, %%mm0				\n\t" // |2L4 - 5L5 + 5L6 - 2L7|

  994. 		"movq %%mm7, %%mm6				\n\t" // 0

  995. 		"psubw %%mm1, %%mm6				\n\t"

  996. 		"pmaxsw %%mm6, %%mm1				\n\t" // |2H4 - 5H5 + 5H6 - 2H7|

  997. 		"movq %%mm7, %%mm6				\n\t" // 0

  998. 		"psubw %%mm2, %%mm6				\n\t"

  999. 		"pmaxsw %%mm6, %%mm2				\n\t" // |2L0 - 5L1 + 5L2 - 2L3|

  1000. 		"movq %%mm7, %%mm6				\n\t" // 0

  1001. 		"psubw %%mm3, %%mm6				\n\t"

  1002. 		"pmaxsw %%mm6, %%mm3				\n\t" // |2H0 - 5H1 + 5H2 - 2H3|

  1003. #else

  1004. 		"movq %%mm7, %%mm6				\n\t" // 0

  1005. 		"pcmpgtw %%mm0, %%mm6				\n\t"

  1006. 		"pxor %%mm6, %%mm0				\n\t"

  1007. 		"psubw %%mm6, %%mm0				\n\t" // |2L4 - 5L5 + 5L6 - 2L7|

  1008. 		"movq %%mm7, %%mm6				\n\t" // 0

  1009. 		"pcmpgtw %%mm1, %%mm6				\n\t"

  1010. 		"pxor %%mm6, %%mm1				\n\t"

  1011. 		"psubw %%mm6, %%mm1				\n\t" // |2H4 - 5H5 + 5H6 - 2H7|

  1012. 		"movq %%mm7, %%mm6				\n\t" // 0

  1013. 		"pcmpgtw %%mm2, %%mm6				\n\t"

  1014. 		"pxor %%mm6, %%mm2				\n\t"

  1015. 		"psubw %%mm6, %%mm2				\n\t" // |2L0 - 5L1 + 5L2 - 2L3|

  1016. 		"movq %%mm7, %%mm6				\n\t" // 0

  1017. 		"pcmpgtw %%mm3, %%mm6				\n\t"

  1018. 		"pxor %%mm6, %%mm3				\n\t"

  1019. 		"psubw %%mm6, %%mm3				\n\t" // |2H0 - 5H1 + 5H2 - 2H3|

  1020. #endif

  1021. 

  1022. #ifdef HAVE_MMX2

  1023. 		"pminsw %%mm2, %%mm0				\n\t"

  1024. 		"pminsw %%mm3, %%mm1				\n\t"

  1025. #else

  1026. 		"movq %%mm0, %%mm6				\n\t"

  1027. 		"psubusw %%mm2, %%mm6				\n\t"

  1028. 		"psubw %%mm6, %%mm0				\n\t"

  1029. 		"movq %%mm1, %%mm6				\n\t"

  1030. 		"psubusw %%mm3, %%mm6				\n\t"

  1031. 		"psubw %%mm6, %%mm1				\n\t"

  1032. #endif

  1033. 

  1034. 		"movq %%mm7, %%mm6				\n\t" // 0

  1035. 		"pcmpgtw %%mm4, %%mm6				\n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)

  1036. 		"pxor %%mm6, %%mm4				\n\t"

  1037. 		"psubw %%mm6, %%mm4				\n\t" // |2L2 - 5L3 + 5L4 - 2L5|

  1038. 		"pcmpgtw %%mm5, %%mm7				\n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)

  1039. 		"pxor %%mm7, %%mm5				\n\t"

  1040. 		"psubw %%mm7, %%mm5				\n\t" // |2H2 - 5H3 + 5H4 - 2H5|

  1041. // 100 opcodes

  1042. 		"movd %2, %%mm2					\n\t" // QP

  1043. 		"psllw $3, %%mm2				\n\t" // 8QP

  1044. 		"movq %%mm2, %%mm3				\n\t" // 8QP

  1045. 		"pcmpgtw %%mm4, %%mm2				\n\t"

  1046. 		"pcmpgtw %%mm5, %%mm3				\n\t"

  1047. 		"pand %%mm2, %%mm4				\n\t"

  1048. 		"pand %%mm3, %%mm5				\n\t"

  1049. 

  1050. 

  1051. 		"psubusw %%mm0, %%mm4				\n\t" // hd

  1052. 		"psubusw %%mm1, %%mm5				\n\t" // ld

  1053. 

  1054. 

  1055. 		"movq "MANGLE(w05)", %%mm2			\n\t" // 5

  1056. 		"pmullw %%mm2, %%mm4				\n\t"

  1057. 		"pmullw %%mm2, %%mm5				\n\t"

  1058. 		"movq "MANGLE(w20)", %%mm2			\n\t" // 32

  1059. 		"paddw %%mm2, %%mm4				\n\t"

  1060. 		"paddw %%mm2, %%mm5				\n\t"

  1061. 		"psrlw $6, %%mm4				\n\t"

  1062. 		"psrlw $6, %%mm5				\n\t"

  1063. 

  1064. 		"movq 16(%%ecx), %%mm0				\n\t" // L3 - L4

  1065. 		"movq 24(%%ecx), %%mm1				\n\t" // H3 - H4

  1066. 

  1067. 		"pxor %%mm2, %%mm2				\n\t"

  1068. 		"pxor %%mm3, %%mm3				\n\t"

  1069. 

  1070. 		"pcmpgtw %%mm0, %%mm2				\n\t" // sign (L3-L4)

  1071. 		"pcmpgtw %%mm1, %%mm3				\n\t" // sign (H3-H4)

  1072. 		"pxor %%mm2, %%mm0				\n\t"

  1073. 		"pxor %%mm3, %%mm1				\n\t"

  1074. 		"psubw %%mm2, %%mm0				\n\t" // |L3-L4|

  1075. 		"psubw %%mm3, %%mm1				\n\t" // |H3-H4|

  1076. 		"psrlw $1, %%mm0				\n\t" // |L3 - L4|/2

  1077. 		"psrlw $1, %%mm1				\n\t" // |H3 - H4|/2

  1078. 

  1079. 		"pxor %%mm6, %%mm2				\n\t"

  1080. 		"pxor %%mm7, %%mm3				\n\t"

  1081. 		"pand %%mm2, %%mm4				\n\t"

  1082. 		"pand %%mm3, %%mm5				\n\t"

  1083. 

  1084. #ifdef HAVE_MMX2

  1085. 		"pminsw %%mm0, %%mm4				\n\t"

  1086. 		"pminsw %%mm1, %%mm5				\n\t"

  1087. #else

  1088. 		"movq %%mm4, %%mm2				\n\t"

  1089. 		"psubusw %%mm0, %%mm2				\n\t"

  1090. 		"psubw %%mm2, %%mm4				\n\t"

  1091. 		"movq %%mm5, %%mm2				\n\t"

  1092. 		"psubusw %%mm1, %%mm2				\n\t"

  1093. 		"psubw %%mm2, %%mm5				\n\t"

  1094. #endif

  1095. 		"pxor %%mm6, %%mm4				\n\t"

  1096. 		"pxor %%mm7, %%mm5				\n\t"

  1097. 		"psubw %%mm6, %%mm4				\n\t"

  1098. 		"psubw %%mm7, %%mm5				\n\t"

  1099. 		"packsswb %%mm5, %%mm4				\n\t"

  1100. 		"movq (%0), %%mm0				\n\t"

  1101. 		"paddb   %%mm4, %%mm0				\n\t"

  1102. 		"movq %%mm0, (%0)				\n\t"

  1103. 		"movq (%0, %1), %%mm0				\n\t"

  1104. 		"psubb %%mm4, %%mm0				\n\t"

  1105. 		"movq %%mm0, (%0, %1)				\n\t"

  1106. 

  1107. 		: "+r" (src)

  1108. 		: "r" (stride), "m" (c->pQPb)

  1109. 		: "%eax", "%ecx"

  1110. 	);

  1111. #else

  1112. 	const int l1= stride;

  1113. 	const int l2= stride + l1;

  1114. 	const int l3= stride + l2;

  1115. 	const int l4= stride + l3;

  1116. 	const int l5= stride + l4;

  1117. 	const int l6= stride + l5;

  1118. 	const int l7= stride + l6;

  1119. 	const int l8= stride + l7;

  1120. //	const int l9= stride + l8;

  1121. 	int x;

  1122. 	src+= stride*3;

  1123. 	for(x=0; x<BLOCK_SIZE; x++)

  1124. 	{

  1125. 		const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);

  1126. 		if(ABS(middleEnergy) < 8*c->QP)

  1127. 		{

  1128. 			const int q=(src[l4] - src[l5])/2;

  1129. 			const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);

  1130. 			const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

  1131. 

  1132. 			int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );

  1133. 			d= MAX(d, 0);

  1134. 

  1135. 			d= (5*d + 32) >> 6;

  1136. 			d*= SIGN(-middleEnergy);

  1137. 

  1138. 			if(q>0)

  1139. 			{

  1140. 				d= d<0 ? 0 : d;

  1141. 				d= d>q ? q : d;

  1142. 			}

  1143. 			else

  1144. 			{

  1145. 				d= d>0 ? 0 : d;

  1146. 				d= d<q ? q : d;

  1147. 			}

  1148. 

  1149.         		src[l4]-= d;

  1150. 	        	src[l5]+= d;

  1151. 		}

  1152. 		src++;

  1153. 	}

  1154. #endif

  1155. }

  1156. #endif //HAVE_ALTIVEC

  1157. 

  1158. #ifndef HAVE_ALTIVEC

  1159. static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)

  1160. {

  1161. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1162. 	asm volatile(

  1163. 		"pxor %%mm6, %%mm6				\n\t"

  1164. 		"pcmpeqb %%mm7, %%mm7				\n\t"

  1165. 		"movq %2, %%mm0					\n\t"

  1166. 		"punpcklbw %%mm6, %%mm0				\n\t"

  1167. 		"psrlw $1, %%mm0				\n\t"

  1168. 		"psubw %%mm7, %%mm0				\n\t"

  1169. 		"packuswb %%mm0, %%mm0				\n\t"

  1170. 		"movq %%mm0, %3					\n\t"

  1171. 

  1172. 		"leal (%0, %1), %%eax				\n\t"

  1173. 		"leal (%%eax, %1, 4), %%edx			\n\t"

  1174. 		

  1175. //	0	1	2	3	4	5	6	7	8	9

  1176. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1

  1177. 

  1178. #undef FIND_MIN_MAX

  1179. #ifdef HAVE_MMX2

  1180. #define FIND_MIN_MAX(addr)\

  1181. 		"movq " #addr ", %%mm0				\n\t"\

  1182. 		"pminub %%mm0, %%mm7				\n\t"\

  1183. 		"pmaxub %%mm0, %%mm6				\n\t"

  1184. #else

  1185. #define FIND_MIN_MAX(addr)\

  1186. 		"movq " #addr ", %%mm0				\n\t"\

  1187. 		"movq %%mm7, %%mm1				\n\t"\

  1188. 		"psubusb %%mm0, %%mm6				\n\t"\

  1189. 		"paddb %%mm0, %%mm6				\n\t"\

  1190. 		"psubusb %%mm0, %%mm1				\n\t"\

  1191. 		"psubb %%mm1, %%mm7				\n\t"

  1192. #endif

  1193. 

  1194. FIND_MIN_MAX((%%eax))

  1195. FIND_MIN_MAX((%%eax, %1))

  1196. FIND_MIN_MAX((%%eax, %1, 2))

  1197. FIND_MIN_MAX((%0, %1, 4))

  1198. FIND_MIN_MAX((%%edx))

  1199. FIND_MIN_MAX((%%edx, %1))

  1200. FIND_MIN_MAX((%%edx, %1, 2))

  1201. FIND_MIN_MAX((%0, %1, 8))

  1202. 

  1203. 		"movq %%mm7, %%mm4				\n\t"

  1204. 		"psrlq $8, %%mm7				\n\t"

  1205. #ifdef HAVE_MMX2

  1206. 		"pminub %%mm4, %%mm7				\n\t" // min of pixels

  1207. 		"pshufw $0xF9, %%mm7, %%mm4			\n\t"

  1208. 		"pminub %%mm4, %%mm7				\n\t" // min of pixels

  1209. 		"pshufw $0xFE, %%mm7, %%mm4			\n\t"

  1210. 		"pminub %%mm4, %%mm7				\n\t"

  1211. #else

  1212. 		"movq %%mm7, %%mm1				\n\t"

  1213. 		"psubusb %%mm4, %%mm1				\n\t"

  1214. 		"psubb %%mm1, %%mm7				\n\t"

  1215. 		"movq %%mm7, %%mm4				\n\t"

  1216. 		"psrlq $16, %%mm7				\n\t"

  1217. 		"movq %%mm7, %%mm1				\n\t"

  1218. 		"psubusb %%mm4, %%mm1				\n\t"

  1219. 		"psubb %%mm1, %%mm7				\n\t"

  1220. 		"movq %%mm7, %%mm4				\n\t"

  1221. 		"psrlq $32, %%mm7				\n\t"

  1222. 		"movq %%mm7, %%mm1				\n\t"

  1223. 		"psubusb %%mm4, %%mm1				\n\t"

  1224. 		"psubb %%mm1, %%mm7				\n\t"

  1225. #endif

  1226. 

  1227. 

  1228. 		"movq %%mm6, %%mm4				\n\t"

  1229. 		"psrlq $8, %%mm6				\n\t"

  1230. #ifdef HAVE_MMX2

  1231. 		"pmaxub %%mm4, %%mm6				\n\t" // max of pixels

  1232. 		"pshufw $0xF9, %%mm6, %%mm4			\n\t"

  1233. 		"pmaxub %%mm4, %%mm6				\n\t"

  1234. 		"pshufw $0xFE, %%mm6, %%mm4			\n\t"

  1235. 		"pmaxub %%mm4, %%mm6				\n\t"

  1236. #else

  1237. 		"psubusb %%mm4, %%mm6				\n\t"

  1238. 		"paddb %%mm4, %%mm6				\n\t"

  1239. 		"movq %%mm6, %%mm4				\n\t"

  1240. 		"psrlq $16, %%mm6				\n\t"

  1241. 		"psubusb %%mm4, %%mm6				\n\t"

  1242. 		"paddb %%mm4, %%mm6				\n\t"

  1243. 		"movq %%mm6, %%mm4				\n\t"

  1244. 		"psrlq $32, %%mm6				\n\t"

  1245. 		"psubusb %%mm4, %%mm6				\n\t"

  1246. 		"paddb %%mm4, %%mm6				\n\t"

  1247. #endif

  1248. 		"movq %%mm6, %%mm0				\n\t" // max

  1249. 		"psubb %%mm7, %%mm6				\n\t" // max - min

  1250. 		"movd %%mm6, %%ecx				\n\t"

  1251. 		"cmpb "MANGLE(deringThreshold)", %%cl		\n\t"

  1252. 		" jb 1f						\n\t"

  1253. 		"leal -24(%%esp), %%ecx				\n\t"

  1254. 		"andl $0xFFFFFFF8, %%ecx			\n\t" 

  1255. 		PAVGB(%%mm0, %%mm7)				      // a=(max + min)/2

  1256. 		"punpcklbw %%mm7, %%mm7				\n\t"

  1257. 		"punpcklbw %%mm7, %%mm7				\n\t"

  1258. 		"punpcklbw %%mm7, %%mm7				\n\t"

  1259. 		"movq %%mm7, (%%ecx)				\n\t"

  1260. 

  1261. 		"movq (%0), %%mm0				\n\t" // L10

  1262. 		"movq %%mm0, %%mm1				\n\t" // L10

  1263. 		"movq %%mm0, %%mm2				\n\t" // L10

  1264. 		"psllq $8, %%mm1				\n\t"

  1265. 		"psrlq $8, %%mm2				\n\t"

  1266. 		"movd -4(%0), %%mm3				\n\t"

  1267. 		"movd 8(%0), %%mm4				\n\t"

  1268. 		"psrlq $24, %%mm3				\n\t"

  1269. 		"psllq $56, %%mm4				\n\t"

  1270. 		"por %%mm3, %%mm1				\n\t" // L00

  1271. 		"por %%mm4, %%mm2				\n\t" // L20

  1272. 		"movq %%mm1, %%mm3				\n\t" // L00

  1273. 		PAVGB(%%mm2, %%mm1)				      // (L20 + L00)/2

  1274. 		PAVGB(%%mm0, %%mm1)				      // (L20 + L00 + 2L10)/4

  1275. 		"psubusb %%mm7, %%mm0				\n\t"

  1276. 		"psubusb %%mm7, %%mm2				\n\t"

  1277. 		"psubusb %%mm7, %%mm3				\n\t"

  1278. 		"pcmpeqb "MANGLE(b00)", %%mm0			\n\t" // L10 > a ? 0 : -1

  1279. 		"pcmpeqb "MANGLE(b00)", %%mm2			\n\t" // L20 > a ? 0 : -1

  1280. 		"pcmpeqb "MANGLE(b00)", %%mm3			\n\t" // L00 > a ? 0 : -1

  1281. 		"paddb %%mm2, %%mm0				\n\t"

  1282. 		"paddb %%mm3, %%mm0				\n\t"

  1283. 

  1284. 		"movq (%%eax), %%mm2				\n\t" // L11

  1285. 		"movq %%mm2, %%mm3				\n\t" // L11

  1286. 		"movq %%mm2, %%mm4				\n\t" // L11

  1287. 		"psllq $8, %%mm3				\n\t"

  1288. 		"psrlq $8, %%mm4				\n\t"

  1289. 		"movd -4(%%eax), %%mm5				\n\t"

  1290. 		"movd 8(%%eax), %%mm6				\n\t"

  1291. 		"psrlq $24, %%mm5				\n\t"

  1292. 		"psllq $56, %%mm6				\n\t"

  1293. 		"por %%mm5, %%mm3				\n\t" // L01

  1294. 		"por %%mm6, %%mm4				\n\t" // L21

  1295. 		"movq %%mm3, %%mm5				\n\t" // L01

  1296. 		PAVGB(%%mm4, %%mm3)				      // (L21 + L01)/2

  1297. 		PAVGB(%%mm2, %%mm3)				      // (L21 + L01 + 2L11)/4

  1298. 		"psubusb %%mm7, %%mm2				\n\t"

  1299. 		"psubusb %%mm7, %%mm4				\n\t"

  1300. 		"psubusb %%mm7, %%mm5				\n\t"

  1301. 		"pcmpeqb "MANGLE(b00)", %%mm2			\n\t" // L11 > a ? 0 : -1

  1302. 		"pcmpeqb "MANGLE(b00)", %%mm4			\n\t" // L21 > a ? 0 : -1

  1303. 		"pcmpeqb "MANGLE(b00)", %%mm5			\n\t" // L01 > a ? 0 : -1

  1304. 		"paddb %%mm4, %%mm2				\n\t"

  1305. 		"paddb %%mm5, %%mm2				\n\t"

  1306. // 0, 2, 3, 1

  1307. #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \

  1308. 		"movq " #src ", " #sx "				\n\t" /* src[0] */\

  1309. 		"movq " #sx ", " #lx "				\n\t" /* src[0] */\

  1310. 		"movq " #sx ", " #t0 "				\n\t" /* src[0] */\

  1311. 		"psllq $8, " #lx "				\n\t"\

  1312. 		"psrlq $8, " #t0 "				\n\t"\

  1313. 		"movd -4" #src ", " #t1 "			\n\t"\

  1314. 		"psrlq $24, " #t1 "				\n\t"\

  1315. 		"por " #t1 ", " #lx "				\n\t" /* src[-1] */\

  1316. 		"movd 8" #src ", " #t1 "			\n\t"\

  1317. 		"psllq $56, " #t1 "				\n\t"\

  1318. 		"por " #t1 ", " #t0 "				\n\t" /* src[+1] */\

  1319. 		"movq " #lx ", " #t1 "				\n\t" /* src[-1] */\

  1320. 		PAVGB(t0, lx)				              /* (src[-1] + src[+1])/2 */\

  1321. 		PAVGB(sx, lx)				      /* (src[-1] + 2src[0] + src[+1])/4 */\

  1322. 		PAVGB(lx, pplx)					     \

  1323. 		"movq " #lx ", 8(%%ecx)				\n\t"\

  1324. 		"movq (%%ecx), " #lx "				\n\t"\

  1325. 		"psubusb " #lx ", " #t1 "			\n\t"\

  1326. 		"psubusb " #lx ", " #t0 "			\n\t"\

  1327. 		"psubusb " #lx ", " #sx "			\n\t"\

  1328. 		"movq "MANGLE(b00)", " #lx "			\n\t"\

  1329. 		"pcmpeqb " #lx ", " #t1 "			\n\t" /* src[-1] > a ? 0 : -1*/\

  1330. 		"pcmpeqb " #lx ", " #t0 "			\n\t" /* src[+1] > a ? 0 : -1*/\

  1331. 		"pcmpeqb " #lx ", " #sx "			\n\t" /* src[0]  > a ? 0 : -1*/\

  1332. 		"paddb " #t1 ", " #t0 "				\n\t"\

  1333. 		"paddb " #t0 ", " #sx "				\n\t"\

  1334. \

  1335. 		PAVGB(plx, pplx)				      /* filtered */\

  1336. 		"movq " #dst ", " #t0 "				\n\t" /* dst */\

  1337. 		"movq " #t0 ", " #t1 "				\n\t" /* dst */\

  1338. 		"psubusb %3, " #t0 "				\n\t"\

  1339. 		"paddusb %3, " #t1 "				\n\t"\

  1340. 		PMAXUB(t0, pplx)\

  1341. 		PMINUB(t1, pplx, t0)\

  1342. 		"paddb " #sx ", " #ppsx "			\n\t"\

  1343. 		"paddb " #psx ", " #ppsx "			\n\t"\

  1344. 		"#paddb "MANGLE(b02)", " #ppsx "		\n\t"\

  1345. 		"pand "MANGLE(b08)", " #ppsx "			\n\t"\

  1346. 		"pcmpeqb " #lx ", " #ppsx "			\n\t"\

  1347. 		"pand " #ppsx ", " #pplx "			\n\t"\

  1348. 		"pandn " #dst ", " #ppsx "			\n\t"\

  1349. 		"por " #pplx ", " #ppsx "			\n\t"\

  1350. 		"movq " #ppsx ", " #dst "			\n\t"\

  1351. 		"movq 8(%%ecx), " #lx "				\n\t"

  1352. 

  1353. /*

  1354. 0000000

  1355. 1111111

  1356. 

  1357. 1111110

  1358. 1111101

  1359. 1111100

  1360. 1111011

  1361. 1111010

  1362. 1111001

  1363. 

  1364. 1111000

  1365. 1110111

  1366. 

  1367. */

  1368. //DERING_CORE(dst,src                  ,ppsx ,psx  ,sx   ,pplx ,plx  ,lx   ,t0   ,t1)

  1369. DERING_CORE((%%eax),(%%eax, %1)        ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1370. DERING_CORE((%%eax, %1),(%%eax, %1, 2) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1371. DERING_CORE((%%eax, %1, 2),(%0, %1, 4) ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)

  1372. DERING_CORE((%0, %1, 4),(%%edx)        ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1373. DERING_CORE((%%edx),(%%edx, %1)        ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1374. DERING_CORE((%%edx, %1), (%%edx, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)

  1375. DERING_CORE((%%edx, %1, 2),(%0, %1, 8) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1376. DERING_CORE((%0, %1, 8),(%%edx, %1, 4) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1377. 

  1378. 		"1:			\n\t"

  1379. 		: : "r" (src), "r" (stride), "m" (c->pQPb), "m"(c->pQPb2)

  1380. 		: "%eax", "%edx", "%ecx"

  1381. 	);

  1382. #else

  1383. 	int y;

  1384. 	int min=255;

  1385. 	int max=0;

  1386. 	int avg;

  1387. 	uint8_t *p;

  1388. 	int s[10];

  1389. 	const int QP2= c->QP/2 + 1;

  1390. 

  1391. 	for(y=1; y<9; y++)

  1392. 	{

  1393. 		int x;

  1394. 		p= src + stride*y;

  1395. 		for(x=1; x<9; x++)

  1396. 		{

  1397. 			p++;

  1398. 			if(*p > max) max= *p;

  1399. 			if(*p < min) min= *p;

  1400. 		}

  1401. 	}

  1402. 	avg= (min + max + 1)>>1;

  1403. 

  1404. 	if(max - min <deringThreshold) return;

  1405. 

  1406. 	for(y=0; y<10; y++)

  1407. 	{

  1408. 		int t = 0;

  1409. 

  1410. 		if(src[stride*y + 0] > avg) t+= 1;

  1411. 		if(src[stride*y + 1] > avg) t+= 2;

  1412. 		if(src[stride*y + 2] > avg) t+= 4;

  1413. 		if(src[stride*y + 3] > avg) t+= 8;

  1414. 		if(src[stride*y + 4] > avg) t+= 16;

  1415. 		if(src[stride*y + 5] > avg) t+= 32;

  1416. 		if(src[stride*y + 6] > avg) t+= 64;

  1417. 		if(src[stride*y + 7] > avg) t+= 128;

  1418. 		if(src[stride*y + 8] > avg) t+= 256;

  1419. 		if(src[stride*y + 9] > avg) t+= 512;

  1420. 		

  1421. 		t |= (~t)<<16;

  1422. 		t &= (t<<1) & (t>>1);

  1423. 		s[y] = t;

  1424. 	}

  1425. 	

  1426. 	for(y=1; y<9; y++)

  1427. 	{

  1428. 		int t = s[y-1] & s[y] & s[y+1];

  1429. 		t|= t>>16;

  1430. 		s[y-1]= t;

  1431. 	}

  1432. 

  1433. 	for(y=1; y<9; y++)

  1434. 	{

  1435. 		int x;

  1436. 		int t = s[y-1];

  1437. 

  1438. 		p= src + stride*y;

  1439. 		for(x=1; x<9; x++)

  1440. 		{

  1441. 			p++;

  1442. 			if(t & (1<<x))

  1443. 			{

  1444. 				int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))

  1445. 				      +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))

  1446. 				      +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));

  1447. 				f= (f + 8)>>4;

  1448. 

  1449. #ifdef DEBUG_DERING_THRESHOLD

  1450. 				asm volatile("emms\n\t":);

  1451. 				{

  1452. 				static long long numPixels=0;

  1453. 				if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;

  1454. //				if((max-min)<20 || (max-min)*QP<200)

  1455. //				if((max-min)*QP < 500)

  1456. //				if(max-min<QP/2)

  1457. 				if(max-min < 20)

  1458. 				{

  1459. 					static int numSkiped=0;

  1460. 					static int errorSum=0;

  1461. 					static int worstQP=0;

  1462. 					static int worstRange=0;

  1463. 					static int worstDiff=0;

  1464. 					int diff= (f - *p);

  1465. 					int absDiff= ABS(diff);

  1466. 					int error= diff*diff;

  1467. 

  1468. 					if(x==1 || x==8 || y==1 || y==8) continue;

  1469. 

  1470. 					numSkiped++;

  1471. 					if(absDiff > worstDiff)

  1472. 					{

  1473. 						worstDiff= absDiff;

  1474. 						worstQP= QP;

  1475. 						worstRange= max-min;

  1476. 					}

  1477. 					errorSum+= error;

  1478. 

  1479. 					if(1024LL*1024LL*1024LL % numSkiped == 0)

  1480. 					{

  1481. 						printf( "sum:%1.3f, skip:%d, wQP:%d, "

  1482. 							"wRange:%d, wDiff:%d, relSkip:%1.3f\n",

  1483. 							(float)errorSum/numSkiped, numSkiped, worstQP, worstRange,

  1484. 							worstDiff, (float)numSkiped/numPixels);

  1485. 					}

  1486. 				}

  1487. 				}

  1488. #endif

  1489. 				if     (*p + QP2 < f) *p= *p + QP2;

  1490. 				else if(*p - QP2 > f) *p= *p - QP2;

  1491. 				else *p=f;

  1492. 			}

  1493. 		}

  1494. 	}

  1495. #ifdef DEBUG_DERING_THRESHOLD

  1496. 	if(max-min < 20)

  1497. 	{

  1498. 		for(y=1; y<9; y++)

  1499. 		{

  1500. 			int x;

  1501. 			int t = 0;

  1502. 			p= src + stride*y;

  1503. 			for(x=1; x<9; x++)

  1504. 			{

  1505. 				p++;

  1506. 				*p = MIN(*p + 20, 255);

  1507. 			}

  1508. 		}

  1509. //		src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;

  1510. 	}

  1511. #endif

  1512. #endif

  1513. }

  1514. #endif //HAVE_ALTIVEC

  1515. 

  1516. /**

  1517.  * Deinterlaces the given block by linearly interpolating every second line.

  1518.  * will be called for every 8x8 block and can read & write from line 4-15

  1519.  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too

  1520.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1521.  */

  1522. static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)

  1523. {

  1524. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1525. 	src+= 4*stride;

  1526. 	asm volatile(

  1527. 		"leal (%0, %1), %%eax				\n\t"

  1528. 		"leal (%%eax, %1, 4), %%ecx			\n\t"

  1529. //	0	1	2	3	4	5	6	7	8	9

  1530. //	%0	eax	eax+%1	eax+2%1	%0+4%1	ecx	ecx+%1	ecx+2%1	%0+8%1	ecx+4%1

  1531. 

  1532. 		"movq (%0), %%mm0				\n\t"

  1533. 		"movq (%%eax, %1), %%mm1			\n\t"

  1534. 		PAVGB(%%mm1, %%mm0)

  1535. 		"movq %%mm0, (%%eax)				\n\t"

  1536. 		"movq (%0, %1, 4), %%mm0			\n\t"

  1537. 		PAVGB(%%mm0, %%mm1)

  1538. 		"movq %%mm1, (%%eax, %1, 2)			\n\t"

  1539. 		"movq (%%ecx, %1), %%mm1			\n\t"

  1540. 		PAVGB(%%mm1, %%mm0)

  1541. 		"movq %%mm0, (%%ecx)				\n\t"

  1542. 		"movq (%0, %1, 8), %%mm0			\n\t"

  1543. 		PAVGB(%%mm0, %%mm1)

  1544. 		"movq %%mm1, (%%ecx, %1, 2)			\n\t"

  1545. 

  1546. 		: : "r" (src), "r" (stride)

  1547. 		: "%eax", "%ecx"

  1548. 	);

  1549. #else

  1550. 	int a, b, x;

  1551. 	src+= 4*stride;

  1552. 

  1553. 	for(x=0; x<2; x++){

  1554. 		a= *(uint32_t*)&src[stride*0];

  1555. 		b= *(uint32_t*)&src[stride*2];

  1556. 		*(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1557. 		a= *(uint32_t*)&src[stride*4];

  1558. 		*(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1559. 		b= *(uint32_t*)&src[stride*6];

  1560. 		*(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1561. 		a= *(uint32_t*)&src[stride*8];

  1562. 		*(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1563. 		src += 4;

  1564. 	}

  1565. #endif

  1566. }

  1567. 

  1568. /**

  1569.  * Deinterlaces the given block by cubic interpolating every second line.

  1570.  * will be called for every 8x8 block and can read & write from line 4-15

  1571.  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too

  1572.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1573.  * this filter will read lines 3-15 and write 7-13

  1574.  */

  1575. static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)

  1576. {

  1577. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1578. 	src+= stride*3;

  1579. 	asm volatile(

  1580. 		"leal (%0, %1), %%eax				\n\t"

  1581. 		"leal (%%eax, %1, 4), %%edx			\n\t"

  1582. 		"leal (%%edx, %1, 4), %%ecx			\n\t"

  1583. 		"addl %1, %%ecx					\n\t"

  1584. 		"pxor %%mm7, %%mm7				\n\t"

  1585. //	0	1	2	3	4	5	6	7	8	9	10

  1586. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1 ecx

  1587. 

  1588. #define DEINT_CUBIC(a,b,c,d,e)\

  1589. 		"movq " #a ", %%mm0				\n\t"\

  1590. 		"movq " #b ", %%mm1				\n\t"\

  1591. 		"movq " #d ", %%mm2				\n\t"\

  1592. 		"movq " #e ", %%mm3				\n\t"\

  1593. 		PAVGB(%%mm2, %%mm1)					/* (b+d) /2 */\

  1594. 		PAVGB(%%mm3, %%mm0)					/* a(a+e) /2 */\

  1595. 		"movq %%mm0, %%mm2				\n\t"\

  1596. 		"punpcklbw %%mm7, %%mm0				\n\t"\

  1597. 		"punpckhbw %%mm7, %%mm2				\n\t"\

  1598. 		"movq %%mm1, %%mm3				\n\t"\

  1599. 		"punpcklbw %%mm7, %%mm1				\n\t"\

  1600. 		"punpckhbw %%mm7, %%mm3				\n\t"\

  1601. 		"psubw %%mm1, %%mm0				\n\t"	/* L(a+e - (b+d))/2 */\

  1602. 		"psubw %%mm3, %%mm2				\n\t"	/* H(a+e - (b+d))/2 */\

  1603. 		"psraw $3, %%mm0				\n\t"	/* L(a+e - (b+d))/16 */\

  1604. 		"psraw $3, %%mm2				\n\t"	/* H(a+e - (b+d))/16 */\

  1605. 		"psubw %%mm0, %%mm1				\n\t"	/* L(9b + 9d - a - e)/16 */\

  1606. 		"psubw %%mm2, %%mm3				\n\t"	/* H(9b + 9d - a - e)/16 */\

  1607. 		"packuswb %%mm3, %%mm1				\n\t"\

  1608. 		"movq %%mm1, " #c "				\n\t"

  1609. 

  1610. DEINT_CUBIC((%0), (%%eax, %1), (%%eax, %1, 2), (%0, %1, 4), (%%edx, %1))

  1611. DEINT_CUBIC((%%eax, %1), (%0, %1, 4), (%%edx), (%%edx, %1), (%0, %1, 8))

  1612. DEINT_CUBIC((%0, %1, 4), (%%edx, %1), (%%edx, %1, 2), (%0, %1, 8), (%%ecx))

  1613. DEINT_CUBIC((%%edx, %1), (%0, %1, 8), (%%edx, %1, 4), (%%ecx), (%%ecx, %1, 2))

  1614. 

  1615. 		: : "r" (src), "r" (stride)

  1616. 		: "%eax", "%edx", "ecx"

  1617. 	);

  1618. #else

  1619. 	int x;

  1620. 	src+= stride*3;

  1621. 	for(x=0; x<8; x++)

  1622. 	{

  1623. 		src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);

  1624. 		src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);

  1625. 		src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);

  1626. 		src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);

  1627. 		src++;

  1628. 	}

  1629. #endif

  1630. }

  1631. 

  1632. /**

  1633.  * Deinterlaces the given block by filtering every second line with a (-1 4 2 4 -1) filter.

  1634.  * will be called for every 8x8 block and can read & write from line 4-15

  1635.  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too

  1636.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1637.  * this filter will read lines 4-13 and write 5-11

  1638.  */

  1639. static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)

  1640. {

  1641. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1642. 	src+= stride*4;

  1643. 	asm volatile(

  1644. 		"leal (%0, %1), %%eax				\n\t"

  1645. 		"leal (%%eax, %1, 4), %%edx			\n\t"

  1646. 		"pxor %%mm7, %%mm7				\n\t"

  1647. 		"movq (%2), %%mm0				\n\t"

  1648. //	0	1	2	3	4	5	6	7	8	9	10

  1649. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1 ecx

  1650. 

  1651. #define DEINT_FF(a,b,c,d)\

  1652. 		"movq " #a ", %%mm1				\n\t"\

  1653. 		"movq " #b ", %%mm2				\n\t"\

  1654. 		"movq " #c ", %%mm3				\n\t"\

  1655. 		"movq " #d ", %%mm4				\n\t"\

  1656. 		PAVGB(%%mm3, %%mm1)					\

  1657. 		PAVGB(%%mm4, %%mm0)					\

  1658. 		"movq %%mm0, %%mm3				\n\t"\

  1659. 		"punpcklbw %%mm7, %%mm0				\n\t"\

  1660. 		"punpckhbw %%mm7, %%mm3				\n\t"\

  1661. 		"movq %%mm1, %%mm4				\n\t"\

  1662. 		"punpcklbw %%mm7, %%mm1				\n\t"\

  1663. 		"punpckhbw %%mm7, %%mm4				\n\t"\

  1664. 		"psllw $2, %%mm1				\n\t"\

  1665. 		"psllw $2, %%mm4				\n\t"\

  1666. 		"psubw %%mm0, %%mm1				\n\t"\

  1667. 		"psubw %%mm3, %%mm4				\n\t"\

  1668. 		"movq %%mm2, %%mm5				\n\t"\

  1669. 		"movq %%mm2, %%mm0				\n\t"\

  1670. 		"punpcklbw %%mm7, %%mm2				\n\t"\

  1671. 		"punpckhbw %%mm7, %%mm5				\n\t"\

  1672. 		"paddw %%mm2, %%mm1				\n\t"\

  1673. 		"paddw %%mm5, %%mm4				\n\t"\

  1674. 		"psraw $2, %%mm1				\n\t"\

  1675. 		"psraw $2, %%mm4				\n\t"\

  1676. 		"packuswb %%mm4, %%mm1				\n\t"\

  1677. 		"movq %%mm1, " #b "				\n\t"\

  1678. 

  1679. DEINT_FF((%0)       , (%%eax)       , (%%eax, %1), (%%eax, %1, 2))

  1680. DEINT_FF((%%eax, %1), (%%eax, %1, 2), (%0, %1, 4), (%%edx)       )

  1681. DEINT_FF((%0, %1, 4), (%%edx)       , (%%edx, %1), (%%edx, %1, 2))

  1682. DEINT_FF((%%edx, %1), (%%edx, %1, 2), (%0, %1, 8), (%%edx, %1, 4))

  1683. 

  1684. 		"movq %%mm0, (%2)				\n\t"

  1685. 		: : "r" (src), "r" (stride), "r"(tmp)

  1686. 		: "%eax", "%edx"

  1687. 	);

  1688. #else

  1689. 	int x;

  1690. 	src+= stride*4;

  1691. 	for(x=0; x<8; x++)

  1692. 	{

  1693. 		int t1= tmp[x];

  1694. 		int t2= src[stride*1];

  1695. 

  1696. 		src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);

  1697. 		t1= src[stride*4];

  1698. 		src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);

  1699. 		t2= src[stride*6];

  1700. 		src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);

  1701. 		t1= src[stride*8];

  1702. 		src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);

  1703. 		tmp[x]= t1;

  1704. 

  1705. 		src++;

  1706. 	}

  1707. #endif

  1708. }

  1709. 

  1710. /**

  1711.  * Deinterlaces the given block by filtering every line with a (-1 2 6 2 -1) filter.

  1712.  * will be called for every 8x8 block and can read & write from line 4-15

  1713.  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too

  1714.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1715.  * this filter will read lines 4-13 and write 4-11

  1716.  */

  1717. static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)

  1718. {

  1719. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1720. 	src+= stride*4;

  1721. 	asm volatile(

  1722. 		"leal (%0, %1), %%eax				\n\t"

  1723. 		"leal (%%eax, %1, 4), %%edx			\n\t"

  1724. 		"pxor %%mm7, %%mm7				\n\t"

  1725. 		"movq (%2), %%mm0				\n\t"

  1726. 		"movq (%3), %%mm1				\n\t"

  1727. //	0	1	2	3	4	5	6	7	8	9	10

  1728. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1 ecx

  1729. 

  1730. #define DEINT_L5(t1,t2,a,b,c)\

  1731. 		"movq " #a ", %%mm2				\n\t"\

  1732. 		"movq " #b ", %%mm3				\n\t"\

  1733. 		"movq " #c ", %%mm4				\n\t"\

  1734. 		PAVGB(t2, %%mm3)					\

  1735. 		PAVGB(t1, %%mm4)					\

  1736. 		"movq %%mm2, %%mm5				\n\t"\

  1737. 		"movq %%mm2, " #t1 "				\n\t"\

  1738. 		"punpcklbw %%mm7, %%mm2				\n\t"\

  1739. 		"punpckhbw %%mm7, %%mm5				\n\t"\

  1740. 		"movq %%mm2, %%mm6				\n\t"\

  1741. 		"paddw %%mm2, %%mm2				\n\t"\

  1742. 		"paddw %%mm6, %%mm2				\n\t"\

  1743. 		"movq %%mm5, %%mm6				\n\t"\

  1744. 		"paddw %%mm5, %%mm5				\n\t"\

  1745. 		"paddw %%mm6, %%mm5				\n\t"\

  1746. 		"movq %%mm3, %%mm6				\n\t"\

  1747. 		"punpcklbw %%mm7, %%mm3				\n\t"\

  1748. 		"punpckhbw %%mm7, %%mm6				\n\t"\

  1749. 		"paddw %%mm3, %%mm3				\n\t"\

  1750. 		"paddw %%mm6, %%mm6				\n\t"\

  1751. 		"paddw %%mm3, %%mm2				\n\t"\

  1752. 		"paddw %%mm6, %%mm5				\n\t"\

  1753. 		"movq %%mm4, %%mm6				\n\t"\

  1754. 		"punpcklbw %%mm7, %%mm4				\n\t"\

  1755. 		"punpckhbw %%mm7, %%mm6				\n\t"\

  1756. 		"psubw %%mm4, %%mm2				\n\t"\

  1757. 		"psubw %%mm6, %%mm5				\n\t"\

  1758. 		"psraw $2, %%mm2				\n\t"\

  1759. 		"psraw $2, %%mm5				\n\t"\

  1760. 		"packuswb %%mm5, %%mm2				\n\t"\

  1761. 		"movq %%mm2, " #a "				\n\t"\

  1762. 

  1763. DEINT_L5(%%mm0, %%mm1, (%0)          , (%%eax)       , (%%eax, %1)   )

  1764. DEINT_L5(%%mm1, %%mm0, (%%eax)       , (%%eax, %1)   , (%%eax, %1, 2))

  1765. DEINT_L5(%%mm0, %%mm1, (%%eax, %1)   , (%%eax, %1, 2), (%0, %1, 4)   )

  1766. DEINT_L5(%%mm1, %%mm0, (%%eax, %1, 2), (%0, %1, 4)   , (%%edx)       )

  1767. DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)   , (%%edx)       , (%%edx, %1)   )  

  1768. DEINT_L5(%%mm1, %%mm0, (%%edx)       , (%%edx, %1)   , (%%edx, %1, 2))

  1769. DEINT_L5(%%mm0, %%mm1, (%%edx, %1)   , (%%edx, %1, 2), (%0, %1, 8)   )

  1770. DEINT_L5(%%mm1, %%mm0, (%%edx, %1, 2), (%0, %1, 8)   , (%%edx, %1, 4))

  1771. 

  1772. 		"movq %%mm0, (%2)				\n\t"

  1773. 		"movq %%mm1, (%3)				\n\t"

  1774. 		: : "r" (src), "r" (stride), "r"(tmp), "r"(tmp2)

  1775. 		: "%eax", "%edx"

  1776. 	);

  1777. #else

  1778. 	int x;

  1779. 	src+= stride*4;

  1780. 	for(x=0; x<8; x++)

  1781. 	{

  1782. 		int t1= tmp[x];

  1783. 		int t2= tmp2[x];

  1784. 		int t3= src[0];

  1785. 

  1786. 		src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);

  1787. 		t1= src[stride*1];

  1788. 		src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);

  1789. 		t2= src[stride*2];

  1790. 		src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);

  1791. 		t3= src[stride*3];

  1792. 		src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);

  1793. 		t1= src[stride*4];

  1794. 		src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);

  1795. 		t2= src[stride*5];

  1796. 		src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);

  1797. 		t3= src[stride*6];

  1798. 		src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);

  1799. 		t1= src[stride*7];

  1800. 		src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);

  1801. 

  1802. 		tmp[x]= t3;

  1803. 		tmp2[x]= t1;

  1804. 

  1805. 		src++;

  1806. 	}

  1807. #endif

  1808. }

  1809. 

  1810. /**

  1811.  * Deinterlaces the given block by filtering all lines with a (1 2 1) filter.

  1812.  * will be called for every 8x8 block and can read & write from line 4-15

  1813.  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too

  1814.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1815.  * this filter will read lines 4-13 and write 4-11

  1816.  */

  1817. static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)

  1818. {

  1819. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1820. 	src+= 4*stride;

  1821. 	asm volatile(

  1822. 		"leal (%0, %1), %%eax				\n\t"

  1823. 		"leal (%%eax, %1, 4), %%edx			\n\t"

  1824. //	0	1	2	3	4	5	6	7	8	9

  1825. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1

  1826. 

  1827. 		"movq (%2), %%mm0				\n\t" // L0

  1828. 		"movq (%%eax), %%mm1				\n\t" // L2

  1829. 		PAVGB(%%mm1, %%mm0)				      // L0+L2

  1830. 		"movq (%0), %%mm2				\n\t" // L1

  1831. 		PAVGB(%%mm2, %%mm0)

  1832. 		"movq %%mm0, (%0)				\n\t"

  1833. 		"movq (%%eax, %1), %%mm0			\n\t" // L3

  1834. 		PAVGB(%%mm0, %%mm2)				      // L1+L3

  1835. 		PAVGB(%%mm1, %%mm2)				      // 2L2 + L1 + L3

  1836. 		"movq %%mm2, (%%eax)				\n\t"

  1837. 		"movq (%%eax, %1, 2), %%mm2			\n\t" // L4

  1838. 		PAVGB(%%mm2, %%mm1)				      // L2+L4

  1839. 		PAVGB(%%mm0, %%mm1)				      // 2L3 + L2 + L4

  1840. 		"movq %%mm1, (%%eax, %1)			\n\t"

  1841. 		"movq (%0, %1, 4), %%mm1			\n\t" // L5

  1842. 		PAVGB(%%mm1, %%mm0)				      // L3+L5

  1843. 		PAVGB(%%mm2, %%mm0)				      // 2L4 + L3 + L5

  1844. 		"movq %%mm0, (%%eax, %1, 2)			\n\t"

  1845. 		"movq (%%edx), %%mm0				\n\t" // L6

  1846. 		PAVGB(%%mm0, %%mm2)				      // L4+L6

  1847. 		PAVGB(%%mm1, %%mm2)				      // 2L5 + L4 + L6

  1848. 		"movq %%mm2, (%0, %1, 4)			\n\t"

  1849. 		"movq (%%edx, %1), %%mm2			\n\t" // L7

  1850. 		PAVGB(%%mm2, %%mm1)				      // L5+L7

  1851. 		PAVGB(%%mm0, %%mm1)				      // 2L6 + L5 + L7

  1852. 		"movq %%mm1, (%%edx)				\n\t"

  1853. 		"movq (%%edx, %1, 2), %%mm1			\n\t" // L8

  1854. 		PAVGB(%%mm1, %%mm0)				      // L6+L8

  1855. 		PAVGB(%%mm2, %%mm0)				      // 2L7 + L6 + L8

  1856. 		"movq %%mm0, (%%edx, %1)			\n\t"

  1857. 		"movq (%0, %1, 8), %%mm0			\n\t" // L9

  1858. 		PAVGB(%%mm0, %%mm2)				      // L7+L9

  1859. 		PAVGB(%%mm1, %%mm2)				      // 2L8 + L7 + L9

  1860. 		"movq %%mm2, (%%edx, %1, 2)			\n\t"

  1861. 		"movq %%mm1, (%2)				\n\t"

  1862. 

  1863. 		: : "r" (src), "r" (stride), "r" (tmp)

  1864. 		: "%eax", "%edx"

  1865. 	);

  1866. #else

  1867. 	int a, b, c, x;

  1868. 	src+= 4*stride;

  1869. 

  1870. 	for(x=0; x<2; x++){

  1871. 		a= *(uint32_t*)&tmp[stride*0];

  1872. 		b= *(uint32_t*)&src[stride*0];

  1873. 		c= *(uint32_t*)&src[stride*1];

  1874. 		a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1875. 		*(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1876. 

  1877. 		a= *(uint32_t*)&src[stride*2];

  1878. 		b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1879. 		*(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1880. 

  1881. 		b= *(uint32_t*)&src[stride*3];

  1882. 		c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);

  1883. 		*(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);

  1884. 

  1885. 		c= *(uint32_t*)&src[stride*4];

  1886. 		a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1887. 		*(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1888. 

  1889. 		a= *(uint32_t*)&src[stride*5];

  1890. 		b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1891. 		*(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1892. 

  1893. 		b= *(uint32_t*)&src[stride*6];

  1894. 		c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);

  1895. 		*(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);

  1896. 

  1897. 		c= *(uint32_t*)&src[stride*7];

  1898. 		a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1899. 		*(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1900. 

  1901. 		a= *(uint32_t*)&src[stride*8];

  1902. 		b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1903. 		*(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1904. 

  1905. 		*(uint32_t*)&tmp[stride*0]= c;

  1906. 		src += 4;

  1907. 		tmp += 4;

  1908. 	}

  1909. #endif

  1910. }

  1911. 

  1912. /**

  1913.  * Deinterlaces the given block by applying a median filter to every second line.

  1914.  * will be called for every 8x8 block and can read & write from line 4-15,

  1915.  * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too

  1916.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1917.  */

  1918. static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)

  1919. {

  1920. #ifdef HAVE_MMX

  1921. 	src+= 4*stride;

  1922. #ifdef HAVE_MMX2

  1923. 	asm volatile(

  1924. 		"leal (%0, %1), %%eax				\n\t"

  1925. 		"leal (%%eax, %1, 4), %%edx			\n\t"

  1926. //	0	1	2	3	4	5	6	7	8	9

  1927. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1

  1928. 

  1929. 		"movq (%0), %%mm0				\n\t" //

  1930. 		"movq (%%eax, %1), %%mm2			\n\t" //

  1931. 		"movq (%%eax), %%mm1				\n\t" //

  1932. 		"movq %%mm0, %%mm3				\n\t"

  1933. 		"pmaxub %%mm1, %%mm0				\n\t" //

  1934. 		"pminub %%mm3, %%mm1				\n\t" //

  1935. 		"pmaxub %%mm2, %%mm1				\n\t" //

  1936. 		"pminub %%mm1, %%mm0				\n\t"

  1937. 		"movq %%mm0, (%%eax)				\n\t"

  1938. 

  1939. 		"movq (%0, %1, 4), %%mm0			\n\t" //

  1940. 		"movq (%%eax, %1, 2), %%mm1			\n\t" //

  1941. 		"movq %%mm2, %%mm3				\n\t"

  1942. 		"pmaxub %%mm1, %%mm2				\n\t" //

  1943. 		"pminub %%mm3, %%mm1				\n\t" //

  1944. 		"pmaxub %%mm0, %%mm1				\n\t" //

  1945. 		"pminub %%mm1, %%mm2				\n\t"

  1946. 		"movq %%mm2, (%%eax, %1, 2)			\n\t"

  1947. 

  1948. 		"movq (%%edx), %%mm2				\n\t" //

  1949. 		"movq (%%edx, %1), %%mm1			\n\t" //

  1950. 		"movq %%mm2, %%mm3				\n\t"

  1951. 		"pmaxub %%mm0, %%mm2				\n\t" //

  1952. 		"pminub %%mm3, %%mm0				\n\t" //

  1953. 		"pmaxub %%mm1, %%mm0				\n\t" //

  1954. 		"pminub %%mm0, %%mm2				\n\t"

  1955. 		"movq %%mm2, (%%edx)				\n\t"

  1956. 

  1957. 		"movq (%%edx, %1, 2), %%mm2			\n\t" //

  1958. 		"movq (%0, %1, 8), %%mm0			\n\t" //

  1959. 		"movq %%mm2, %%mm3				\n\t"

  1960. 		"pmaxub %%mm0, %%mm2				\n\t" //

  1961. 		"pminub %%mm3, %%mm0				\n\t" //

  1962. 		"pmaxub %%mm1, %%mm0				\n\t" //

  1963. 		"pminub %%mm0, %%mm2				\n\t"

  1964. 		"movq %%mm2, (%%edx, %1, 2)			\n\t"

  1965. 

  1966. 

  1967. 		: : "r" (src), "r" (stride)

  1968. 		: "%eax", "%edx"

  1969. 	);

  1970. 

  1971. #else // MMX without MMX2

  1972. 	asm volatile(

  1973. 		"leal (%0, %1), %%eax				\n\t"

  1974. 		"leal (%%eax, %1, 4), %%edx			\n\t"

  1975. //	0	1	2	3	4	5	6	7	8	9

  1976. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1

  1977. 		"pxor %%mm7, %%mm7				\n\t"

  1978. 

  1979. #define MEDIAN(a,b,c)\

  1980. 		"movq " #a ", %%mm0				\n\t"\

  1981. 		"movq " #b ", %%mm2				\n\t"\

  1982. 		"movq " #c ", %%mm1				\n\t"\

  1983. 		"movq %%mm0, %%mm3				\n\t"\

  1984. 		"movq %%mm1, %%mm4				\n\t"\

  1985. 		"movq %%mm2, %%mm5				\n\t"\

  1986. 		"psubusb %%mm1, %%mm3				\n\t"\

  1987. 		"psubusb %%mm2, %%mm4				\n\t"\

  1988. 		"psubusb %%mm0, %%mm5				\n\t"\

  1989. 		"pcmpeqb %%mm7, %%mm3				\n\t"\

  1990. 		"pcmpeqb %%mm7, %%mm4				\n\t"\

  1991. 		"pcmpeqb %%mm7, %%mm5				\n\t"\

  1992. 		"movq %%mm3, %%mm6				\n\t"\

  1993. 		"pxor %%mm4, %%mm3				\n\t"\

  1994. 		"pxor %%mm5, %%mm4				\n\t"\

  1995. 		"pxor %%mm6, %%mm5				\n\t"\

  1996. 		"por %%mm3, %%mm1				\n\t"\

  1997. 		"por %%mm4, %%mm2				\n\t"\

  1998. 		"por %%mm5, %%mm0				\n\t"\

  1999. 		"pand %%mm2, %%mm0				\n\t"\

  2000. 		"pand %%mm1, %%mm0				\n\t"\

  2001. 		"movq %%mm0, " #b "				\n\t"

  2002. 

  2003. MEDIAN((%0), (%%eax), (%%eax, %1))

  2004. MEDIAN((%%eax, %1), (%%eax, %1, 2), (%0, %1, 4))

  2005. MEDIAN((%0, %1, 4), (%%edx), (%%edx, %1))

  2006. MEDIAN((%%edx, %1), (%%edx, %1, 2), (%0, %1, 8))

  2007. 

  2008. 		: : "r" (src), "r" (stride)

  2009. 		: "%eax", "%edx"

  2010. 	);

  2011. #endif // MMX

  2012. #else

  2013. 	int x, y;

  2014. 	src+= 4*stride;

  2015. 	// FIXME - there should be a way to do a few columns in parallel like w/mmx

  2016. 	for(x=0; x<8; x++)

  2017. 	{

  2018. 		uint8_t *colsrc = src;

  2019. 		for (y=0; y<4; y++)

  2020. 		{

  2021. 			int a, b, c, d, e, f;

  2022. 			a = colsrc[0       ];

  2023. 			b = colsrc[stride  ];

  2024. 			c = colsrc[stride*2];

  2025. 			d = (a-b)>>31;

  2026. 			e = (b-c)>>31;

  2027. 			f = (c-a)>>31;

  2028. 			colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));

  2029. 			colsrc += stride*2;

  2030. 		}

  2031. 		src++;

  2032. 	}

  2033. #endif

  2034. }

  2035. 

  2036. #ifdef HAVE_MMX

  2037. /**

  2038.  * transposes and shift the given 8x8 Block into dst1 and dst2

  2039.  */

  2040. static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)

  2041. {

  2042. 	asm(

  2043. 		"leal (%0, %1), %%eax				\n\t"

  2044. //	0	1	2	3	4	5	6	7	8	9

  2045. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1

  2046. 		"movq (%0), %%mm0		\n\t" // 12345678

  2047. 		"movq (%%eax), %%mm1		\n\t" // abcdefgh

  2048. 		"movq %%mm0, %%mm2		\n\t" // 12345678

  2049. 		"punpcklbw %%mm1, %%mm0		\n\t" // 1a2b3c4d

  2050. 		"punpckhbw %%mm1, %%mm2		\n\t" // 5e6f7g8h

  2051. 

  2052. 		"movq (%%eax, %1), %%mm1	\n\t"

  2053. 		"movq (%%eax, %1, 2), %%mm3	\n\t"

  2054. 		"movq %%mm1, %%mm4		\n\t"

  2055. 		"punpcklbw %%mm3, %%mm1		\n\t"

  2056. 		"punpckhbw %%mm3, %%mm4		\n\t"

  2057. 

  2058. 		"movq %%mm0, %%mm3		\n\t"

  2059. 		"punpcklwd %%mm1, %%mm0		\n\t"

  2060. 		"punpckhwd %%mm1, %%mm3		\n\t"

  2061. 		"movq %%mm2, %%mm1		\n\t"

  2062. 		"punpcklwd %%mm4, %%mm2		\n\t"

  2063. 		"punpckhwd %%mm4, %%mm1		\n\t"

  2064. 

  2065. 		"movd %%mm0, 128(%2)		\n\t"

  2066. 		"psrlq $32, %%mm0		\n\t"

  2067. 		"movd %%mm0, 144(%2)		\n\t"

  2068. 		"movd %%mm3, 160(%2)		\n\t"

  2069. 		"psrlq $32, %%mm3		\n\t"

  2070. 		"movd %%mm3, 176(%2)		\n\t"

  2071. 		"movd %%mm3, 48(%3)		\n\t"

  2072. 		"movd %%mm2, 192(%2)		\n\t"

  2073. 		"movd %%mm2, 64(%3)		\n\t"

  2074. 		"psrlq $32, %%mm2		\n\t"

  2075. 		"movd %%mm2, 80(%3)		\n\t"

  2076. 		"movd %%mm1, 96(%3)		\n\t"

  2077. 		"psrlq $32, %%mm1		\n\t"

  2078. 		"movd %%mm1, 112(%3)		\n\t"

  2079. 

  2080. 		"leal (%%eax, %1, 4), %%eax	\n\t"

  2081. 		

  2082. 		"movq (%0, %1, 4), %%mm0	\n\t" // 12345678

  2083. 		"movq (%%eax), %%mm1		\n\t" // abcdefgh

  2084. 		"movq %%mm0, %%mm2		\n\t" // 12345678

  2085. 		"punpcklbw %%mm1, %%mm0		\n\t" // 1a2b3c4d

  2086. 		"punpckhbw %%mm1, %%mm2		\n\t" // 5e6f7g8h

  2087. 

  2088. 		"movq (%%eax, %1), %%mm1	\n\t"

  2089. 		"movq (%%eax, %1, 2), %%mm3	\n\t"

  2090. 		"movq %%mm1, %%mm4		\n\t"

  2091. 		"punpcklbw %%mm3, %%mm1		\n\t"

  2092. 		"punpckhbw %%mm3, %%mm4		\n\t"

  2093. 

  2094. 		"movq %%mm0, %%mm3		\n\t"

  2095. 		"punpcklwd %%mm1, %%mm0		\n\t"

  2096. 		"punpckhwd %%mm1, %%mm3		\n\t"

  2097. 		"movq %%mm2, %%mm1		\n\t"

  2098. 		"punpcklwd %%mm4, %%mm2		\n\t"

  2099. 		"punpckhwd %%mm4, %%mm1		\n\t"

  2100. 

  2101. 		"movd %%mm0, 132(%2)		\n\t"

  2102. 		"psrlq $32, %%mm0		\n\t"

  2103. 		"movd %%mm0, 148(%2)		\n\t"

  2104. 		"movd %%mm3, 164(%2)		\n\t"

  2105. 		"psrlq $32, %%mm3		\n\t"

  2106. 		"movd %%mm3, 180(%2)		\n\t"

  2107. 		"movd %%mm3, 52(%3)		\n\t"

  2108. 		"movd %%mm2, 196(%2)		\n\t"

  2109. 		"movd %%mm2, 68(%3)		\n\t"

  2110. 		"psrlq $32, %%mm2		\n\t"

  2111. 		"movd %%mm2, 84(%3)		\n\t"

  2112. 		"movd %%mm1, 100(%3)		\n\t"

  2113. 		"psrlq $32, %%mm1		\n\t"

  2114. 		"movd %%mm1, 116(%3)		\n\t"

  2115. 

  2116. 

  2117. 	:: "r" (src), "r" (srcStride), "r" (dst1), "r" (dst2)

  2118. 	: "%eax"

  2119. 	);

  2120. }

  2121. 

  2122. /**

  2123.  * transposes the given 8x8 block

  2124.  */

  2125. static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)

  2126. {

  2127. 	asm(

  2128. 		"leal (%0, %1), %%eax				\n\t"

  2129. 		"leal (%%eax, %1, 4), %%edx			\n\t"

  2130. //	0	1	2	3	4	5	6	7	8	9

  2131. //	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1

  2132. 		"movq (%2), %%mm0		\n\t" // 12345678

  2133. 		"movq 16(%2), %%mm1		\n\t" // abcdefgh

  2134. 		"movq %%mm0, %%mm2		\n\t" // 12345678

  2135. 		"punpcklbw %%mm1, %%mm0		\n\t" // 1a2b3c4d

  2136. 		"punpckhbw %%mm1, %%mm2		\n\t" // 5e6f7g8h

  2137. 

  2138. 		"movq 32(%2), %%mm1		\n\t"

  2139. 		"movq 48(%2), %%mm3		\n\t"

  2140. 		"movq %%mm1, %%mm4		\n\t"

  2141. 		"punpcklbw %%mm3, %%mm1		\n\t"

  2142. 		"punpckhbw %%mm3, %%mm4		\n\t"

  2143. 

  2144. 		"movq %%mm0, %%mm3		\n\t"

  2145. 		"punpcklwd %%mm1, %%mm0		\n\t"

  2146. 		"punpckhwd %%mm1, %%mm3		\n\t"

  2147. 		"movq %%mm2, %%mm1		\n\t"

  2148. 		"punpcklwd %%mm4, %%mm2		\n\t"

  2149. 		"punpckhwd %%mm4, %%mm1		\n\t"

  2150. 

  2151. 		"movd %%mm0, (%0)		\n\t"

  2152. 		"psrlq $32, %%mm0		\n\t"

  2153. 		"movd %%mm0, (%%eax)		\n\t"

  2154. 		"movd %%mm3, (%%eax, %1)	\n\t"

  2155. 		"psrlq $32, %%mm3		\n\t"

  2156. 		"movd %%mm3, (%%eax, %1, 2)	\n\t"

  2157. 		"movd %%mm2, (%0, %1, 4)	\n\t"

  2158. 		"psrlq $32, %%mm2		\n\t"

  2159. 		"movd %%mm2, (%%edx)		\n\t"

  2160. 		"movd %%mm1, (%%edx, %1)	\n\t"

  2161. 		"psrlq $32, %%mm1		\n\t"

  2162. 		"movd %%mm1, (%%edx, %1, 2)	\n\t"

  2163. 

  2164. 

  2165. 		"movq 64(%2), %%mm0		\n\t" // 12345678

  2166. 		"movq 80(%2), %%mm1		\n\t" // abcdefgh

  2167. 		"movq %%mm0, %%mm2		\n\t" // 12345678

  2168. 		"punpcklbw %%mm1, %%mm0		\n\t" // 1a2b3c4d

  2169. 		"punpckhbw %%mm1, %%mm2		\n\t" // 5e6f7g8h

  2170. 

  2171. 		"movq 96(%2), %%mm1		\n\t"

  2172. 		"movq 112(%2), %%mm3		\n\t"

  2173. 		"movq %%mm1, %%mm4		\n\t"

  2174. 		"punpcklbw %%mm3, %%mm1		\n\t"

  2175. 		"punpckhbw %%mm3, %%mm4		\n\t"

  2176. 

  2177. 		"movq %%mm0, %%mm3		\n\t"

  2178. 		"punpcklwd %%mm1, %%mm0		\n\t"

  2179. 		"punpckhwd %%mm1, %%mm3		\n\t"

  2180. 		"movq %%mm2, %%mm1		\n\t"

  2181. 		"punpcklwd %%mm4, %%mm2		\n\t"

  2182. 		"punpckhwd %%mm4, %%mm1		\n\t"

  2183. 

  2184. 		"movd %%mm0, 4(%0)		\n\t"

  2185. 		"psrlq $32, %%mm0		\n\t"

  2186. 		"movd %%mm0, 4(%%eax)		\n\t"

  2187. 		"movd %%mm3, 4(%%eax, %1)	\n\t"

  2188. 		"psrlq $32, %%mm3		\n\t"

  2189. 		"movd %%mm3, 4(%%eax, %1, 2)	\n\t"

  2190. 		"movd %%mm2, 4(%0, %1, 4)	\n\t"

  2191. 		"psrlq $32, %%mm2		\n\t"

  2192. 		"movd %%mm2, 4(%%edx)		\n\t"

  2193. 		"movd %%mm1, 4(%%edx, %1)	\n\t"

  2194. 		"psrlq $32, %%mm1		\n\t"

  2195. 		"movd %%mm1, 4(%%edx, %1, 2)	\n\t"

  2196. 

  2197. 	:: "r" (dst), "r" (dstStride), "r" (src)

  2198. 	: "%eax", "%edx"

  2199. 	);

  2200. }

  2201. #endif

  2202. //static int test=0;

  2203. 

  2204. static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,

  2205. 				    uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)

  2206. {

  2207. 	// to save a register (FIXME do this outside of the loops)

  2208. 	tempBluredPast[127]= maxNoise[0];

  2209. 	tempBluredPast[128]= maxNoise[1];

  2210. 	tempBluredPast[129]= maxNoise[2];

  2211.         

  2212. #define FAST_L2_DIFF

  2213. //#define L1_DIFF //u should change the thresholds too if u try that one

  2214. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  2215. 	asm volatile(

  2216. 		"leal (%2, %2, 2), %%eax			\n\t" // 3*stride

  2217. 		"leal (%2, %2, 4), %%edx			\n\t" // 5*stride

  2218. 		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride

  2219. //	0	1	2	3	4	5	6	7	8	9

  2220. //	%x	%x+%2	%x+2%2	%x+eax	%x+4%2	%x+edx	%x+2eax	%x+ecx	%x+8%2

  2221. //FIXME reorder?

  2222. #ifdef L1_DIFF //needs mmx2

  2223. 		"movq (%0), %%mm0				\n\t" // L0

  2224. 		"psadbw (%1), %%mm0				\n\t" // |L0-R0|

  2225. 		"movq (%0, %2), %%mm1				\n\t" // L1

  2226. 		"psadbw (%1, %2), %%mm1				\n\t" // |L1-R1|

  2227. 		"movq (%0, %2, 2), %%mm2			\n\t" // L2

  2228. 		"psadbw (%1, %2, 2), %%mm2			\n\t" // |L2-R2|

  2229. 		"movq (%0, %%eax), %%mm3			\n\t" // L3

  2230. 		"psadbw (%1, %%eax), %%mm3			\n\t" // |L3-R3|

  2231. 

  2232. 		"movq (%0, %2, 4), %%mm4			\n\t" // L4

  2233. 		"paddw %%mm1, %%mm0				\n\t"

  2234. 		"psadbw (%1, %2, 4), %%mm4			\n\t" // |L4-R4|

  2235. 		"movq (%0, %%edx), %%mm5			\n\t" // L5

  2236. 		"paddw %%mm2, %%mm0				\n\t"

  2237. 		"psadbw (%1, %%edx), %%mm5			\n\t" // |L5-R5|

  2238. 		"movq (%0, %%eax, 2), %%mm6			\n\t" // L6

  2239. 		"paddw %%mm3, %%mm0				\n\t"

  2240. 		"psadbw (%1, %%eax, 2), %%mm6			\n\t" // |L6-R6|

  2241. 		"movq (%0, %%ecx), %%mm7			\n\t" // L7

  2242. 		"paddw %%mm4, %%mm0				\n\t"

  2243. 		"psadbw (%1, %%ecx), %%mm7			\n\t" // |L7-R7|

  2244. 		"paddw %%mm5, %%mm6				\n\t"

  2245. 		"paddw %%mm7, %%mm6				\n\t"

  2246. 		"paddw %%mm6, %%mm0				\n\t"

  2247. #elif defined (FAST_L2_DIFF)

  2248. 		"pcmpeqb %%mm7, %%mm7				\n\t"

  2249. 		"movq "MANGLE(b80)", %%mm6			\n\t"

  2250. 		"pxor %%mm0, %%mm0				\n\t"

  2251. #define L2_DIFF_CORE(a, b)\

  2252. 		"movq " #a ", %%mm5				\n\t"\

  2253. 		"movq " #b ", %%mm2				\n\t"\

  2254. 		"pxor %%mm7, %%mm2				\n\t"\

  2255. 		PAVGB(%%mm2, %%mm5)\

  2256. 		"paddb %%mm6, %%mm5				\n\t"\

  2257. 		"movq %%mm5, %%mm2				\n\t"\

  2258. 		"psllw $8, %%mm5				\n\t"\

  2259. 		"pmaddwd %%mm5, %%mm5				\n\t"\

  2260. 		"pmaddwd %%mm2, %%mm2				\n\t"\

  2261. 		"paddd %%mm2, %%mm5				\n\t"\

  2262. 		"psrld $14, %%mm5				\n\t"\

  2263. 		"paddd %%mm5, %%mm0				\n\t"

  2264. 

  2265. L2_DIFF_CORE((%0), (%1))

  2266. L2_DIFF_CORE((%0, %2), (%1, %2))

  2267. L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2))

  2268. L2_DIFF_CORE((%0, %%eax), (%1, %%eax))

  2269. L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4))

  2270. L2_DIFF_CORE((%0, %%edx), (%1, %%edx))

  2271. L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2))

  2272. L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx))

  2273. 

  2274. #else

  2275. 		"pxor %%mm7, %%mm7				\n\t"

  2276. 		"pxor %%mm0, %%mm0				\n\t"

  2277. #define L2_DIFF_CORE(a, b)\

  2278. 		"movq " #a ", %%mm5				\n\t"\

  2279. 		"movq " #b ", %%mm2				\n\t"\

  2280. 		"movq %%mm5, %%mm1				\n\t"\

  2281. 		"movq %%mm2, %%mm3				\n\t"\

  2282. 		"punpcklbw %%mm7, %%mm5				\n\t"\

  2283. 		"punpckhbw %%mm7, %%mm1				\n\t"\

  2284. 		"punpcklbw %%mm7, %%mm2				\n\t"\

  2285. 		"punpckhbw %%mm7, %%mm3				\n\t"\

  2286. 		"psubw %%mm2, %%mm5				\n\t"\

  2287. 		"psubw %%mm3, %%mm1				\n\t"\

  2288. 		"pmaddwd %%mm5, %%mm5				\n\t"\

  2289. 		"pmaddwd %%mm1, %%mm1				\n\t"\

  2290. 		"paddd %%mm1, %%mm5				\n\t"\

  2291. 		"paddd %%mm5, %%mm0				\n\t"

  2292. 

  2293. L2_DIFF_CORE((%0), (%1))

  2294. L2_DIFF_CORE((%0, %2), (%1, %2))

  2295. L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2))

  2296. L2_DIFF_CORE((%0, %%eax), (%1, %%eax))

  2297. L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4))

  2298. L2_DIFF_CORE((%0, %%edx), (%1, %%edx))

  2299. L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2))

  2300. L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx))

  2301. 

  2302. #endif

  2303. 

  2304. 		"movq %%mm0, %%mm4				\n\t"

  2305. 		"psrlq $32, %%mm0				\n\t"

  2306. 		"paddd %%mm0, %%mm4				\n\t"

  2307. 		"movd %%mm4, %%ecx				\n\t"

  2308. 		"shll $2, %%ecx					\n\t"

  2309. 		"movl %3, %%edx					\n\t"

  2310. 		"addl -4(%%edx), %%ecx				\n\t"

  2311. 		"addl 4(%%edx), %%ecx				\n\t"

  2312. 		"addl -1024(%%edx), %%ecx			\n\t"

  2313. 		"addl $4, %%ecx					\n\t"

  2314. 		"addl 1024(%%edx), %%ecx			\n\t"

  2315. 		"shrl $3, %%ecx					\n\t"

  2316. 		"movl %%ecx, (%%edx)				\n\t"

  2317. 

  2318. //		"movl %3, %%ecx					\n\t"

  2319. //		"movl %%ecx, test				\n\t"

  2320. //		"jmp 4f \n\t"

  2321. 		"cmpl 512(%%edx), %%ecx				\n\t"

  2322. 		" jb 2f						\n\t"

  2323. 		"cmpl 516(%%edx), %%ecx				\n\t"

  2324. 		" jb 1f						\n\t"

  2325. 

  2326. 		"leal (%%eax, %2, 2), %%edx			\n\t" // 5*stride

  2327. 		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride

  2328. 		"movq (%0), %%mm0				\n\t" // L0

  2329. 		"movq (%0, %2), %%mm1				\n\t" // L1

  2330. 		"movq (%0, %2, 2), %%mm2			\n\t" // L2

  2331. 		"movq (%0, %%eax), %%mm3			\n\t" // L3

  2332. 		"movq (%0, %2, 4), %%mm4			\n\t" // L4

  2333. 		"movq (%0, %%edx), %%mm5			\n\t" // L5

  2334. 		"movq (%0, %%eax, 2), %%mm6			\n\t" // L6

  2335. 		"movq (%0, %%ecx), %%mm7			\n\t" // L7

  2336. 		"movq %%mm0, (%1)				\n\t" // L0

  2337. 		"movq %%mm1, (%1, %2)				\n\t" // L1

  2338. 		"movq %%mm2, (%1, %2, 2)			\n\t" // L2

  2339. 		"movq %%mm3, (%1, %%eax)			\n\t" // L3

  2340. 		"movq %%mm4, (%1, %2, 4)			\n\t" // L4

  2341. 		"movq %%mm5, (%1, %%edx)			\n\t" // L5

  2342. 		"movq %%mm6, (%1, %%eax, 2)			\n\t" // L6

  2343. 		"movq %%mm7, (%1, %%ecx)			\n\t" // L7

  2344. 		"jmp 4f						\n\t"

  2345. 

  2346. 		"1:						\n\t"

  2347. 		"leal (%%eax, %2, 2), %%edx			\n\t" // 5*stride

  2348. 		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride

  2349. 		"movq (%0), %%mm0				\n\t" // L0

  2350. 		PAVGB((%1), %%mm0)				      // L0

  2351. 		"movq (%0, %2), %%mm1				\n\t" // L1

  2352. 		PAVGB((%1, %2), %%mm1)				      // L1

  2353. 		"movq (%0, %2, 2), %%mm2			\n\t" // L2

  2354. 		PAVGB((%1, %2, 2), %%mm2)			      // L2

  2355. 		"movq (%0, %%eax), %%mm3			\n\t" // L3

  2356. 		PAVGB((%1, %%eax), %%mm3)			      // L3

  2357. 		"movq (%0, %2, 4), %%mm4			\n\t" // L4

  2358. 		PAVGB((%1, %2, 4), %%mm4)			      // L4

  2359. 		"movq (%0, %%edx), %%mm5			\n\t" // L5

  2360. 		PAVGB((%1, %%edx), %%mm5)			      // L5

  2361. 		"movq (%0, %%eax, 2), %%mm6			\n\t" // L6

  2362. 		PAVGB((%1, %%eax, 2), %%mm6)			      // L6

  2363. 		"movq (%0, %%ecx), %%mm7			\n\t" // L7

  2364. 		PAVGB((%1, %%ecx), %%mm7)			      // L7

  2365. 		"movq %%mm0, (%1)				\n\t" // R0

  2366. 		"movq %%mm1, (%1, %2)				\n\t" // R1

  2367. 		"movq %%mm2, (%1, %2, 2)			\n\t" // R2

  2368. 		"movq %%mm3, (%1, %%eax)			\n\t" // R3

  2369. 		"movq %%mm4, (%1, %2, 4)			\n\t" // R4

  2370. 		"movq %%mm5, (%1, %%edx)			\n\t" // R5

  2371. 		"movq %%mm6, (%1, %%eax, 2)			\n\t" // R6

  2372. 		"movq %%mm7, (%1, %%ecx)			\n\t" // R7

  2373. 		"movq %%mm0, (%0)				\n\t" // L0

  2374. 		"movq %%mm1, (%0, %2)				\n\t" // L1

  2375. 		"movq %%mm2, (%0, %2, 2)			\n\t" // L2

  2376. 		"movq %%mm3, (%0, %%eax)			\n\t" // L3

  2377. 		"movq %%mm4, (%0, %2, 4)			\n\t" // L4

  2378. 		"movq %%mm5, (%0, %%edx)			\n\t" // L5

  2379. 		"movq %%mm6, (%0, %%eax, 2)			\n\t" // L6

  2380. 		"movq %%mm7, (%0, %%ecx)			\n\t" // L7

  2381. 		"jmp 4f						\n\t"

  2382. 

  2383. 		"2:						\n\t"

  2384. 		"cmpl 508(%%edx), %%ecx				\n\t"

  2385. 		" jb 3f						\n\t"

  2386. 

  2387. 		"leal (%%eax, %2, 2), %%edx			\n\t" // 5*stride

  2388. 		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride

  2389. 		"movq (%0), %%mm0				\n\t" // L0

  2390. 		"movq (%0, %2), %%mm1				\n\t" // L1

  2391. 		"movq (%0, %2, 2), %%mm2			\n\t" // L2

  2392. 		"movq (%0, %%eax), %%mm3			\n\t" // L3

  2393. 		"movq (%1), %%mm4				\n\t" // R0

  2394. 		"movq (%1, %2), %%mm5				\n\t" // R1

  2395. 		"movq (%1, %2, 2), %%mm6			\n\t" // R2

  2396. 		"movq (%1, %%eax), %%mm7			\n\t" // R3

  2397. 		PAVGB(%%mm4, %%mm0)

  2398. 		PAVGB(%%mm5, %%mm1)

  2399. 		PAVGB(%%mm6, %%mm2)

  2400. 		PAVGB(%%mm7, %%mm3)

  2401. 		PAVGB(%%mm4, %%mm0)

  2402. 		PAVGB(%%mm5, %%mm1)

  2403. 		PAVGB(%%mm6, %%mm2)

  2404. 		PAVGB(%%mm7, %%mm3)

  2405. 		"movq %%mm0, (%1)				\n\t" // R0

  2406. 		"movq %%mm1, (%1, %2)				\n\t" // R1

  2407. 		"movq %%mm2, (%1, %2, 2)			\n\t" // R2

  2408. 		"movq %%mm3, (%1, %%eax)			\n\t" // R3

  2409. 		"movq %%mm0, (%0)				\n\t" // L0

  2410. 		"movq %%mm1, (%0, %2)				\n\t" // L1

  2411. 		"movq %%mm2, (%0, %2, 2)			\n\t" // L2

  2412. 		"movq %%mm3, (%0, %%eax)			\n\t" // L3

  2413. 

  2414. 		"movq (%0, %2, 4), %%mm0			\n\t" // L4

  2415. 		"movq (%0, %%edx), %%mm1			\n\t" // L5

  2416. 		"movq (%0, %%eax, 2), %%mm2			\n\t" // L6

  2417. 		"movq (%0, %%ecx), %%mm3			\n\t" // L7

  2418. 		"movq (%1, %2, 4), %%mm4			\n\t" // R4

  2419. 		"movq (%1, %%edx), %%mm5			\n\t" // R5

  2420. 		"movq (%1, %%eax, 2), %%mm6			\n\t" // R6

  2421. 		"movq (%1, %%ecx), %%mm7			\n\t" // R7

  2422. 		PAVGB(%%mm4, %%mm0)

  2423. 		PAVGB(%%mm5, %%mm1)

  2424. 		PAVGB(%%mm6, %%mm2)

  2425. 		PAVGB(%%mm7, %%mm3)

  2426. 		PAVGB(%%mm4, %%mm0)

  2427. 		PAVGB(%%mm5, %%mm1)

  2428. 		PAVGB(%%mm6, %%mm2)

  2429. 		PAVGB(%%mm7, %%mm3)

  2430. 		"movq %%mm0, (%1, %2, 4)			\n\t" // R4

  2431. 		"movq %%mm1, (%1, %%edx)			\n\t" // R5

  2432. 		"movq %%mm2, (%1, %%eax, 2)			\n\t" // R6

  2433. 		"movq %%mm3, (%1, %%ecx)			\n\t" // R7

  2434. 		"movq %%mm0, (%0, %2, 4)			\n\t" // L4

  2435. 		"movq %%mm1, (%0, %%edx)			\n\t" // L5

  2436. 		"movq %%mm2, (%0, %%eax, 2)			\n\t" // L6

  2437. 		"movq %%mm3, (%0, %%ecx)			\n\t" // L7

  2438. 		"jmp 4f						\n\t"

  2439. 

  2440. 		"3:						\n\t"

  2441. 		"leal (%%eax, %2, 2), %%edx			\n\t" // 5*stride

  2442. 		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride

  2443. 		"movq (%0), %%mm0				\n\t" // L0

  2444. 		"movq (%0, %2), %%mm1				\n\t" // L1

  2445. 		"movq (%0, %2, 2), %%mm2			\n\t" // L2

  2446. 		"movq (%0, %%eax), %%mm3			\n\t" // L3

  2447. 		"movq (%1), %%mm4				\n\t" // R0

  2448. 		"movq (%1, %2), %%mm5				\n\t" // R1

  2449. 		"movq (%1, %2, 2), %%mm6			\n\t" // R2

  2450. 		"movq (%1, %%eax), %%mm7			\n\t" // R3

  2451. 		PAVGB(%%mm4, %%mm0)

  2452. 		PAVGB(%%mm5, %%mm1)

  2453. 		PAVGB(%%mm6, %%mm2)

  2454. 		PAVGB(%%mm7, %%mm3)

  2455. 		PAVGB(%%mm4, %%mm0)

  2456. 		PAVGB(%%mm5, %%mm1)

  2457. 		PAVGB(%%mm6, %%mm2)

  2458. 		PAVGB(%%mm7, %%mm3)

  2459. 		PAVGB(%%mm4, %%mm0)

  2460. 		PAVGB(%%mm5, %%mm1)

  2461. 		PAVGB(%%mm6, %%mm2)

  2462. 		PAVGB(%%mm7, %%mm3)

  2463. 		"movq %%mm0, (%1)				\n\t" // R0

  2464. 		"movq %%mm1, (%1, %2)				\n\t" // R1

  2465. 		"movq %%mm2, (%1, %2, 2)			\n\t" // R2

  2466. 		"movq %%mm3, (%1, %%eax)			\n\t" // R3

  2467. 		"movq %%mm0, (%0)				\n\t" // L0

  2468. 		"movq %%mm1, (%0, %2)				\n\t" // L1

  2469. 		"movq %%mm2, (%0, %2, 2)			\n\t" // L2

  2470. 		"movq %%mm3, (%0, %%eax)			\n\t" // L3

  2471. 

  2472. 		"movq (%0, %2, 4), %%mm0			\n\t" // L4

  2473. 		"movq (%0, %%edx), %%mm1			\n\t" // L5

  2474. 		"movq (%0, %%eax, 2), %%mm2			\n\t" // L6

  2475. 		"movq (%0, %%ecx), %%mm3			\n\t" // L7

  2476. 		"movq (%1, %2, 4), %%mm4			\n\t" // R4

  2477. 		"movq (%1, %%edx), %%mm5			\n\t" // R5

  2478. 		"movq (%1, %%eax, 2), %%mm6			\n\t" // R6

  2479. 		"movq (%1, %%ecx), %%mm7			\n\t" // R7

  2480. 		PAVGB(%%mm4, %%mm0)

  2481. 		PAVGB(%%mm5, %%mm1)

  2482. 		PAVGB(%%mm6, %%mm2)

  2483. 		PAVGB(%%mm7, %%mm3)

  2484. 		PAVGB(%%mm4, %%mm0)

  2485. 		PAVGB(%%mm5, %%mm1)

  2486. 		PAVGB(%%mm6, %%mm2)

  2487. 		PAVGB(%%mm7, %%mm3)

  2488. 		PAVGB(%%mm4, %%mm0)

  2489. 		PAVGB(%%mm5, %%mm1)

  2490. 		PAVGB(%%mm6, %%mm2)

  2491. 		PAVGB(%%mm7, %%mm3)

  2492. 		"movq %%mm0, (%1, %2, 4)			\n\t" // R4

  2493. 		"movq %%mm1, (%1, %%edx)			\n\t" // R5

  2494. 		"movq %%mm2, (%1, %%eax, 2)			\n\t" // R6

  2495. 		"movq %%mm3, (%1, %%ecx)			\n\t" // R7

  2496. 		"movq %%mm0, (%0, %2, 4)			\n\t" // L4

  2497. 		"movq %%mm1, (%0, %%edx)			\n\t" // L5

  2498. 		"movq %%mm2, (%0, %%eax, 2)			\n\t" // L6

  2499. 		"movq %%mm3, (%0, %%ecx)			\n\t" // L7

  2500. 

  2501. 		"4:						\n\t"

  2502. 

  2503. 		:: "r" (src), "r" (tempBlured), "r"(stride), "m" (tempBluredPast)

  2504. 		: "%eax", "%edx", "%ecx", "memory"

  2505. 		);

  2506. //printf("%d\n", test);

  2507. #else

  2508. {

  2509. 	int y;

  2510. 	int d=0;

  2511. 	int sysd=0;

  2512. 	int i;

  2513. 

  2514. 	for(y=0; y<8; y++)

  2515. 	{

  2516. 		int x;

  2517. 		for(x=0; x<8; x++)

  2518. 		{

  2519. 			int ref= tempBlured[ x + y*stride ];

  2520. 			int cur= src[ x + y*stride ];

  2521. 			int d1=ref - cur;

  2522. //			if(x==0 || x==7) d1+= d1>>1;

  2523. //			if(y==0 || y==7) d1+= d1>>1;

  2524. //			d+= ABS(d1);

  2525. 			d+= d1*d1;

  2526. 			sysd+= d1;

  2527. 		}

  2528. 	}

  2529. 	i=d;

  2530. 	d= 	(

  2531. 		4*d

  2532. 		+(*(tempBluredPast-256))

  2533. 		+(*(tempBluredPast-1))+ (*(tempBluredPast+1))

  2534. 		+(*(tempBluredPast+256))

  2535. 		+4)>>3;

  2536. 	*tempBluredPast=i;

  2537. //	((*tempBluredPast)*3 + d + 2)>>2;

  2538. 

  2539. //printf("%d %d %d\n", maxNoise[0], maxNoise[1], maxNoise[2]);

  2540. /*

  2541. Switch between

  2542.  1  0  0  0  0  0  0  (0)

  2543. 64 32 16  8  4  2  1  (1)

  2544. 64 48 36 27 20 15 11 (33) (approx)

  2545. 64 56 49 43 37 33 29 (200) (approx)

  2546. */

  2547. 	if(d > maxNoise[1])

  2548. 	{

  2549. 		if(d < maxNoise[2])

  2550. 		{

  2551. 			for(y=0; y<8; y++)

  2552. 			{

  2553. 				int x;

  2554. 				for(x=0; x<8; x++)

  2555. 				{

  2556. 					int ref= tempBlured[ x + y*stride ];

  2557. 					int cur= src[ x + y*stride ];

  2558. 					tempBlured[ x + y*stride ]=

  2559. 					src[ x + y*stride ]=

  2560. 						(ref + cur + 1)>>1;

  2561. 				}

  2562. 			}

  2563. 		}

  2564. 		else

  2565. 		{

  2566. 			for(y=0; y<8; y++)

  2567. 			{

  2568. 				int x;

  2569. 				for(x=0; x<8; x++)

  2570. 				{

  2571. 					tempBlured[ x + y*stride ]= src[ x + y*stride ];

  2572. 				}

  2573. 			}

  2574. 		}

  2575. 	}

  2576. 	else

  2577. 	{

  2578. 		if(d < maxNoise[0])

  2579. 		{

  2580. 			for(y=0; y<8; y++)

  2581. 			{

  2582. 				int x;

  2583. 				for(x=0; x<8; x++)

  2584. 				{

  2585. 					int ref= tempBlured[ x + y*stride ];

  2586. 					int cur= src[ x + y*stride ];

  2587. 					tempBlured[ x + y*stride ]=

  2588. 					src[ x + y*stride ]=

  2589. 						(ref*7 + cur + 4)>>3;

  2590. 				}

  2591. 			}

  2592. 		}

  2593. 		else

  2594. 		{

  2595. 			for(y=0; y<8; y++)

  2596. 			{

  2597. 				int x;

  2598. 				for(x=0; x<8; x++)

  2599. 				{

  2600. 					int ref= tempBlured[ x + y*stride ];

  2601. 					int cur= src[ x + y*stride ];

  2602. 					tempBlured[ x + y*stride ]=

  2603. 					src[ x + y*stride ]=

  2604. 						(ref*3 + cur + 2)>>2;

  2605. 				}

  2606. 			}

  2607. 		}

  2608. 	}

  2609. }

  2610. #endif

  2611. }

  2612. 

  2613. static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  2614. 	QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);

  2615. 

  2616. /**

  2617.  * Copies a block from src to dst and fixes the blacklevel

  2618.  * levelFix == 0 -> dont touch the brighness & contrast

  2619.  */

  2620. #undef SCALED_CPY

  2621. 

  2622. static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,

  2623. 	int levelFix, int64_t *packedOffsetAndScale)

  2624. {

  2625. #ifndef HAVE_MMX

  2626. 	int i;

  2627. #endif

  2628. 	if(levelFix)

  2629. 	{

  2630. #ifdef HAVE_MMX

  2631. 					asm volatile(

  2632. 						"movq (%%eax), %%mm2	\n\t" // packedYOffset

  2633. 						"movq 8(%%eax), %%mm3	\n\t" // packedYScale

  2634. 						"leal (%2,%4), %%eax	\n\t"

  2635. 						"leal (%3,%5), %%edx	\n\t"

  2636. 						"pxor %%mm4, %%mm4	\n\t"

  2637. #ifdef HAVE_MMX2

  2638. #define SCALED_CPY(src1, src2, dst1, dst2)					\

  2639. 						"movq " #src1 ", %%mm0	\n\t"\

  2640. 						"movq " #src1 ", %%mm5	\n\t"\

  2641. 						"movq " #src2 ", %%mm1	\n\t"\

  2642. 						"movq " #src2 ", %%mm6	\n\t"\

  2643. 						"punpcklbw %%mm0, %%mm0 \n\t"\

  2644. 						"punpckhbw %%mm5, %%mm5 \n\t"\

  2645. 						"punpcklbw %%mm1, %%mm1 \n\t"\

  2646. 						"punpckhbw %%mm6, %%mm6 \n\t"\

  2647. 						"pmulhuw %%mm3, %%mm0	\n\t"\

  2648. 						"pmulhuw %%mm3, %%mm5	\n\t"\

  2649. 						"pmulhuw %%mm3, %%mm1	\n\t"\

  2650. 						"pmulhuw %%mm3, %%mm6	\n\t"\

  2651. 						"psubw %%mm2, %%mm0	\n\t"\

  2652. 						"psubw %%mm2, %%mm5	\n\t"\

  2653. 						"psubw %%mm2, %%mm1	\n\t"\

  2654. 						"psubw %%mm2, %%mm6	\n\t"\

  2655. 						"packuswb %%mm5, %%mm0	\n\t"\

  2656. 						"packuswb %%mm6, %%mm1	\n\t"\

  2657. 						"movq %%mm0, " #dst1 "	\n\t"\

  2658. 						"movq %%mm1, " #dst2 "	\n\t"\

  2659. 

  2660. #else //HAVE_MMX2

  2661. #define SCALED_CPY(src1, src2, dst1, dst2)					\

  2662. 						"movq " #src1 ", %%mm0	\n\t"\

  2663. 						"movq " #src1 ", %%mm5	\n\t"\

  2664. 						"punpcklbw %%mm4, %%mm0 \n\t"\

  2665. 						"punpckhbw %%mm4, %%mm5 \n\t"\

  2666. 						"psubw %%mm2, %%mm0	\n\t"\

  2667. 						"psubw %%mm2, %%mm5	\n\t"\

  2668. 						"movq " #src2 ", %%mm1	\n\t"\

  2669. 						"psllw $6, %%mm0	\n\t"\

  2670. 						"psllw $6, %%mm5	\n\t"\

  2671. 						"pmulhw %%mm3, %%mm0	\n\t"\

  2672. 						"movq " #src2 ", %%mm6	\n\t"\

  2673. 						"pmulhw %%mm3, %%mm5	\n\t"\

  2674. 						"punpcklbw %%mm4, %%mm1 \n\t"\

  2675. 						"punpckhbw %%mm4, %%mm6 \n\t"\

  2676. 						"psubw %%mm2, %%mm1	\n\t"\

  2677. 						"psubw %%mm2, %%mm6	\n\t"\

  2678. 						"psllw $6, %%mm1	\n\t"\

  2679. 						"psllw $6, %%mm6	\n\t"\

  2680. 						"pmulhw %%mm3, %%mm1	\n\t"\

  2681. 						"pmulhw %%mm3, %%mm6	\n\t"\

  2682. 						"packuswb %%mm5, %%mm0	\n\t"\

  2683. 						"packuswb %%mm6, %%mm1	\n\t"\

  2684. 						"movq %%mm0, " #dst1 "	\n\t"\

  2685. 						"movq %%mm1, " #dst2 "	\n\t"\

  2686. 

  2687. #endif //!HAVE_MMX2

  2688. 

  2689. SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))

  2690. SCALED_CPY((%2, %4, 2), (%%eax, %4, 2), (%3, %5, 2), (%%edx, %5, 2))

  2691. SCALED_CPY((%2, %4, 4), (%%eax, %4, 4), (%3, %5, 4), (%%edx, %5, 4))

  2692. 						"leal (%%eax,%4,4), %%eax	\n\t"

  2693. 						"leal (%%edx,%5,4), %%edx	\n\t"

  2694. SCALED_CPY((%%eax, %4), (%%eax, %4, 2), (%%edx, %5), (%%edx, %5, 2))

  2695. 

  2696. 

  2697. 						: "=&a" (packedOffsetAndScale)

  2698. 						: "0" (packedOffsetAndScale),

  2699. 						"r"(src),

  2700. 						"r"(dst),

  2701. 						"r" (srcStride),

  2702. 						"r" (dstStride)

  2703. 						: "%edx"

  2704. 					);

  2705. #else

  2706. 				for(i=0; i<8; i++)

  2707. 					memcpy(	&(dst[dstStride*i]),

  2708. 						&(src[srcStride*i]), BLOCK_SIZE);

  2709. #endif

  2710. 	}

  2711. 	else

  2712. 	{

  2713. #ifdef HAVE_MMX

  2714. 					asm volatile(

  2715. 						"leal (%0,%2), %%eax	\n\t"

  2716. 						"leal (%1,%3), %%edx	\n\t"

  2717. 

  2718. #define SIMPLE_CPY(src1, src2, dst1, dst2)				\

  2719. 						"movq " #src1 ", %%mm0	\n\t"\

  2720. 						"movq " #src2 ", %%mm1	\n\t"\

  2721. 						"movq %%mm0, " #dst1 "	\n\t"\

  2722. 						"movq %%mm1, " #dst2 "	\n\t"\

  2723. 

  2724. SIMPLE_CPY((%0)       , (%0, %2)      , (%1)       , (%1, %3))

  2725. SIMPLE_CPY((%0, %2, 2), (%%eax, %2, 2), (%1, %3, 2), (%%edx, %3, 2))

  2726. SIMPLE_CPY((%0, %2, 4), (%%eax, %2, 4), (%1, %3, 4), (%%edx, %3, 4))

  2727. 						"leal (%%eax,%2,4), %%eax	\n\t"

  2728. 						"leal (%%edx,%3,4), %%edx	\n\t"

  2729. SIMPLE_CPY((%%eax, %2), (%%eax, %2, 2), (%%edx, %3), (%%edx, %3, 2))

  2730. 

  2731. 						: : "r" (src),

  2732. 						"r" (dst),

  2733. 						"r" (srcStride),

  2734. 						"r" (dstStride)

  2735. 						: "%eax", "%edx"

  2736. 					);

  2737. #else

  2738. 				for(i=0; i<8; i++)

  2739. 					memcpy(	&(dst[dstStride*i]),

  2740. 						&(src[srcStride*i]), BLOCK_SIZE);

  2741. #endif

  2742. 	}

  2743. }

  2744. 

  2745. /**

  2746.  * Duplicates the given 8 src pixels ? times upward

  2747.  */

  2748. static inline void RENAME(duplicate)(uint8_t src[], int stride)

  2749. {

  2750. #ifdef HAVE_MMX

  2751. 	asm volatile(

  2752. 		"movq (%0), %%mm0		\n\t"

  2753. 		"addl %1, %0			\n\t"

  2754. 		"movq %%mm0, (%0)		\n\t"

  2755. 		"movq %%mm0, (%0, %1)		\n\t"

  2756. 		"movq %%mm0, (%0, %1, 2)	\n\t"

  2757. 		: "+r" (src)

  2758. 		: "r" (-stride)

  2759. 	);

  2760. #else

  2761. 	int i;

  2762. 	uint8_t *p=src;

  2763. 	for(i=0; i<3; i++)

  2764. 	{

  2765. 		p-= stride;

  2766. 		memcpy(p, src, 8);

  2767. 	}

  2768. #endif

  2769. }

  2770. 

  2771. /**

  2772.  * Filters array of bytes (Y or U or V values)

  2773.  */

  2774. static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  2775. 	QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)

  2776. {

  2777. 	PPContext __attribute__((aligned(8))) c= *c2; //copy to stack for faster access

  2778. 	int x,y;

  2779. #ifdef COMPILE_TIME_MODE

  2780. 	const int mode= COMPILE_TIME_MODE;

  2781. #else

  2782. 	const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;

  2783. #endif

  2784. 	int black=0, white=255; // blackest black and whitest white in the picture

  2785. 	int QPCorrecture= 256*256;

  2786. 

  2787. 	int copyAhead;

  2788. #ifdef HAVE_MMX

  2789. 	int i;

  2790. #endif

  2791. 

  2792. 	const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;

  2793. 	const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;

  2794. 

  2795. 	//FIXME remove

  2796. 	uint64_t * const yHistogram= c.yHistogram;

  2797. 	uint8_t * const tempSrc= c.tempSrc;

  2798. 	uint8_t * const tempDst= c.tempDst;

  2799. 	//const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;

  2800. 

  2801. #ifdef HAVE_MMX

  2802. 	for(i=0; i<57; i++){

  2803. 		int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;

  2804. 		int threshold= offset*2 + 1;

  2805. 		c.mmxDcOffset[i]= 0x7F - offset;

  2806. 		c.mmxDcThreshold[i]= 0x7F - threshold;

  2807. 		c.mmxDcOffset[i]*= 0x0101010101010101LL;

  2808. 		c.mmxDcThreshold[i]*= 0x0101010101010101LL;

  2809. 	}

  2810. #endif

  2811. 

  2812. 	if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;

  2813. 	else if(   (mode & LINEAR_BLEND_DEINT_FILTER)

  2814. 		|| (mode & FFMPEG_DEINT_FILTER)

  2815. 		|| (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;

  2816. 	else if(   (mode & V_DEBLOCK)

  2817. 		|| (mode & LINEAR_IPOL_DEINT_FILTER)

  2818. 		|| (mode & MEDIAN_DEINT_FILTER)

  2819. 		|| (mode & V_A_DEBLOCK)) copyAhead=13;

  2820. 	else if(mode & V_X1_FILTER) copyAhead=11;

  2821. //	else if(mode & V_RK1_FILTER) copyAhead=10;

  2822. 	else if(mode & DERING) copyAhead=9;

  2823. 	else copyAhead=8;

  2824. 

  2825. 	copyAhead-= 8;

  2826. 

  2827. 	if(!isColor)

  2828. 	{

  2829. 		uint64_t sum= 0;

  2830. 		int i;

  2831. 		uint64_t maxClipped;

  2832. 		uint64_t clipped;

  2833. 		double scale;

  2834. 

  2835. 		c.frameNum++;

  2836. 		// first frame is fscked so we ignore it

  2837. 		if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;

  2838. 

  2839. 		for(i=0; i<256; i++)

  2840. 		{

  2841. 			sum+= yHistogram[i];

  2842. //			printf("%d ", yHistogram[i]);

  2843. 		}

  2844. //		printf("\n\n");

  2845. 

  2846. 		/* we allways get a completly black picture first */

  2847. 		maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);

  2848. 

  2849. 		clipped= sum;

  2850. 		for(black=255; black>0; black--)

  2851. 		{

  2852. 			if(clipped < maxClipped) break;

  2853. 			clipped-= yHistogram[black];

  2854. 		}

  2855. 

  2856. 		clipped= sum;

  2857. 		for(white=0; white<256; white++)

  2858. 		{

  2859. 			if(clipped < maxClipped) break;

  2860. 			clipped-= yHistogram[white];

  2861. 		}

  2862. 

  2863. 		scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);

  2864. 

  2865. #ifdef HAVE_MMX2

  2866. 		c.packedYScale= (uint16_t)(scale*256.0 + 0.5);

  2867. 		c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;

  2868. #else

  2869. 		c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);

  2870. 		c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;

  2871. #endif

  2872. 

  2873. 		c.packedYOffset|= c.packedYOffset<<32;

  2874. 		c.packedYOffset|= c.packedYOffset<<16;

  2875. 

  2876. 		c.packedYScale|= c.packedYScale<<32;

  2877. 		c.packedYScale|= c.packedYScale<<16;

  2878. 		

  2879. 		if(mode & LEVEL_FIX)	QPCorrecture= (int)(scale*256*256 + 0.5);

  2880. 		else			QPCorrecture= 256*256;

  2881. 	}

  2882. 	else

  2883. 	{

  2884. 		c.packedYScale= 0x0100010001000100LL;

  2885. 		c.packedYOffset= 0;

  2886. 		QPCorrecture= 256*256;

  2887. 	}

  2888. 

  2889. 	/* copy & deinterlace first row of blocks */

  2890. 	y=-BLOCK_SIZE;

  2891. 	{

  2892. 		uint8_t *srcBlock= &(src[y*srcStride]);

  2893. 		uint8_t *dstBlock= tempDst + dstStride;

  2894. 

  2895. 		// From this point on it is guranteed that we can read and write 16 lines downward

  2896. 		// finish 1 block before the next otherwise we´ll might have a problem

  2897. 		// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing

  2898. 		for(x=0; x<width; x+=BLOCK_SIZE)

  2899. 		{

  2900. 

  2901. #ifdef HAVE_MMX2

  2902. /*

  2903. 			prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);

  2904. 			prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);

  2905. 			prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);

  2906. 			prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);

  2907. */

  2908. 

  2909. 			asm(

  2910. 				"movl %4, %%eax			\n\t"

  2911. 				"shrl $2, %%eax			\n\t"

  2912. 				"andl $6, %%eax			\n\t"

  2913. 				"addl %5, %%eax			\n\t"

  2914. 				"movl %%eax, %%edx		\n\t"

  2915. 				"imul %1, %%eax			\n\t"

  2916. 				"imul %3, %%edx			\n\t"

  2917. 				"prefetchnta 32(%%eax, %0)	\n\t"

  2918. 				"prefetcht0 32(%%edx, %2)	\n\t"

  2919. 				"addl %1, %%eax			\n\t"

  2920. 				"addl %3, %%edx			\n\t"

  2921. 				"prefetchnta 32(%%eax, %0)	\n\t"

  2922. 				"prefetcht0 32(%%edx, %2)	\n\t"

  2923. 			:: "r" (srcBlock), "r" (srcStride), "r" (dstBlock), "r" (dstStride),

  2924. 			"m" (x), "m" (copyAhead)

  2925. 			: "%eax", "%edx"

  2926. 			);

  2927. 

  2928. #elif defined(HAVE_3DNOW)

  2929. //FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...

  2930. /*			prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);

  2931. 			prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);

  2932. 			prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);

  2933. 			prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);

  2934. */

  2935. #endif

  2936. 

  2937. 			RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,

  2938. 				srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

  2939. 

  2940. 			RENAME(duplicate)(dstBlock + dstStride*8, dstStride);

  2941. 

  2942. 			if(mode & LINEAR_IPOL_DEINT_FILTER)

  2943. 				RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);

  2944. 			else if(mode & LINEAR_BLEND_DEINT_FILTER)

  2945. 				RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);

  2946. 			else if(mode & MEDIAN_DEINT_FILTER)

  2947. 				RENAME(deInterlaceMedian)(dstBlock, dstStride);

  2948. 			else if(mode & CUBIC_IPOL_DEINT_FILTER)

  2949. 				RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);

  2950. 			else if(mode & FFMPEG_DEINT_FILTER)

  2951. 				RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);

  2952. 			else if(mode & LOWPASS5_DEINT_FILTER)

  2953. 				RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);

  2954. /*			else if(mode & CUBIC_BLEND_DEINT_FILTER)

  2955. 				RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);

  2956. */

  2957. 			dstBlock+=8;

  2958. 			srcBlock+=8;

  2959. 		}

  2960. 		if(width==dstStride)

  2961. 			memcpy(dst, tempDst + 9*dstStride, copyAhead*dstStride);

  2962. 		else

  2963. 		{

  2964. 			int i;

  2965. 			for(i=0; i<copyAhead; i++)

  2966. 			{

  2967. 				memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);

  2968. 			}

  2969. 		}

  2970. 	}

  2971. 

  2972. //printf("\n");

  2973. 	for(y=0; y<height; y+=BLOCK_SIZE)

  2974. 	{

  2975. 		//1% speedup if these are here instead of the inner loop

  2976. 		uint8_t *srcBlock= &(src[y*srcStride]);

  2977. 		uint8_t *dstBlock= &(dst[y*dstStride]);

  2978. #ifdef HAVE_MMX

  2979. 		uint8_t *tempBlock1= c.tempBlocks;

  2980. 		uint8_t *tempBlock2= c.tempBlocks + 8;

  2981. #endif

  2982. 		int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];

  2983. 		int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*QPStride];

  2984. 		int QP=0;

  2985. 		/* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards

  2986. 		   if not than use a temporary buffer */

  2987. 		if(y+15 >= height)

  2988. 		{

  2989. 			int i;

  2990. 			/* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with

  2991. 			   blockcopy to dst later */

  2992. 			memcpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,

  2993. 				srcStride*MAX(height-y-copyAhead, 0) );

  2994. 

  2995. 			/* duplicate last line of src to fill the void upto line (copyAhead+7) */

  2996. 			for(i=MAX(height-y, 8); i<copyAhead+8; i++)

  2997. 				memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), srcStride);

  2998. 

  2999. 			/* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/

  3000. 			memcpy(tempDst, dstBlock - dstStride, dstStride*MIN(height-y+1, copyAhead+1) );

  3001. 

  3002. 			/* duplicate last line of dst to fill the void upto line (copyAhead) */

  3003. 			for(i=height-y+1; i<=copyAhead; i++)

  3004. 				memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), dstStride);

  3005. 

  3006. 			dstBlock= tempDst + dstStride;

  3007. 			srcBlock= tempSrc;

  3008. 		}

  3009. //printf("\n");

  3010. 

  3011. 		// From this point on it is guranteed that we can read and write 16 lines downward

  3012. 		// finish 1 block before the next otherwise we´ll might have a problem

  3013. 		// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing

  3014. 		for(x=0; x<width; x+=BLOCK_SIZE)

  3015. 		{

  3016. 			const int stride= dstStride;

  3017. #ifdef HAVE_MMX

  3018. 			uint8_t *tmpXchg;

  3019. #endif

  3020. 			if(isColor)

  3021. 			{

  3022. 				QP= QPptr[x>>qpHShift];

  3023. 				c.nonBQP= nonBQPptr[x>>qpHShift];

  3024. 			}

  3025. 			else

  3026. 			{

  3027. 				QP= QPptr[x>>4];

  3028. 				QP= (QP* QPCorrecture + 256*128)>>16;

  3029. 				c.nonBQP= nonBQPptr[x>>4];

  3030. 				c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;

  3031. 				yHistogram[ srcBlock[srcStride*12 + 4] ]++;

  3032. 			}

  3033. 			c.QP= QP;

  3034. #ifdef HAVE_MMX

  3035. 			asm volatile(

  3036. 				"movd %1, %%mm7					\n\t"

  3037. 				"packuswb %%mm7, %%mm7				\n\t" // 0, 0, 0, QP, 0, 0, 0, QP

  3038. 				"packuswb %%mm7, %%mm7				\n\t" // 0,QP, 0, QP, 0,QP, 0, QP

  3039. 				"packuswb %%mm7, %%mm7				\n\t" // QP,..., QP

  3040. 				"movq %%mm7, %0			\n\t"

  3041. 				: "=m" (c.pQPb) 

  3042. 				: "r" (QP)

  3043. 			);

  3044. #endif

  3045. 

  3046. 

  3047. #ifdef HAVE_MMX2

  3048. /*

  3049. 			prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);

  3050. 			prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);

  3051. 			prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);

  3052. 			prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);

  3053. */

  3054. 

  3055. 			asm(

  3056. 				"movl %4, %%eax			\n\t"

  3057. 				"shrl $2, %%eax			\n\t"

  3058. 				"andl $6, %%eax			\n\t"

  3059. 				"addl %5, %%eax			\n\t"

  3060. 				"movl %%eax, %%edx		\n\t"

  3061. 				"imul %1, %%eax			\n\t"

  3062. 				"imul %3, %%edx			\n\t"

  3063. 				"prefetchnta 32(%%eax, %0)	\n\t"

  3064. 				"prefetcht0 32(%%edx, %2)	\n\t"

  3065. 				"addl %1, %%eax			\n\t"

  3066. 				"addl %3, %%edx			\n\t"

  3067. 				"prefetchnta 32(%%eax, %0)	\n\t"

  3068. 				"prefetcht0 32(%%edx, %2)	\n\t"

  3069. 			:: "r" (srcBlock), "r" (srcStride), "r" (dstBlock), "r" (dstStride),

  3070. 			"m" (x), "m" (copyAhead)

  3071. 			: "%eax", "%edx"

  3072. 			);

  3073. 

  3074. #elif defined(HAVE_3DNOW)

  3075. //FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...

  3076. /*			prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);

  3077. 			prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);

  3078. 			prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);

  3079. 			prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);

  3080. */

  3081. #endif

  3082. 

  3083. 			RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,

  3084. 				srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

  3085. 

  3086. 			if(mode & LINEAR_IPOL_DEINT_FILTER)

  3087. 				RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);

  3088. 			else if(mode & LINEAR_BLEND_DEINT_FILTER)

  3089. 				RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);

  3090. 			else if(mode & MEDIAN_DEINT_FILTER)

  3091. 				RENAME(deInterlaceMedian)(dstBlock, dstStride);

  3092. 			else if(mode & CUBIC_IPOL_DEINT_FILTER)

  3093. 				RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);

  3094. 			else if(mode & FFMPEG_DEINT_FILTER)

  3095. 				RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);

  3096. 			else if(mode & LOWPASS5_DEINT_FILTER)

  3097. 				RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);

  3098. /*			else if(mode & CUBIC_BLEND_DEINT_FILTER)

  3099. 				RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);

  3100. */

  3101. 

  3102. 			/* only deblock if we have 2 blocks */

  3103. 			if(y + 8 < height)

  3104. 			{

  3105. 				if(mode & V_X1_FILTER)

  3106. 					RENAME(vertX1Filter)(dstBlock, stride, &c);

  3107. 				else if(mode & V_DEBLOCK)

  3108. 				{

  3109. 					const int t= RENAME(vertClassify)(dstBlock, stride, &c);

  3110. 

  3111. 					if(t==1)

  3112. 						RENAME(doVertLowPass)(dstBlock, stride, &c);

  3113. 					else if(t==2)

  3114. 						RENAME(doVertDefFilter)(dstBlock, stride, &c);

  3115. 				}else if(mode & V_A_DEBLOCK){

  3116. 					do_a_deblock(dstBlock, stride, 1, &c);

  3117. 				}

  3118. 			}

  3119. 

  3120. #ifdef HAVE_MMX

  3121. 			RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);

  3122. #endif

  3123. 			/* check if we have a previous block to deblock it with dstBlock */

  3124. 			if(x - 8 >= 0)

  3125. 			{

  3126. #ifdef HAVE_MMX

  3127. 				if(mode & H_X1_FILTER)

  3128. 					RENAME(vertX1Filter)(tempBlock1, 16, &c);

  3129. 				else if(mode & H_DEBLOCK)

  3130. 				{

  3131. //START_TIMER

  3132. 					const int t= RENAME(vertClassify)(tempBlock1, 16, &c);

  3133. //STOP_TIMER("dc & minmax")

  3134.                                         if(t==1)

  3135. 						RENAME(doVertLowPass)(tempBlock1, 16, &c);

  3136. 					else if(t==2)

  3137. 						RENAME(doVertDefFilter)(tempBlock1, 16, &c);

  3138. 				}else if(mode & H_A_DEBLOCK){

  3139. 					do_a_deblock(tempBlock1, 16, 1, &c);

  3140. 				}

  3141. 

  3142. 				RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);

  3143. 

  3144. #else

  3145. 				if(mode & H_X1_FILTER)

  3146. 					horizX1Filter(dstBlock-4, stride, QP);

  3147. 				else if(mode & H_DEBLOCK)

  3148. 				{

  3149. 					const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);

  3150. 

  3151. 					if(t==1)

  3152. 						RENAME(doHorizLowPass)(dstBlock-4, stride, &c);

  3153. 					else if(t==2)

  3154. 						RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);

  3155. 				}else if(mode & H_A_DEBLOCK){

  3156. 					do_a_deblock(dstBlock-8, 1, stride, &c);

  3157. 				}

  3158. #endif

  3159. 				if(mode & DERING)

  3160. 				{

  3161. 				//FIXME filter first line

  3162. 					if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);

  3163. 				}

  3164. 

  3165. 				if(mode & TEMP_NOISE_FILTER)

  3166. 				{

  3167. 					RENAME(tempNoiseReducer)(dstBlock-8, stride,

  3168. 						c.tempBlured[isColor] + y*dstStride + x,

  3169. 						c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),

  3170. 						c.ppMode.maxTmpNoise);

  3171. 				}

  3172. 			}

  3173. 

  3174. 			dstBlock+=8;

  3175. 			srcBlock+=8;

  3176. 

  3177. #ifdef HAVE_MMX

  3178. 			tmpXchg= tempBlock1;

  3179. 			tempBlock1= tempBlock2;

  3180. 			tempBlock2 = tmpXchg;

  3181. #endif

  3182. 		}

  3183. 

  3184. 		if(mode & DERING)

  3185. 		{

  3186. 				if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);

  3187. 		}

  3188. 

  3189. 		if((mode & TEMP_NOISE_FILTER))

  3190. 		{

  3191. 			RENAME(tempNoiseReducer)(dstBlock-8, dstStride,

  3192. 				c.tempBlured[isColor] + y*dstStride + x,

  3193. 				c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),

  3194. 				c.ppMode.maxTmpNoise);

  3195. 		}

  3196. 

  3197. 		/* did we use a tmp buffer for the last lines*/

  3198. 		if(y+15 >= height)

  3199. 		{

  3200. 			uint8_t *dstBlock= &(dst[y*dstStride]);

  3201. 			if(width==dstStride)

  3202. 				memcpy(dstBlock, tempDst + dstStride, dstStride*(height-y));

  3203. 			else

  3204. 			{

  3205. 				int i;

  3206. 				for(i=0; i<height-y; i++)

  3207. 				{

  3208. 					memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);

  3209. 				}

  3210. 			}

  3211. 		}

  3212. /*

  3213. 		for(x=0; x<width; x+=32)

  3214. 		{

  3215. 			volatile int i;

  3216. 			i+=	+ dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]

  3217. 				+ dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]

  3218. 				+ dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];

  3219. //				+ dstBlock[x +13*dstStride]

  3220. //				+ dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];

  3221. 		}*/

  3222. 	}

  3223. #ifdef HAVE_3DNOW

  3224. 	asm volatile("femms");

  3225. #elif defined (HAVE_MMX)

  3226. 	asm volatile("emms");

  3227. #endif

  3228. 

  3229. #ifdef DEBUG_BRIGHTNESS

  3230. 	if(!isColor)

  3231. 	{

  3232. 		int max=1;

  3233. 		int i;

  3234. 		for(i=0; i<256; i++)

  3235. 			if(yHistogram[i] > max) max=yHistogram[i];

  3236. 

  3237. 		for(i=1; i<256; i++)

  3238. 		{

  3239. 			int x;

  3240. 			int start=yHistogram[i-1]/(max/256+1);

  3241. 			int end=yHistogram[i]/(max/256+1);

  3242. 			int inc= end > start ? 1 : -1;

  3243. 			for(x=start; x!=end+inc; x+=inc)

  3244. 				dst[ i*dstStride + x]+=128;

  3245. 		}

  3246. 

  3247. 		for(i=0; i<100; i+=2)

  3248. 		{

  3249. 			dst[ (white)*dstStride + i]+=128;

  3250. 			dst[ (black)*dstStride + i]+=128;

  3251. 		}

  3252. 

  3253. 	}

  3254. #endif

  3255. 

  3256. 	*c2= c; //copy local context back

  3257. 

  3258. }