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.
1155 Commits
32 Branches
283MB
Tree: d46aba2642
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'd46aba2642'
${ noResults }
FFmpeg/postproc/rgb2rgb_template.c

2410 lines
60KB
Raw Blame History 

  1. /*

  2.  *

  3.  *  rgb2rgb.c, Software RGB to RGB convertor

  4.  *  pluralize by Software PAL8 to RGB convertor

  5.  *               Software YUV to YUV convertor

  6.  *               Software YUV to RGB convertor

  7.  *  Written by Nick Kurshev.

  8.  *  palette & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)

  9.  */

  10. 

  11. #include <stddef.h>

  12. #include <inttypes.h> /* for __WORDSIZE */

  13. 

  14. #ifndef __WORDSIZE

  15. // #warning You have misconfigured system and probably will lose performance!

  16. #define __WORDSIZE MP_WORDSIZE

  17. #endif

  18. 

  19. #undef PREFETCH

  20. #undef MOVNTQ

  21. #undef EMMS

  22. #undef SFENCE

  23. #undef MMREG_SIZE

  24. #undef PREFETCHW

  25. #undef PAVGB

  26. 

  27. #ifdef HAVE_SSE2

  28. #define MMREG_SIZE 16

  29. #else

  30. #define MMREG_SIZE 8

  31. #endif

  32. 

  33. #ifdef HAVE_3DNOW

  34. #define PREFETCH  "prefetch"

  35. #define PREFETCHW "prefetchw"

  36. #define PAVGB	  "pavgusb"

  37. #elif defined ( HAVE_MMX2 )

  38. #define PREFETCH "prefetchnta"

  39. #define PREFETCHW "prefetcht0"

  40. #define PAVGB	  "pavgb"

  41. #else

  42. #define PREFETCH "/nop"

  43. #define PREFETCHW "/nop"

  44. #endif

  45. 

  46. #ifdef HAVE_3DNOW

  47. /* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */

  48. #define EMMS     "femms"

  49. #else

  50. #define EMMS     "emms"

  51. #endif

  52. 

  53. #ifdef HAVE_MMX2

  54. #define MOVNTQ "movntq"

  55. #define SFENCE "sfence"

  56. #else

  57. #define MOVNTQ "movq"

  58. #define SFENCE "/nop"

  59. #endif

  60. 

  61. static inline void RENAME(rgb24to32)(const uint8_t *src,uint8_t *dst,unsigned src_size)

  62. {

  63.   uint8_t *dest = dst;

  64.   const uint8_t *s = src;

  65.   const uint8_t *end;

  66. #ifdef HAVE_MMX

  67.   const uint8_t *mm_end;

  68. #endif

  69.   end = s + src_size;

  70. #ifdef HAVE_MMX

  71.   __asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");

  72.   mm_end = end - 23;

  73.   __asm __volatile("movq	%0, %%mm7"::"m"(mask32):"memory");

  74.   while(s < mm_end)

  75.   {

  76.     __asm __volatile(

  77. 	PREFETCH"	32%1\n\t"

  78. 	"movd	%1, %%mm0\n\t"

  79. 	"punpckldq 3%1, %%mm0\n\t"

  80. 	"movd	6%1, %%mm1\n\t"

  81. 	"punpckldq 9%1, %%mm1\n\t"

  82. 	"movd	12%1, %%mm2\n\t"

  83. 	"punpckldq 15%1, %%mm2\n\t"

  84. 	"movd	18%1, %%mm3\n\t"

  85. 	"punpckldq 21%1, %%mm3\n\t"

  86. 	"pand	%%mm7, %%mm0\n\t"

  87. 	"pand	%%mm7, %%mm1\n\t"

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

  89. 	"pand	%%mm7, %%mm3\n\t"

  90. 	MOVNTQ"	%%mm0, %0\n\t"

  91. 	MOVNTQ"	%%mm1, 8%0\n\t"

  92. 	MOVNTQ"	%%mm2, 16%0\n\t"

  93. 	MOVNTQ"	%%mm3, 24%0"

  94. 	:"=m"(*dest)

  95. 	:"m"(*s)

  96. 	:"memory");

  97.     dest += 32;

  98.     s += 24;

  99.   }

  100.   __asm __volatile(SFENCE:::"memory");

  101.   __asm __volatile(EMMS:::"memory");

  102. #endif

  103.   while(s < end)

  104.   {

  105.     *dest++ = *s++;

  106.     *dest++ = *s++;

  107.     *dest++ = *s++;

  108.     *dest++ = 0;

  109.   }

  110. }

  111. 

  112. static inline void RENAME(rgb32to24)(const uint8_t *src,uint8_t *dst,unsigned src_size)

  113. {

  114.   uint8_t *dest = dst;

  115.   const uint8_t *s = src;

  116.   const uint8_t *end;

  117. #ifdef HAVE_MMX

  118.   const uint8_t *mm_end;

  119. #endif

  120.   end = s + src_size;

  121. #ifdef HAVE_MMX

  122.   __asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");

  123.   mm_end = end - 31;

  124.   while(s < mm_end)

  125.   {

  126.     __asm __volatile(

  127. 	PREFETCH"	32%1\n\t"

  128. 	"movq	%1, %%mm0\n\t"

  129. 	"movq	8%1, %%mm1\n\t"

  130. 	"movq	16%1, %%mm4\n\t"

  131. 	"movq	24%1, %%mm5\n\t"

  132. 	"movq	%%mm0, %%mm2\n\t"

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

  134. 	"movq	%%mm4, %%mm6\n\t"

  135. 	"movq	%%mm5, %%mm7\n\t"

  136. 	"psrlq	$8, %%mm2\n\t"

  137. 	"psrlq	$8, %%mm3\n\t"

  138. 	"psrlq	$8, %%mm6\n\t"

  139. 	"psrlq	$8, %%mm7\n\t"

  140. 	"pand	%2, %%mm0\n\t"

  141. 	"pand	%2, %%mm1\n\t"

  142. 	"pand	%2, %%mm4\n\t"

  143. 	"pand	%2, %%mm5\n\t"

  144. 	"pand	%3, %%mm2\n\t"

  145. 	"pand	%3, %%mm3\n\t"

  146. 	"pand	%3, %%mm6\n\t"

  147. 	"pand	%3, %%mm7\n\t"

  148. 	"por	%%mm2, %%mm0\n\t"

  149. 	"por	%%mm3, %%mm1\n\t"

  150. 	"por	%%mm6, %%mm4\n\t"

  151. 	"por	%%mm7, %%mm5\n\t"

  152. 

  153. 	"movq	%%mm1, %%mm2\n\t"

  154. 	"movq	%%mm4, %%mm3\n\t"

  155. 	"psllq	$48, %%mm2\n\t"

  156. 	"psllq	$32, %%mm3\n\t"

  157. 	"pand	%4, %%mm2\n\t"

  158. 	"pand	%5, %%mm3\n\t"

  159. 	"por	%%mm2, %%mm0\n\t"

  160. 	"psrlq	$16, %%mm1\n\t"

  161. 	"psrlq	$32, %%mm4\n\t"

  162. 	"psllq	$16, %%mm5\n\t"

  163. 	"por	%%mm3, %%mm1\n\t"

  164. 	"pand	%6, %%mm5\n\t"

  165. 	"por	%%mm5, %%mm4\n\t"

  166. 

  167. 	MOVNTQ"	%%mm0, %0\n\t"

  168. 	MOVNTQ"	%%mm1, 8%0\n\t"

  169. 	MOVNTQ"	%%mm4, 16%0"

  170. 	:"=m"(*dest)

  171. 	:"m"(*s),"m"(mask24l),

  172. 	 "m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)

  173. 	:"memory");

  174.     dest += 24;

  175.     s += 32;

  176.   }

  177.   __asm __volatile(SFENCE:::"memory");

  178.   __asm __volatile(EMMS:::"memory");

  179. #endif

  180.   while(s < end)

  181.   {

  182.     *dest++ = *s++;

  183.     *dest++ = *s++;

  184.     *dest++ = *s++;

  185.     s++;

  186.   }

  187. }

  188. 

  189. /*

  190.  Original by Strepto/Astral

  191.  ported to gcc & bugfixed : A'rpi

  192.  MMX2, 3DNOW optimization by Nick Kurshev

  193.  32bit c version, and and&add trick by Michael Niedermayer

  194. */

  195. static inline void RENAME(rgb15to16)(const uint8_t *src,uint8_t *dst,unsigned src_size)

  196. {

  197.   register const uint8_t* s=src;

  198.   register uint8_t* d=dst;

  199.   register const uint8_t *end;

  200.   const uint8_t *mm_end;

  201.   end = s + src_size;

  202. #ifdef HAVE_MMX

  203.   __asm __volatile(PREFETCH"	%0"::"m"(*s));

  204.   __asm __volatile("movq	%0, %%mm4"::"m"(mask15s));

  205.   mm_end = end - 15;

  206.   while(s<mm_end)

  207.   {

  208. 	__asm __volatile(

  209. 		PREFETCH"	32%1\n\t"

  210. 		"movq	%1, %%mm0\n\t"

  211. 		"movq	8%1, %%mm2\n\t"

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

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

  214. 		"pand	%%mm4, %%mm0\n\t"

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

  216. 		"paddw	%%mm1, %%mm0\n\t"

  217. 		"paddw	%%mm3, %%mm2\n\t"

  218. 		MOVNTQ"	%%mm0, %0\n\t"

  219. 		MOVNTQ"	%%mm2, 8%0"

  220. 		:"=m"(*d)

  221. 		:"m"(*s)

  222. 		);

  223. 	d+=16;

  224. 	s+=16;

  225.   }

  226.   __asm __volatile(SFENCE:::"memory");

  227.   __asm __volatile(EMMS:::"memory");

  228. #endif

  229.     mm_end = end - 3;

  230.     while(s < mm_end)

  231.     {

  232. 	register unsigned x= *((uint32_t *)s);

  233. 	*((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);

  234. 	d+=4;

  235. 	s+=4;

  236.     }

  237.     if(s < end)

  238.     {

  239. 	register unsigned short x= *((uint16_t *)s);

  240. 	*((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);

  241.     }

  242. }

  243. 

  244. static inline void RENAME(bgr24torgb24)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  245. {

  246. 	unsigned j,i,num_pixels=src_size/3;

  247. 	for(i=0,j=0; j<num_pixels; i+=3,j+=3)

  248. 	{

  249. 		dst[j+0] = src[i+2];

  250. 		dst[j+1] = src[i+1];

  251. 		dst[j+2] = src[i+0];

  252. 	}

  253. }

  254. 

  255. static inline void RENAME(rgb16to15)(const uint8_t *src,uint8_t *dst,unsigned src_size)

  256. {

  257.   register const uint8_t* s=src;

  258.   register uint8_t* d=dst;

  259.   register const uint8_t *end;

  260.   const uint8_t *mm_end;

  261.   end = s + src_size;

  262. #ifdef HAVE_MMX

  263.   __asm __volatile(PREFETCH"	%0"::"m"(*s));

  264.   __asm __volatile("movq	%0, %%mm7"::"m"(mask15rg));

  265.   __asm __volatile("movq	%0, %%mm6"::"m"(mask15b));

  266.   mm_end = end - 15;

  267.   while(s<mm_end)

  268.   {

  269. 	__asm __volatile(

  270. 		PREFETCH"	32%1\n\t"

  271. 		"movq	%1, %%mm0\n\t"

  272. 		"movq	8%1, %%mm2\n\t"

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

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

  275. 		"psrlq	$1, %%mm0\n\t"

  276. 		"psrlq	$1, %%mm2\n\t"

  277. 		"pand	%%mm7, %%mm0\n\t"

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

  279. 		"pand	%%mm6, %%mm1\n\t"

  280. 		"pand	%%mm6, %%mm3\n\t"

  281. 		"por	%%mm1, %%mm0\n\t"

  282. 		"por	%%mm3, %%mm2\n\t"

  283. 		MOVNTQ"	%%mm0, %0\n\t"

  284. 		MOVNTQ"	%%mm2, 8%0"

  285. 		:"=m"(*d)

  286. 		:"m"(*s)

  287. 		);

  288. 	d+=16;

  289. 	s+=16;

  290.   }

  291.   __asm __volatile(SFENCE:::"memory");

  292.   __asm __volatile(EMMS:::"memory");

  293. #endif

  294.     mm_end = end - 3;

  295.     while(s < mm_end)

  296.     {

  297. 	register uint32_t x= *((uint32_t *)s);

  298. 	*((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F);

  299. 	s+=4;

  300. 	d+=4;

  301.     }

  302.     if(s < end)

  303.     {

  304. 	register uint16_t x= *((uint16_t *)s);

  305. 	*((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F);

  306. 	s+=2;

  307. 	d+=2;

  308.     }

  309. }

  310. 

  311. static inline void RENAME(rgb32to16)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  312. {

  313. 	const uint8_t *s = src;

  314. 	const uint8_t *end;

  315. #ifdef HAVE_MMX

  316. 	const uint8_t *mm_end;

  317. #endif

  318. 	uint16_t *d = (uint16_t *)dst;

  319. 	end = s + src_size;

  320. #ifdef HAVE_MMX

  321. 	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");

  322. 	__asm __volatile(

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

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

  325. 	    ::"m"(red_16mask),"m"(green_16mask));

  326. 	mm_end = end - 15;

  327. 	while(s < mm_end)

  328. 	{

  329. 	    __asm __volatile(

  330. 		PREFETCH" 32%1\n\t"

  331. 		"movd	%1, %%mm0\n\t"

  332. 		"movd	4%1, %%mm3\n\t"

  333. 		"punpckldq 8%1, %%mm0\n\t"

  334. 		"punpckldq 12%1, %%mm3\n\t"

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

  336. 		"movq	%%mm0, %%mm2\n\t"

  337. 		"movq	%%mm3, %%mm4\n\t"

  338. 		"movq	%%mm3, %%mm5\n\t"

  339. 		"psrlq	$3, %%mm0\n\t"

  340. 		"psrlq	$3, %%mm3\n\t"

  341. 		"pand	%2, %%mm0\n\t"

  342. 		"pand	%2, %%mm3\n\t"

  343. 		"psrlq	$5, %%mm1\n\t"

  344. 		"psrlq	$5, %%mm4\n\t"

  345. 		"pand	%%mm6, %%mm1\n\t"

  346. 		"pand	%%mm6, %%mm4\n\t"

  347. 		"psrlq	$8, %%mm2\n\t"

  348. 		"psrlq	$8, %%mm5\n\t"

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

  350. 		"pand	%%mm7, %%mm5\n\t"

  351. 		"por	%%mm1, %%mm0\n\t"

  352. 		"por	%%mm4, %%mm3\n\t"

  353. 		"por	%%mm2, %%mm0\n\t"

  354. 		"por	%%mm5, %%mm3\n\t"

  355. 		"psllq	$16, %%mm3\n\t"

  356. 		"por	%%mm3, %%mm0\n\t"

  357. 		MOVNTQ"	%%mm0, %0\n\t"

  358. 		:"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");

  359. 		d += 4;

  360. 		s += 16;

  361. 	}

  362. 	__asm __volatile(SFENCE:::"memory");

  363. 	__asm __volatile(EMMS:::"memory");

  364. #endif

  365. 	while(s < end)

  366. 	{

  367. 		const int b= *s++;

  368. 		const int g= *s++;

  369. 		const int r= *s++;

  370. 		*d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);

  371. 		s++;

  372. 	}

  373. }

  374. 

  375. static inline void RENAME(rgb32tobgr16)(const uint8_t *src, uint8_t *dst, unsigned int src_size)

  376. {

  377. 	const uint8_t *s = src;

  378. 	const uint8_t *end;

  379. #ifdef HAVE_MMX

  380. 	const uint8_t *mm_end;

  381. #endif

  382. 	uint16_t *d = (uint16_t *)dst;

  383. 	end = s + src_size;

  384. #ifdef HAVE_MMX

  385. 	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");

  386. 	__asm __volatile(

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

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

  389. 	    ::"m"(red_16mask),"m"(green_16mask));

  390. 	mm_end = end - 15;

  391. 	while(s < mm_end)

  392. 	{

  393. 	    __asm __volatile(

  394. 		PREFETCH" 32%1\n\t"

  395. 		"movd	%1, %%mm0\n\t"

  396. 		"movd	4%1, %%mm3\n\t"

  397. 		"punpckldq 8%1, %%mm0\n\t"

  398. 		"punpckldq 12%1, %%mm3\n\t"

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

  400. 		"movq	%%mm0, %%mm2\n\t"

  401. 		"movq	%%mm3, %%mm4\n\t"

  402. 		"movq	%%mm3, %%mm5\n\t"

  403. 		"psllq	$8, %%mm0\n\t"

  404. 		"psllq	$8, %%mm3\n\t"

  405. 		"pand	%%mm7, %%mm0\n\t"

  406. 		"pand	%%mm7, %%mm3\n\t"

  407. 		"psrlq	$5, %%mm1\n\t"

  408. 		"psrlq	$5, %%mm4\n\t"

  409. 		"pand	%%mm6, %%mm1\n\t"

  410. 		"pand	%%mm6, %%mm4\n\t"

  411. 		"psrlq	$19, %%mm2\n\t"

  412. 		"psrlq	$19, %%mm5\n\t"

  413. 		"pand	%2, %%mm2\n\t"

  414. 		"pand	%2, %%mm5\n\t"

  415. 		"por	%%mm1, %%mm0\n\t"

  416. 		"por	%%mm4, %%mm3\n\t"

  417. 		"por	%%mm2, %%mm0\n\t"

  418. 		"por	%%mm5, %%mm3\n\t"

  419. 		"psllq	$16, %%mm3\n\t"

  420. 		"por	%%mm3, %%mm0\n\t"

  421. 		MOVNTQ"	%%mm0, %0\n\t"

  422. 		:"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");

  423. 		d += 4;

  424. 		s += 16;

  425. 	}

  426. 	__asm __volatile(SFENCE:::"memory");

  427. 	__asm __volatile(EMMS:::"memory");

  428. #endif

  429. 	while(s < end)

  430. 	{

  431. 		const int r= *s++;

  432. 		const int g= *s++;

  433. 		const int b= *s++;

  434. 		*d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);

  435. 		s++;

  436. 	}

  437. }

  438. 

  439. static inline void RENAME(rgb32to15)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  440. {

  441. 	const uint8_t *s = src;

  442. 	const uint8_t *end;

  443. #ifdef HAVE_MMX

  444. 	const uint8_t *mm_end;

  445. #endif

  446. 	uint16_t *d = (uint16_t *)dst;

  447. 	end = s + src_size;

  448. #ifdef HAVE_MMX

  449. 	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");

  450. 	__asm __volatile(

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

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

  453. 	    ::"m"(red_15mask),"m"(green_15mask));

  454. 	mm_end = end - 15;

  455. 	while(s < mm_end)

  456. 	{

  457. 	    __asm __volatile(

  458. 		PREFETCH" 32%1\n\t"

  459. 		"movd	%1, %%mm0\n\t"

  460. 		"movd	4%1, %%mm3\n\t"

  461. 		"punpckldq 8%1, %%mm0\n\t"

  462. 		"punpckldq 12%1, %%mm3\n\t"

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

  464. 		"movq	%%mm0, %%mm2\n\t"

  465. 		"movq	%%mm3, %%mm4\n\t"

  466. 		"movq	%%mm3, %%mm5\n\t"

  467. 		"psrlq	$3, %%mm0\n\t"

  468. 		"psrlq	$3, %%mm3\n\t"

  469. 		"pand	%2, %%mm0\n\t"

  470. 		"pand	%2, %%mm3\n\t"

  471. 		"psrlq	$6, %%mm1\n\t"

  472. 		"psrlq	$6, %%mm4\n\t"

  473. 		"pand	%%mm6, %%mm1\n\t"

  474. 		"pand	%%mm6, %%mm4\n\t"

  475. 		"psrlq	$9, %%mm2\n\t"

  476. 		"psrlq	$9, %%mm5\n\t"

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

  478. 		"pand	%%mm7, %%mm5\n\t"

  479. 		"por	%%mm1, %%mm0\n\t"

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

  481. 		"por	%%mm2, %%mm0\n\t"

  482. 		"por	%%mm5, %%mm3\n\t"

  483. 		"psllq	$16, %%mm3\n\t"

  484. 		"por	%%mm3, %%mm0\n\t"

  485. 		MOVNTQ"	%%mm0, %0\n\t"

  486. 		:"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");

  487. 		d += 4;

  488. 		s += 16;

  489. 	}

  490. 	__asm __volatile(SFENCE:::"memory");

  491. 	__asm __volatile(EMMS:::"memory");

  492. #endif

  493. 	while(s < end)

  494. 	{

  495. 		const int b= *s++;

  496. 		const int g= *s++;

  497. 		const int r= *s++;

  498. 		*d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);

  499. 		s++;

  500. 	}

  501. }

  502. 

  503. static inline void RENAME(rgb32tobgr15)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  504. {

  505. 	const uint8_t *s = src;

  506. 	const uint8_t *end;

  507. #ifdef HAVE_MMX

  508. 	const uint8_t *mm_end;

  509. #endif

  510. 	uint16_t *d = (uint16_t *)dst;

  511. 	end = s + src_size;

  512. #ifdef HAVE_MMX

  513. 	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");

  514. 	__asm __volatile(

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

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

  517. 	    ::"m"(red_15mask),"m"(green_15mask));

  518. 	mm_end = end - 15;

  519. 	while(s < mm_end)

  520. 	{

  521. 	    __asm __volatile(

  522. 		PREFETCH" 32%1\n\t"

  523. 		"movd	%1, %%mm0\n\t"

  524. 		"movd	4%1, %%mm3\n\t"

  525. 		"punpckldq 8%1, %%mm0\n\t"

  526. 		"punpckldq 12%1, %%mm3\n\t"

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

  528. 		"movq	%%mm0, %%mm2\n\t"

  529. 		"movq	%%mm3, %%mm4\n\t"

  530. 		"movq	%%mm3, %%mm5\n\t"

  531. 		"psllq	$7, %%mm0\n\t"

  532. 		"psllq	$7, %%mm3\n\t"

  533. 		"pand	%%mm7, %%mm0\n\t"

  534. 		"pand	%%mm7, %%mm3\n\t"

  535. 		"psrlq	$6, %%mm1\n\t"

  536. 		"psrlq	$6, %%mm4\n\t"

  537. 		"pand	%%mm6, %%mm1\n\t"

  538. 		"pand	%%mm6, %%mm4\n\t"

  539. 		"psrlq	$19, %%mm2\n\t"

  540. 		"psrlq	$19, %%mm5\n\t"

  541. 		"pand	%2, %%mm2\n\t"

  542. 		"pand	%2, %%mm5\n\t"

  543. 		"por	%%mm1, %%mm0\n\t"

  544. 		"por	%%mm4, %%mm3\n\t"

  545. 		"por	%%mm2, %%mm0\n\t"

  546. 		"por	%%mm5, %%mm3\n\t"

  547. 		"psllq	$16, %%mm3\n\t"

  548. 		"por	%%mm3, %%mm0\n\t"

  549. 		MOVNTQ"	%%mm0, %0\n\t"

  550. 		:"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");

  551. 		d += 4;

  552. 		s += 16;

  553. 	}

  554. 	__asm __volatile(SFENCE:::"memory");

  555. 	__asm __volatile(EMMS:::"memory");

  556. #endif

  557. 	while(s < end)

  558. 	{

  559. 		const int r= *s++;

  560. 		const int g= *s++;

  561. 		const int b= *s++;

  562. 		*d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);

  563. 		s++;

  564. 	}

  565. }

  566. 

  567. static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  568. {

  569. 	const uint8_t *s = src;

  570. 	const uint8_t *end;

  571. #ifdef HAVE_MMX

  572. 	const uint8_t *mm_end;

  573. #endif

  574. 	uint16_t *d = (uint16_t *)dst;

  575. 	end = s + src_size;

  576. #ifdef HAVE_MMX

  577. 	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");

  578. 	__asm __volatile(

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

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

  581. 	    ::"m"(red_16mask),"m"(green_16mask));

  582. 	mm_end = end - 11;

  583. 	while(s < mm_end)

  584. 	{

  585. 	    __asm __volatile(

  586. 		PREFETCH" 32%1\n\t"

  587. 		"movd	%1, %%mm0\n\t"

  588. 		"movd	3%1, %%mm3\n\t"

  589. 		"punpckldq 6%1, %%mm0\n\t"

  590. 		"punpckldq 9%1, %%mm3\n\t"

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

  592. 		"movq	%%mm0, %%mm2\n\t"

  593. 		"movq	%%mm3, %%mm4\n\t"

  594. 		"movq	%%mm3, %%mm5\n\t"

  595. 		"psrlq	$3, %%mm0\n\t"

  596. 		"psrlq	$3, %%mm3\n\t"

  597. 		"pand	%2, %%mm0\n\t"

  598. 		"pand	%2, %%mm3\n\t"

  599. 		"psrlq	$5, %%mm1\n\t"

  600. 		"psrlq	$5, %%mm4\n\t"

  601. 		"pand	%%mm6, %%mm1\n\t"

  602. 		"pand	%%mm6, %%mm4\n\t"

  603. 		"psrlq	$8, %%mm2\n\t"

  604. 		"psrlq	$8, %%mm5\n\t"

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

  606. 		"pand	%%mm7, %%mm5\n\t"

  607. 		"por	%%mm1, %%mm0\n\t"

  608. 		"por	%%mm4, %%mm3\n\t"

  609. 		"por	%%mm2, %%mm0\n\t"

  610. 		"por	%%mm5, %%mm3\n\t"

  611. 		"psllq	$16, %%mm3\n\t"

  612. 		"por	%%mm3, %%mm0\n\t"

  613. 		MOVNTQ"	%%mm0, %0\n\t"

  614. 		:"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");

  615. 		d += 4;

  616. 		s += 12;

  617. 	}

  618. 	__asm __volatile(SFENCE:::"memory");

  619. 	__asm __volatile(EMMS:::"memory");

  620. #endif

  621. 	while(s < end)

  622. 	{

  623. 		const int b= *s++;

  624. 		const int g= *s++;

  625. 		const int r= *s++;

  626. 		*d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);

  627. 	}

  628. }

  629. 

  630. static inline void RENAME(rgb24tobgr16)(const uint8_t *src, uint8_t *dst, unsigned int src_size)

  631. {

  632. 	const uint8_t *s = src;

  633. 	const uint8_t *end;

  634. #ifdef HAVE_MMX

  635. 	const uint8_t *mm_end;

  636. #endif

  637. 	uint16_t *d = (uint16_t *)dst;

  638. 	end = s + src_size;

  639. #ifdef HAVE_MMX

  640. 	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");

  641. 	__asm __volatile(

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

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

  644. 	    ::"m"(red_16mask),"m"(green_16mask));

  645. 	mm_end = end - 15;

  646. 	while(s < mm_end)

  647. 	{

  648. 	    __asm __volatile(

  649. 		PREFETCH" 32%1\n\t"

  650. 		"movd	%1, %%mm0\n\t"

  651. 		"movd	3%1, %%mm3\n\t"

  652. 		"punpckldq 6%1, %%mm0\n\t"

  653. 		"punpckldq 9%1, %%mm3\n\t"

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

  655. 		"movq	%%mm0, %%mm2\n\t"

  656. 		"movq	%%mm3, %%mm4\n\t"

  657. 		"movq	%%mm3, %%mm5\n\t"

  658. 		"psllq	$8, %%mm0\n\t"

  659. 		"psllq	$8, %%mm3\n\t"

  660. 		"pand	%%mm7, %%mm0\n\t"

  661. 		"pand	%%mm7, %%mm3\n\t"

  662. 		"psrlq	$5, %%mm1\n\t"

  663. 		"psrlq	$5, %%mm4\n\t"

  664. 		"pand	%%mm6, %%mm1\n\t"

  665. 		"pand	%%mm6, %%mm4\n\t"

  666. 		"psrlq	$19, %%mm2\n\t"

  667. 		"psrlq	$19, %%mm5\n\t"

  668. 		"pand	%2, %%mm2\n\t"

  669. 		"pand	%2, %%mm5\n\t"

  670. 		"por	%%mm1, %%mm0\n\t"

  671. 		"por	%%mm4, %%mm3\n\t"

  672. 		"por	%%mm2, %%mm0\n\t"

  673. 		"por	%%mm5, %%mm3\n\t"

  674. 		"psllq	$16, %%mm3\n\t"

  675. 		"por	%%mm3, %%mm0\n\t"

  676. 		MOVNTQ"	%%mm0, %0\n\t"

  677. 		:"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");

  678. 		d += 4;

  679. 		s += 12;

  680. 	}

  681. 	__asm __volatile(SFENCE:::"memory");

  682. 	__asm __volatile(EMMS:::"memory");

  683. #endif

  684. 	while(s < end)

  685. 	{

  686. 		const int r= *s++;

  687. 		const int g= *s++;

  688. 		const int b= *s++;

  689. 		*d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);

  690. 	}

  691. }

  692. 

  693. static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  694. {

  695. 	const uint8_t *s = src;

  696. 	const uint8_t *end;

  697. #ifdef HAVE_MMX

  698. 	const uint8_t *mm_end;

  699. #endif

  700. 	uint16_t *d = (uint16_t *)dst;

  701. 	end = s + src_size;

  702. #ifdef HAVE_MMX

  703. 	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");

  704. 	__asm __volatile(

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

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

  707. 	    ::"m"(red_15mask),"m"(green_15mask));

  708. 	mm_end = end - 11;

  709. 	while(s < mm_end)

  710. 	{

  711. 	    __asm __volatile(

  712. 		PREFETCH" 32%1\n\t"

  713. 		"movd	%1, %%mm0\n\t"

  714. 		"movd	3%1, %%mm3\n\t"

  715. 		"punpckldq 6%1, %%mm0\n\t"

  716. 		"punpckldq 9%1, %%mm3\n\t"

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

  718. 		"movq	%%mm0, %%mm2\n\t"

  719. 		"movq	%%mm3, %%mm4\n\t"

  720. 		"movq	%%mm3, %%mm5\n\t"

  721. 		"psrlq	$3, %%mm0\n\t"

  722. 		"psrlq	$3, %%mm3\n\t"

  723. 		"pand	%2, %%mm0\n\t"

  724. 		"pand	%2, %%mm3\n\t"

  725. 		"psrlq	$6, %%mm1\n\t"

  726. 		"psrlq	$6, %%mm4\n\t"

  727. 		"pand	%%mm6, %%mm1\n\t"

  728. 		"pand	%%mm6, %%mm4\n\t"

  729. 		"psrlq	$9, %%mm2\n\t"

  730. 		"psrlq	$9, %%mm5\n\t"

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

  732. 		"pand	%%mm7, %%mm5\n\t"

  733. 		"por	%%mm1, %%mm0\n\t"

  734. 		"por	%%mm4, %%mm3\n\t"

  735. 		"por	%%mm2, %%mm0\n\t"

  736. 		"por	%%mm5, %%mm3\n\t"

  737. 		"psllq	$16, %%mm3\n\t"

  738. 		"por	%%mm3, %%mm0\n\t"

  739. 		MOVNTQ"	%%mm0, %0\n\t"

  740. 		:"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");

  741. 		d += 4;

  742. 		s += 12;

  743. 	}

  744. 	__asm __volatile(SFENCE:::"memory");

  745. 	__asm __volatile(EMMS:::"memory");

  746. #endif

  747. 	while(s < end)

  748. 	{

  749. 		const int b= *s++;

  750. 		const int g= *s++;

  751. 		const int r= *s++;

  752. 		*d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);

  753. 	}

  754. }

  755. 

  756. static inline void RENAME(rgb24tobgr15)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  757. {

  758. 	const uint8_t *s = src;

  759. 	const uint8_t *end;

  760. #ifdef HAVE_MMX

  761. 	const uint8_t *mm_end;

  762. #endif

  763. 	uint16_t *d = (uint16_t *)dst;

  764. 	end = s + src_size;

  765. #ifdef HAVE_MMX

  766. 	__asm __volatile(PREFETCH"	%0"::"m"(*src):"memory");

  767. 	__asm __volatile(

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

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

  770. 	    ::"m"(red_15mask),"m"(green_15mask));

  771. 	mm_end = end - 15;

  772. 	while(s < mm_end)

  773. 	{

  774. 	    __asm __volatile(

  775. 		PREFETCH" 32%1\n\t"

  776. 		"movd	%1, %%mm0\n\t"

  777. 		"movd	3%1, %%mm3\n\t"

  778. 		"punpckldq 6%1, %%mm0\n\t"

  779. 		"punpckldq 9%1, %%mm3\n\t"

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

  781. 		"movq	%%mm0, %%mm2\n\t"

  782. 		"movq	%%mm3, %%mm4\n\t"

  783. 		"movq	%%mm3, %%mm5\n\t"

  784. 		"psllq	$7, %%mm0\n\t"

  785. 		"psllq	$7, %%mm3\n\t"

  786. 		"pand	%%mm7, %%mm0\n\t"

  787. 		"pand	%%mm7, %%mm3\n\t"

  788. 		"psrlq	$6, %%mm1\n\t"

  789. 		"psrlq	$6, %%mm4\n\t"

  790. 		"pand	%%mm6, %%mm1\n\t"

  791. 		"pand	%%mm6, %%mm4\n\t"

  792. 		"psrlq	$19, %%mm2\n\t"

  793. 		"psrlq	$19, %%mm5\n\t"

  794. 		"pand	%2, %%mm2\n\t"

  795. 		"pand	%2, %%mm5\n\t"

  796. 		"por	%%mm1, %%mm0\n\t"

  797. 		"por	%%mm4, %%mm3\n\t"

  798. 		"por	%%mm2, %%mm0\n\t"

  799. 		"por	%%mm5, %%mm3\n\t"

  800. 		"psllq	$16, %%mm3\n\t"

  801. 		"por	%%mm3, %%mm0\n\t"

  802. 		MOVNTQ"	%%mm0, %0\n\t"

  803. 		:"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");

  804. 		d += 4;

  805. 		s += 12;

  806. 	}

  807. 	__asm __volatile(SFENCE:::"memory");

  808. 	__asm __volatile(EMMS:::"memory");

  809. #endif

  810. 	while(s < end)

  811. 	{

  812. 		const int r= *s++;

  813. 		const int g= *s++;

  814. 		const int b= *s++;

  815. 		*d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);

  816. 	}

  817. }

  818. 

  819. /*

  820.   I use here less accurate approximation by simply

  821.  left-shifting the input

  822.   value and filling the low order bits with

  823.  zeroes. This method improves png's

  824.   compression but this scheme cannot reproduce white exactly, since it does not

  825.   generate an all-ones maximum value; the net effect is to darken the

  826.   image slightly.

  827. 

  828.   The better method should be "left bit replication":

  829. 

  830.    4 3 2 1 0

  831.    ---------

  832.    1 1 0 1 1

  833. 

  834.    7 6 5 4 3  2 1 0

  835.    ----------------

  836.    1 1 0 1 1  1 1 0

  837.    |=======|  |===|

  838.        |      Leftmost Bits Repeated to Fill Open Bits

  839.        |

  840.    Original Bits

  841. */

  842. static inline void RENAME(rgb15to24)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  843. {

  844. 	const uint16_t *end;

  845. #ifdef HAVE_MMX

  846. 	const uint16_t *mm_end;

  847. #endif

  848. 	uint8_t *d = (uint8_t *)dst;

  849. 	const uint16_t *s = (uint16_t *)src;

  850. 	end = s + src_size/2;

  851. #ifdef HAVE_MMX

  852. 	__asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");

  853. 	mm_end = end - 7;

  854. 	while(s < mm_end)

  855. 	{

  856. 	    __asm __volatile(

  857. 		PREFETCH" 32%1\n\t"

  858. 		"movq	%1, %%mm0\n\t"

  859. 		"movq	%1, %%mm1\n\t"

  860. 		"movq	%1, %%mm2\n\t"

  861. 		"pand	%2, %%mm0\n\t"

  862. 		"pand	%3, %%mm1\n\t"

  863. 		"pand	%4, %%mm2\n\t"

  864. 		"psllq	$3, %%mm0\n\t"

  865. 		"psrlq	$2, %%mm1\n\t"

  866. 		"psrlq	$7, %%mm2\n\t"

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

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

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

  870. 		"punpcklwd %5, %%mm0\n\t"

  871. 		"punpcklwd %5, %%mm1\n\t"

  872. 		"punpcklwd %5, %%mm2\n\t"

  873. 		"punpckhwd %5, %%mm3\n\t"

  874. 		"punpckhwd %5, %%mm4\n\t"

  875. 		"punpckhwd %5, %%mm5\n\t"

  876. 		"psllq	$8, %%mm1\n\t"

  877. 		"psllq	$16, %%mm2\n\t"

  878. 		"por	%%mm1, %%mm0\n\t"

  879. 		"por	%%mm2, %%mm0\n\t"

  880. 		"psllq	$8, %%mm4\n\t"

  881. 		"psllq	$16, %%mm5\n\t"

  882. 		"por	%%mm4, %%mm3\n\t"

  883. 		"por	%%mm5, %%mm3\n\t"

  884. 

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

  886. 		"movq	%%mm3, %%mm7\n\t"

  887. 		

  888. 		"movq	8%1, %%mm0\n\t"

  889. 		"movq	8%1, %%mm1\n\t"

  890. 		"movq	8%1, %%mm2\n\t"

  891. 		"pand	%2, %%mm0\n\t"

  892. 		"pand	%3, %%mm1\n\t"

  893. 		"pand	%4, %%mm2\n\t"

  894. 		"psllq	$3, %%mm0\n\t"

  895. 		"psrlq	$2, %%mm1\n\t"

  896. 		"psrlq	$7, %%mm2\n\t"

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

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

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

  900. 		"punpcklwd %5, %%mm0\n\t"

  901. 		"punpcklwd %5, %%mm1\n\t"

  902. 		"punpcklwd %5, %%mm2\n\t"

  903. 		"punpckhwd %5, %%mm3\n\t"

  904. 		"punpckhwd %5, %%mm4\n\t"

  905. 		"punpckhwd %5, %%mm5\n\t"

  906. 		"psllq	$8, %%mm1\n\t"

  907. 		"psllq	$16, %%mm2\n\t"

  908. 		"por	%%mm1, %%mm0\n\t"

  909. 		"por	%%mm2, %%mm0\n\t"

  910. 		"psllq	$8, %%mm4\n\t"

  911. 		"psllq	$16, %%mm5\n\t"

  912. 		"por	%%mm4, %%mm3\n\t"

  913. 		"por	%%mm5, %%mm3\n\t"

  914. 

  915. 		:"=m"(*d)

  916. 		:"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null)

  917. 		:"memory");

  918. 	    /* Borrowed 32 to 24 */

  919. 	    __asm __volatile(

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

  921. 		"movq	%%mm3, %%mm5\n\t"

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

  923. 		"movq	%%mm7, %%mm1\n\t"

  924. 		

  925. 		"movq	%%mm4, %%mm6\n\t"

  926. 		"movq	%%mm5, %%mm7\n\t"

  927. 		"movq	%%mm0, %%mm2\n\t"

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

  929. 

  930. 		"psrlq	$8, %%mm2\n\t"

  931. 		"psrlq	$8, %%mm3\n\t"

  932. 		"psrlq	$8, %%mm6\n\t"

  933. 		"psrlq	$8, %%mm7\n\t"

  934. 		"pand	%2, %%mm0\n\t"

  935. 		"pand	%2, %%mm1\n\t"

  936. 		"pand	%2, %%mm4\n\t"

  937. 		"pand	%2, %%mm5\n\t"

  938. 		"pand	%3, %%mm2\n\t"

  939. 		"pand	%3, %%mm3\n\t"

  940. 		"pand	%3, %%mm6\n\t"

  941. 		"pand	%3, %%mm7\n\t"

  942. 		"por	%%mm2, %%mm0\n\t"

  943. 		"por	%%mm3, %%mm1\n\t"

  944. 		"por	%%mm6, %%mm4\n\t"

  945. 		"por	%%mm7, %%mm5\n\t"

  946. 

  947. 		"movq	%%mm1, %%mm2\n\t"

  948. 		"movq	%%mm4, %%mm3\n\t"

  949. 		"psllq	$48, %%mm2\n\t"

  950. 		"psllq	$32, %%mm3\n\t"

  951. 		"pand	%4, %%mm2\n\t"

  952. 		"pand	%5, %%mm3\n\t"

  953. 		"por	%%mm2, %%mm0\n\t"

  954. 		"psrlq	$16, %%mm1\n\t"

  955. 		"psrlq	$32, %%mm4\n\t"

  956. 		"psllq	$16, %%mm5\n\t"

  957. 		"por	%%mm3, %%mm1\n\t"

  958. 		"pand	%6, %%mm5\n\t"

  959. 		"por	%%mm5, %%mm4\n\t"

  960. 

  961. 		MOVNTQ"	%%mm0, %0\n\t"

  962. 		MOVNTQ"	%%mm1, 8%0\n\t"

  963. 		MOVNTQ"	%%mm4, 16%0"

  964. 

  965. 		:"=m"(*d)

  966. 		:"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)

  967. 		:"memory");

  968. 		d += 24;

  969. 		s += 8;

  970. 	}

  971. 	__asm __volatile(SFENCE:::"memory");

  972. 	__asm __volatile(EMMS:::"memory");

  973. #endif

  974. 	while(s < end)

  975. 	{

  976. 		register uint16_t bgr;

  977. 		bgr = *s++;

  978. 		*d++ = (bgr&0x1F)<<3;

  979. 		*d++ = (bgr&0x3E0)>>2;

  980. 		*d++ = (bgr&0x7C00)>>7;

  981. 	}

  982. }

  983. 

  984. static inline void RENAME(rgb16to24)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  985. {

  986. 	const uint16_t *end;

  987. #ifdef HAVE_MMX

  988. 	const uint16_t *mm_end;

  989. #endif

  990. 	uint8_t *d = (uint8_t *)dst;

  991. 	const uint16_t *s = (const uint16_t *)src;

  992. 	end = s + src_size/2;

  993. #ifdef HAVE_MMX

  994. 	__asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");

  995. 	mm_end = end - 7;

  996. 	while(s < mm_end)

  997. 	{

  998. 	    __asm __volatile(

  999. 		PREFETCH" 32%1\n\t"

  1000. 		"movq	%1, %%mm0\n\t"

  1001. 		"movq	%1, %%mm1\n\t"

  1002. 		"movq	%1, %%mm2\n\t"

  1003. 		"pand	%2, %%mm0\n\t"

  1004. 		"pand	%3, %%mm1\n\t"

  1005. 		"pand	%4, %%mm2\n\t"

  1006. 		"psllq	$3, %%mm0\n\t"

  1007. 		"psrlq	$3, %%mm1\n\t"

  1008. 		"psrlq	$8, %%mm2\n\t"

  1009. 		"movq	%%mm0, %%mm3\n\t"

  1010. 		"movq	%%mm1, %%mm4\n\t"

  1011. 		"movq	%%mm2, %%mm5\n\t"

  1012. 		"punpcklwd %5, %%mm0\n\t"

  1013. 		"punpcklwd %5, %%mm1\n\t"

  1014. 		"punpcklwd %5, %%mm2\n\t"

  1015. 		"punpckhwd %5, %%mm3\n\t"

  1016. 		"punpckhwd %5, %%mm4\n\t"

  1017. 		"punpckhwd %5, %%mm5\n\t"

  1018. 		"psllq	$8, %%mm1\n\t"

  1019. 		"psllq	$16, %%mm2\n\t"

  1020. 		"por	%%mm1, %%mm0\n\t"

  1021. 		"por	%%mm2, %%mm0\n\t"

  1022. 		"psllq	$8, %%mm4\n\t"

  1023. 		"psllq	$16, %%mm5\n\t"

  1024. 		"por	%%mm4, %%mm3\n\t"

  1025. 		"por	%%mm5, %%mm3\n\t"

  1026. 		

  1027. 		"movq	%%mm0, %%mm6\n\t"

  1028. 		"movq	%%mm3, %%mm7\n\t"

  1029. 

  1030. 		"movq	8%1, %%mm0\n\t"

  1031. 		"movq	8%1, %%mm1\n\t"

  1032. 		"movq	8%1, %%mm2\n\t"

  1033. 		"pand	%2, %%mm0\n\t"

  1034. 		"pand	%3, %%mm1\n\t"

  1035. 		"pand	%4, %%mm2\n\t"

  1036. 		"psllq	$3, %%mm0\n\t"

  1037. 		"psrlq	$3, %%mm1\n\t"

  1038. 		"psrlq	$8, %%mm2\n\t"

  1039. 		"movq	%%mm0, %%mm3\n\t"

  1040. 		"movq	%%mm1, %%mm4\n\t"

  1041. 		"movq	%%mm2, %%mm5\n\t"

  1042. 		"punpcklwd %5, %%mm0\n\t"

  1043. 		"punpcklwd %5, %%mm1\n\t"

  1044. 		"punpcklwd %5, %%mm2\n\t"

  1045. 		"punpckhwd %5, %%mm3\n\t"

  1046. 		"punpckhwd %5, %%mm4\n\t"

  1047. 		"punpckhwd %5, %%mm5\n\t"

  1048. 		"psllq	$8, %%mm1\n\t"

  1049. 		"psllq	$16, %%mm2\n\t"

  1050. 		"por	%%mm1, %%mm0\n\t"

  1051. 		"por	%%mm2, %%mm0\n\t"

  1052. 		"psllq	$8, %%mm4\n\t"

  1053. 		"psllq	$16, %%mm5\n\t"

  1054. 		"por	%%mm4, %%mm3\n\t"

  1055. 		"por	%%mm5, %%mm3\n\t"

  1056. 		:"=m"(*d)

  1057. 		:"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null)		

  1058. 		:"memory");

  1059. 	    /* Borrowed 32 to 24 */

  1060. 	    __asm __volatile(

  1061. 		"movq	%%mm0, %%mm4\n\t"

  1062. 		"movq	%%mm3, %%mm5\n\t"

  1063. 		"movq	%%mm6, %%mm0\n\t"

  1064. 		"movq	%%mm7, %%mm1\n\t"

  1065. 		

  1066. 		"movq	%%mm4, %%mm6\n\t"

  1067. 		"movq	%%mm5, %%mm7\n\t"

  1068. 		"movq	%%mm0, %%mm2\n\t"

  1069. 		"movq	%%mm1, %%mm3\n\t"

  1070. 

  1071. 		"psrlq	$8, %%mm2\n\t"

  1072. 		"psrlq	$8, %%mm3\n\t"

  1073. 		"psrlq	$8, %%mm6\n\t"

  1074. 		"psrlq	$8, %%mm7\n\t"

  1075. 		"pand	%2, %%mm0\n\t"

  1076. 		"pand	%2, %%mm1\n\t"

  1077. 		"pand	%2, %%mm4\n\t"

  1078. 		"pand	%2, %%mm5\n\t"

  1079. 		"pand	%3, %%mm2\n\t"

  1080. 		"pand	%3, %%mm3\n\t"

  1081. 		"pand	%3, %%mm6\n\t"

  1082. 		"pand	%3, %%mm7\n\t"

  1083. 		"por	%%mm2, %%mm0\n\t"

  1084. 		"por	%%mm3, %%mm1\n\t"

  1085. 		"por	%%mm6, %%mm4\n\t"

  1086. 		"por	%%mm7, %%mm5\n\t"

  1087. 

  1088. 		"movq	%%mm1, %%mm2\n\t"

  1089. 		"movq	%%mm4, %%mm3\n\t"

  1090. 		"psllq	$48, %%mm2\n\t"

  1091. 		"psllq	$32, %%mm3\n\t"

  1092. 		"pand	%4, %%mm2\n\t"

  1093. 		"pand	%5, %%mm3\n\t"

  1094. 		"por	%%mm2, %%mm0\n\t"

  1095. 		"psrlq	$16, %%mm1\n\t"

  1096. 		"psrlq	$32, %%mm4\n\t"

  1097. 		"psllq	$16, %%mm5\n\t"

  1098. 		"por	%%mm3, %%mm1\n\t"

  1099. 		"pand	%6, %%mm5\n\t"

  1100. 		"por	%%mm5, %%mm4\n\t"

  1101. 

  1102. 		MOVNTQ"	%%mm0, %0\n\t"

  1103. 		MOVNTQ"	%%mm1, 8%0\n\t"

  1104. 		MOVNTQ"	%%mm4, 16%0"

  1105. 

  1106. 		:"=m"(*d)

  1107. 		:"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)

  1108. 		:"memory");

  1109. 		d += 24;

  1110. 		s += 8;

  1111. 	}

  1112. 	__asm __volatile(SFENCE:::"memory");

  1113. 	__asm __volatile(EMMS:::"memory");

  1114. #endif

  1115. 	while(s < end)

  1116. 	{

  1117. 		register uint16_t bgr;

  1118. 		bgr = *s++;

  1119. 		*d++ = (bgr&0x1F)<<3;

  1120. 		*d++ = (bgr&0x7E0)>>3;

  1121. 		*d++ = (bgr&0xF800)>>8;

  1122. 	}

  1123. }

  1124. 

  1125. static inline void RENAME(rgb15to32)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  1126. {

  1127. 	const uint16_t *end;

  1128. #ifdef HAVE_MMX

  1129. 	const uint16_t *mm_end;

  1130. #endif

  1131. 	uint8_t *d = (uint8_t *)dst;

  1132. 	const uint16_t *s = (const uint16_t *)src;

  1133. 	end = s + src_size/2;

  1134. #ifdef HAVE_MMX

  1135. 	__asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");

  1136. 	__asm __volatile("pxor	%%mm7,%%mm7\n\t":::"memory");

  1137. 	mm_end = end - 3;

  1138. 	while(s < mm_end)

  1139. 	{

  1140. 	    __asm __volatile(

  1141. 		PREFETCH" 32%1\n\t"

  1142. 		"movq	%1, %%mm0\n\t"

  1143. 		"movq	%1, %%mm1\n\t"

  1144. 		"movq	%1, %%mm2\n\t"

  1145. 		"pand	%2, %%mm0\n\t"

  1146. 		"pand	%3, %%mm1\n\t"

  1147. 		"pand	%4, %%mm2\n\t"

  1148. 		"psllq	$3, %%mm0\n\t"

  1149. 		"psrlq	$2, %%mm1\n\t"

  1150. 		"psrlq	$7, %%mm2\n\t"

  1151. 		"movq	%%mm0, %%mm3\n\t"

  1152. 		"movq	%%mm1, %%mm4\n\t"

  1153. 		"movq	%%mm2, %%mm5\n\t"

  1154. 		"punpcklwd %%mm7, %%mm0\n\t"

  1155. 		"punpcklwd %%mm7, %%mm1\n\t"

  1156. 		"punpcklwd %%mm7, %%mm2\n\t"

  1157. 		"punpckhwd %%mm7, %%mm3\n\t"

  1158. 		"punpckhwd %%mm7, %%mm4\n\t"

  1159. 		"punpckhwd %%mm7, %%mm5\n\t"

  1160. 		"psllq	$8, %%mm1\n\t"

  1161. 		"psllq	$16, %%mm2\n\t"

  1162. 		"por	%%mm1, %%mm0\n\t"

  1163. 		"por	%%mm2, %%mm0\n\t"

  1164. 		"psllq	$8, %%mm4\n\t"

  1165. 		"psllq	$16, %%mm5\n\t"

  1166. 		"por	%%mm4, %%mm3\n\t"

  1167. 		"por	%%mm5, %%mm3\n\t"

  1168. 		MOVNTQ"	%%mm0, %0\n\t"

  1169. 		MOVNTQ"	%%mm3, 8%0\n\t"

  1170. 		:"=m"(*d)

  1171. 		:"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r)

  1172. 		:"memory");

  1173. 		d += 16;

  1174. 		s += 4;

  1175. 	}

  1176. 	__asm __volatile(SFENCE:::"memory");

  1177. 	__asm __volatile(EMMS:::"memory");

  1178. #endif

  1179. 	while(s < end)

  1180. 	{

  1181. 		register uint16_t bgr;

  1182. 		bgr = *s++;

  1183. 		*d++ = (bgr&0x1F)<<3;

  1184. 		*d++ = (bgr&0x3E0)>>2;

  1185. 		*d++ = (bgr&0x7C00)>>7;

  1186. 		*d++ = 0;

  1187. 	}

  1188. }

  1189. 

  1190. static inline void RENAME(rgb16to32)(const uint8_t *src, uint8_t *dst, unsigned src_size)

  1191. {

  1192. 	const uint16_t *end;

  1193. #ifdef HAVE_MMX

  1194. 	const uint16_t *mm_end;

  1195. #endif

  1196. 	uint8_t *d = (uint8_t *)dst;

  1197. 	const uint16_t *s = (uint16_t *)src;

  1198. 	end = s + src_size/2;

  1199. #ifdef HAVE_MMX

  1200. 	__asm __volatile(PREFETCH"	%0"::"m"(*s):"memory");

  1201. 	__asm __volatile("pxor	%%mm7,%%mm7\n\t":::"memory");

  1202. 	mm_end = end - 3;

  1203. 	while(s < mm_end)

  1204. 	{

  1205. 	    __asm __volatile(

  1206. 		PREFETCH" 32%1\n\t"

  1207. 		"movq	%1, %%mm0\n\t"

  1208. 		"movq	%1, %%mm1\n\t"

  1209. 		"movq	%1, %%mm2\n\t"

  1210. 		"pand	%2, %%mm0\n\t"

  1211. 		"pand	%3, %%mm1\n\t"

  1212. 		"pand	%4, %%mm2\n\t"

  1213. 		"psllq	$3, %%mm0\n\t"

  1214. 		"psrlq	$3, %%mm1\n\t"

  1215. 		"psrlq	$8, %%mm2\n\t"

  1216. 		"movq	%%mm0, %%mm3\n\t"

  1217. 		"movq	%%mm1, %%mm4\n\t"

  1218. 		"movq	%%mm2, %%mm5\n\t"

  1219. 		"punpcklwd %%mm7, %%mm0\n\t"

  1220. 		"punpcklwd %%mm7, %%mm1\n\t"

  1221. 		"punpcklwd %%mm7, %%mm2\n\t"

  1222. 		"punpckhwd %%mm7, %%mm3\n\t"

  1223. 		"punpckhwd %%mm7, %%mm4\n\t"

  1224. 		"punpckhwd %%mm7, %%mm5\n\t"

  1225. 		"psllq	$8, %%mm1\n\t"

  1226. 		"psllq	$16, %%mm2\n\t"

  1227. 		"por	%%mm1, %%mm0\n\t"

  1228. 		"por	%%mm2, %%mm0\n\t"

  1229. 		"psllq	$8, %%mm4\n\t"

  1230. 		"psllq	$16, %%mm5\n\t"

  1231. 		"por	%%mm4, %%mm3\n\t"

  1232. 		"por	%%mm5, %%mm3\n\t"

  1233. 		MOVNTQ"	%%mm0, %0\n\t"

  1234. 		MOVNTQ"	%%mm3, 8%0\n\t"

  1235. 		:"=m"(*d)

  1236. 		:"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r)

  1237. 		:"memory");

  1238. 		d += 16;

  1239. 		s += 4;

  1240. 	}

  1241. 	__asm __volatile(SFENCE:::"memory");

  1242. 	__asm __volatile(EMMS:::"memory");

  1243. #endif

  1244. 	while(s < end)

  1245. 	{

  1246. 		register uint16_t bgr;

  1247. 		bgr = *s++;

  1248. 		*d++ = (bgr&0x1F)<<3;

  1249. 		*d++ = (bgr&0x7E0)>>3;

  1250. 		*d++ = (bgr&0xF800)>>8;

  1251. 		*d++ = 0;

  1252. 	}

  1253. }

  1254. 

  1255. static inline void RENAME(rgb32tobgr32)(const uint8_t *src, uint8_t *dst, unsigned int src_size)

  1256. {

  1257. #ifdef HAVE_MMX

  1258. /* TODO: unroll this loop */

  1259. 	asm volatile (

  1260. 		"xorl %%eax, %%eax		\n\t"

  1261. 		".balign 16			\n\t"

  1262. 		"1:				\n\t"

  1263. 		PREFETCH" 32(%0, %%eax)		\n\t"

  1264. 		"movq (%0, %%eax), %%mm0	\n\t"

  1265. 		"movq %%mm0, %%mm1		\n\t"

  1266. 		"movq %%mm0, %%mm2		\n\t"

  1267. 		"pslld $16, %%mm0		\n\t"

  1268. 		"psrld $16, %%mm1		\n\t"

  1269. 		"pand "MANGLE(mask32r)", %%mm0	\n\t"

  1270. 		"pand "MANGLE(mask32g)", %%mm2	\n\t"

  1271. 		"pand "MANGLE(mask32b)", %%mm1	\n\t"

  1272. 		"por %%mm0, %%mm2		\n\t"

  1273. 		"por %%mm1, %%mm2		\n\t"

  1274. 		MOVNTQ" %%mm2, (%1, %%eax)	\n\t"

  1275. 		"addl $8, %%eax			\n\t"

  1276. 		"cmpl %2, %%eax			\n\t"

  1277. 		" jb 1b				\n\t"

  1278. 		:: "r" (src), "r"(dst), "r" (src_size-7)

  1279. 		: "%eax"

  1280. 	);

  1281. 

  1282. 	__asm __volatile(SFENCE:::"memory");

  1283. 	__asm __volatile(EMMS:::"memory");

  1284. #else

  1285. 	unsigned i;

  1286. 	unsigned num_pixels = src_size >> 2;

  1287. 	for(i=0; i<num_pixels; i++)

  1288. 	{

  1289. 		dst[4*i + 0] = src[4*i + 2];

  1290. 		dst[4*i + 1] = src[4*i + 1];

  1291. 		dst[4*i + 2] = src[4*i + 0];

  1292. 	}

  1293. #endif

  1294. }

  1295. 

  1296. static inline void RENAME(rgb24tobgr24)(const uint8_t *src, uint8_t *dst, unsigned int src_size)

  1297. {

  1298. 	unsigned i;

  1299. #ifdef HAVE_MMX

  1300. 	int mmx_size= 23 - src_size;

  1301. 	asm volatile (

  1302. 		"movq "MANGLE(mask24r)", %%mm5	\n\t"

  1303. 		"movq "MANGLE(mask24g)", %%mm6	\n\t"

  1304. 		"movq "MANGLE(mask24b)", %%mm7	\n\t"

  1305. 		".balign 16			\n\t"

  1306. 		"1:				\n\t"

  1307. 		PREFETCH" 32(%1, %%eax)		\n\t"

  1308. 		"movq   (%1, %%eax), %%mm0	\n\t" // BGR BGR BG

  1309. 		"movq   (%1, %%eax), %%mm1	\n\t" // BGR BGR BG

  1310. 		"movq  2(%1, %%eax), %%mm2	\n\t" // R BGR BGR B

  1311. 		"psllq $16, %%mm0		\n\t" // 00 BGR BGR

  1312. 		"pand %%mm5, %%mm0		\n\t"

  1313. 		"pand %%mm6, %%mm1		\n\t"

  1314. 		"pand %%mm7, %%mm2		\n\t"

  1315. 		"por %%mm0, %%mm1		\n\t"

  1316. 		"por %%mm2, %%mm1		\n\t"                

  1317. 		"movq  6(%1, %%eax), %%mm0	\n\t" // BGR BGR BG

  1318. 		MOVNTQ" %%mm1,   (%2, %%eax)	\n\t" // RGB RGB RG

  1319. 		"movq  8(%1, %%eax), %%mm1	\n\t" // R BGR BGR B

  1320. 		"movq 10(%1, %%eax), %%mm2	\n\t" // GR BGR BGR

  1321. 		"pand %%mm7, %%mm0		\n\t"

  1322. 		"pand %%mm5, %%mm1		\n\t"

  1323. 		"pand %%mm6, %%mm2		\n\t"

  1324. 		"por %%mm0, %%mm1		\n\t"

  1325. 		"por %%mm2, %%mm1		\n\t"                

  1326. 		"movq 14(%1, %%eax), %%mm0	\n\t" // R BGR BGR B

  1327. 		MOVNTQ" %%mm1,  8(%2, %%eax)	\n\t" // B RGB RGB R

  1328. 		"movq 16(%1, %%eax), %%mm1	\n\t" // GR BGR BGR

  1329. 		"movq 18(%1, %%eax), %%mm2	\n\t" // BGR BGR BG

  1330. 		"pand %%mm6, %%mm0		\n\t"

  1331. 		"pand %%mm7, %%mm1		\n\t"

  1332. 		"pand %%mm5, %%mm2		\n\t"

  1333. 		"por %%mm0, %%mm1		\n\t"

  1334. 		"por %%mm2, %%mm1		\n\t"                

  1335. 		MOVNTQ" %%mm1, 16(%2, %%eax)	\n\t"

  1336. 		"addl $24, %%eax		\n\t"

  1337. 		" js 1b				\n\t"

  1338. 		: "+a" (mmx_size)

  1339. 		: "r" (src-mmx_size), "r"(dst-mmx_size)

  1340. 	);

  1341. 

  1342. 	__asm __volatile(SFENCE:::"memory");

  1343. 	__asm __volatile(EMMS:::"memory");

  1344. 

  1345. 	if(mmx_size==23) return; //finihsed, was multiple of 8

  1346. 

  1347. 	src+= src_size;

  1348. 	dst+= src_size;

  1349. 	src_size= 23-mmx_size;

  1350. 	src-= src_size;

  1351. 	dst-= src_size;

  1352. #endif

  1353. 	for(i=0; i<src_size; i+=3)

  1354. 	{

  1355. 		register uint8_t x;

  1356. 		x          = src[i + 2];

  1357. 		dst[i + 1] = src[i + 1];

  1358. 		dst[i + 2] = src[i + 0];

  1359. 		dst[i + 0] = x;

  1360. 	}

  1361. }

  1362. 

  1363. static inline void RENAME(yuvPlanartoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1364. 	unsigned int width, unsigned int height,

  1365. 	unsigned int lumStride, unsigned int chromStride, unsigned int dstStride, int vertLumPerChroma)

  1366. {

  1367. 	unsigned y;

  1368. 	const unsigned chromWidth= width>>1;

  1369. 	for(y=0; y<height; y++)

  1370. 	{

  1371. #ifdef HAVE_MMX

  1372. //FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)

  1373. 		asm volatile(

  1374. 			"xorl %%eax, %%eax		\n\t"

  1375. 			".balign 16			\n\t"

  1376. 			"1:				\n\t"

  1377. 			PREFETCH" 32(%1, %%eax, 2)	\n\t"

  1378. 			PREFETCH" 32(%2, %%eax)		\n\t"

  1379. 			PREFETCH" 32(%3, %%eax)		\n\t"

  1380. 			"movq (%2, %%eax), %%mm0	\n\t" // U(0)

  1381. 			"movq %%mm0, %%mm2		\n\t" // U(0)

  1382. 			"movq (%3, %%eax), %%mm1	\n\t" // V(0)

  1383. 			"punpcklbw %%mm1, %%mm0		\n\t" // UVUV UVUV(0)

  1384. 			"punpckhbw %%mm1, %%mm2		\n\t" // UVUV UVUV(8)

  1385. 

  1386. 			"movq (%1, %%eax,2), %%mm3	\n\t" // Y(0)

  1387. 			"movq 8(%1, %%eax,2), %%mm5	\n\t" // Y(8)

  1388. 			"movq %%mm3, %%mm4		\n\t" // Y(0)

  1389. 			"movq %%mm5, %%mm6		\n\t" // Y(8)

  1390. 			"punpcklbw %%mm0, %%mm3		\n\t" // YUYV YUYV(0)

  1391. 			"punpckhbw %%mm0, %%mm4		\n\t" // YUYV YUYV(4)

  1392. 			"punpcklbw %%mm2, %%mm5		\n\t" // YUYV YUYV(8)

  1393. 			"punpckhbw %%mm2, %%mm6		\n\t" // YUYV YUYV(12)

  1394. 

  1395. 			MOVNTQ" %%mm3, (%0, %%eax, 4)	\n\t"

  1396. 			MOVNTQ" %%mm4, 8(%0, %%eax, 4)	\n\t"

  1397. 			MOVNTQ" %%mm5, 16(%0, %%eax, 4)	\n\t"

  1398. 			MOVNTQ" %%mm6, 24(%0, %%eax, 4)	\n\t"

  1399. 

  1400. 			"addl $8, %%eax			\n\t"

  1401. 			"cmpl %4, %%eax			\n\t"

  1402. 			" jb 1b				\n\t"

  1403. 			::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "r" (chromWidth)

  1404. 			: "%eax"

  1405. 		);

  1406. #else

  1407. #if __WORDSIZE >= 64

  1408. 		int i;

  1409. 		uint64_t *ldst = (uint64_t *) dst;

  1410. 		const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;

  1411. 		for(i = 0; i < chromWidth; i += 2){

  1412. 			uint64_t k, l;

  1413. 			k = yc[0] + (uc[0] << 8) +

  1414. 			    (yc[1] << 16) + (vc[0] << 24);

  1415. 			l = yc[2] + (uc[1] << 8) +

  1416. 			    (yc[3] << 16) + (vc[1] << 24);

  1417. 			*ldst++ = k + (l << 32);

  1418. 			yc += 4;

  1419. 			uc += 2;

  1420. 			vc += 2;

  1421. 		}

  1422. 

  1423. #else

  1424. 		int i, *idst = (int32_t *) dst;

  1425. 		const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;

  1426. 		for(i = 0; i < chromWidth; i++){

  1427. 			*idst++ = yc[0] + (uc[0] << 8) +

  1428. 			    (yc[1] << 16) + (vc[0] << 24);

  1429. 			yc += 2;

  1430. 			uc++;

  1431. 			vc++;

  1432. 		}

  1433. #endif

  1434. #endif

  1435. 		if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )

  1436. 		{

  1437. 			usrc += chromStride;

  1438. 			vsrc += chromStride;

  1439. 		}

  1440. 		ysrc += lumStride;

  1441. 		dst += dstStride;

  1442. 	}

  1443. #ifdef HAVE_MMX

  1444. asm(    EMMS" \n\t"

  1445.         SFENCE" \n\t"

  1446.         :::"memory");

  1447. #endif

  1448. }

  1449. 

  1450. /**

  1451.  *

  1452.  * height should be a multiple of 2 and width should be a multiple of 16 (if this is a

  1453.  * problem for anyone then tell me, and ill fix it)

  1454.  */

  1455. static inline void RENAME(yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1456. 	unsigned int width, unsigned int height,

  1457. 	unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)

  1458. {

  1459. 	//FIXME interpolate chroma

  1460. 	RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);

  1461. }

  1462. 

  1463. /**

  1464.  *

  1465.  * width should be a multiple of 16

  1466.  */

  1467. static inline void RENAME(yuv422ptoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1468. 	unsigned int width, unsigned int height,

  1469. 	unsigned int lumStride, unsigned int chromStride, unsigned int dstStride)

  1470. {

  1471. 	RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1);

  1472. }

  1473. 

  1474. /**

  1475.  *

  1476.  * height should be a multiple of 2 and width should be a multiple of 16 (if this is a

  1477.  * problem for anyone then tell me, and ill fix it)

  1478.  */

  1479. static inline void RENAME(yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,

  1480. 	unsigned int width, unsigned int height,

  1481. 	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)

  1482. {

  1483. 	unsigned y;

  1484. 	const unsigned chromWidth= width>>1;

  1485. 	for(y=0; y<height; y+=2)

  1486. 	{

  1487. #ifdef HAVE_MMX

  1488. 		asm volatile(

  1489. 			"xorl %%eax, %%eax		\n\t"

  1490. 			"pcmpeqw %%mm7, %%mm7		\n\t"

  1491. 			"psrlw $8, %%mm7		\n\t" // FF,00,FF,00...

  1492. 			".balign 16			\n\t"

  1493. 			"1:				\n\t"

  1494. 			PREFETCH" 64(%0, %%eax, 4)	\n\t"

  1495. 			"movq (%0, %%eax, 4), %%mm0	\n\t" // YUYV YUYV(0)

  1496. 			"movq 8(%0, %%eax, 4), %%mm1	\n\t" // YUYV YUYV(4)

  1497. 			"movq %%mm0, %%mm2		\n\t" // YUYV YUYV(0)

  1498. 			"movq %%mm1, %%mm3		\n\t" // YUYV YUYV(4)

  1499. 			"psrlw $8, %%mm0		\n\t" // U0V0 U0V0(0)

  1500. 			"psrlw $8, %%mm1		\n\t" // U0V0 U0V0(4)

  1501. 			"pand %%mm7, %%mm2		\n\t" // Y0Y0 Y0Y0(0)

  1502. 			"pand %%mm7, %%mm3		\n\t" // Y0Y0 Y0Y0(4)

  1503. 			"packuswb %%mm1, %%mm0		\n\t" // UVUV UVUV(0)

  1504. 			"packuswb %%mm3, %%mm2		\n\t" // YYYY YYYY(0)

  1505. 

  1506. 			MOVNTQ" %%mm2, (%1, %%eax, 2)	\n\t"

  1507. 

  1508. 			"movq 16(%0, %%eax, 4), %%mm1	\n\t" // YUYV YUYV(8)

  1509. 			"movq 24(%0, %%eax, 4), %%mm2	\n\t" // YUYV YUYV(12)

  1510. 			"movq %%mm1, %%mm3		\n\t" // YUYV YUYV(8)

  1511. 			"movq %%mm2, %%mm4		\n\t" // YUYV YUYV(12)

  1512. 			"psrlw $8, %%mm1		\n\t" // U0V0 U0V0(8)

  1513. 			"psrlw $8, %%mm2		\n\t" // U0V0 U0V0(12)

  1514. 			"pand %%mm7, %%mm3		\n\t" // Y0Y0 Y0Y0(8)

  1515. 			"pand %%mm7, %%mm4		\n\t" // Y0Y0 Y0Y0(12)

  1516. 			"packuswb %%mm2, %%mm1		\n\t" // UVUV UVUV(8)

  1517. 			"packuswb %%mm4, %%mm3		\n\t" // YYYY YYYY(8)

  1518. 

  1519. 			MOVNTQ" %%mm3, 8(%1, %%eax, 2)	\n\t"

  1520. 

  1521. 			"movq %%mm0, %%mm2		\n\t" // UVUV UVUV(0)

  1522. 			"movq %%mm1, %%mm3		\n\t" // UVUV UVUV(8)

  1523. 			"psrlw $8, %%mm0		\n\t" // V0V0 V0V0(0)

  1524. 			"psrlw $8, %%mm1		\n\t" // V0V0 V0V0(8)

  1525. 			"pand %%mm7, %%mm2		\n\t" // U0U0 U0U0(0)

  1526. 			"pand %%mm7, %%mm3		\n\t" // U0U0 U0U0(8)

  1527. 			"packuswb %%mm1, %%mm0		\n\t" // VVVV VVVV(0)

  1528. 			"packuswb %%mm3, %%mm2		\n\t" // UUUU UUUU(0)

  1529. 

  1530. 			MOVNTQ" %%mm0, (%3, %%eax)	\n\t"

  1531. 			MOVNTQ" %%mm2, (%2, %%eax)	\n\t"

  1532. 

  1533. 			"addl $8, %%eax			\n\t"

  1534. 			"cmpl %4, %%eax			\n\t"

  1535. 			" jb 1b				\n\t"

  1536. 			::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (chromWidth)

  1537. 			: "memory", "%eax"

  1538. 		);

  1539. 

  1540. 		ydst += lumStride;

  1541. 		src  += srcStride;

  1542. 

  1543. 		asm volatile(

  1544. 			"xorl %%eax, %%eax		\n\t"

  1545. 			".balign 16			\n\t"

  1546. 			"1:				\n\t"

  1547. 			PREFETCH" 64(%0, %%eax, 4)	\n\t"

  1548. 			"movq (%0, %%eax, 4), %%mm0	\n\t" // YUYV YUYV(0)

  1549. 			"movq 8(%0, %%eax, 4), %%mm1	\n\t" // YUYV YUYV(4)

  1550. 			"movq 16(%0, %%eax, 4), %%mm2	\n\t" // YUYV YUYV(8)

  1551. 			"movq 24(%0, %%eax, 4), %%mm3	\n\t" // YUYV YUYV(12)

  1552. 			"pand %%mm7, %%mm0		\n\t" // Y0Y0 Y0Y0(0)

  1553. 			"pand %%mm7, %%mm1		\n\t" // Y0Y0 Y0Y0(4)

  1554. 			"pand %%mm7, %%mm2		\n\t" // Y0Y0 Y0Y0(8)

  1555. 			"pand %%mm7, %%mm3		\n\t" // Y0Y0 Y0Y0(12)

  1556. 			"packuswb %%mm1, %%mm0		\n\t" // YYYY YYYY(0)

  1557. 			"packuswb %%mm3, %%mm2		\n\t" // YYYY YYYY(8)

  1558. 

  1559. 			MOVNTQ" %%mm0, (%1, %%eax, 2)	\n\t"

  1560. 			MOVNTQ" %%mm2, 8(%1, %%eax, 2)	\n\t"

  1561. 

  1562. 			"addl $8, %%eax			\n\t"

  1563. 			"cmpl %4, %%eax			\n\t"

  1564. 			" jb 1b				\n\t"

  1565. 

  1566. 			::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (chromWidth)

  1567. 			: "memory", "%eax"

  1568. 		);

  1569. #else

  1570. 		unsigned i;

  1571. 		for(i=0; i<chromWidth; i++)

  1572. 		{

  1573. 			ydst[2*i+0] 	= src[4*i+0];

  1574. 			udst[i] 	= src[4*i+1];

  1575. 			ydst[2*i+1] 	= src[4*i+2];

  1576. 			vdst[i] 	= src[4*i+3];

  1577. 		}

  1578. 		ydst += lumStride;

  1579. 		src  += srcStride;

  1580. 

  1581. 		for(i=0; i<chromWidth; i++)

  1582. 		{

  1583. 			ydst[2*i+0] 	= src[4*i+0];

  1584. 			ydst[2*i+1] 	= src[4*i+2];

  1585. 		}

  1586. #endif

  1587. 		udst += chromStride;

  1588. 		vdst += chromStride;

  1589. 		ydst += lumStride;

  1590. 		src  += srcStride;

  1591. 	}

  1592. #ifdef HAVE_MMX

  1593. asm volatile(   EMMS" \n\t"

  1594.         	SFENCE" \n\t"

  1595.         	:::"memory");

  1596. #endif

  1597. }

  1598. 

  1599. static inline void RENAME(yvu9toyv12)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc,

  1600. 	uint8_t *ydst, uint8_t *udst, uint8_t *vdst,

  1601. 	unsigned int width, unsigned int height, unsigned int lumStride, unsigned int chromStride)

  1602. {

  1603. 	/* Y Plane */

  1604. 	memcpy(ydst, ysrc, width*height);

  1605. 

  1606. 	/* XXX: implement upscaling for U,V */

  1607. }

  1608. 

  1609. static inline void RENAME(planar2x)(const uint8_t *src, uint8_t *dst, int srcWidth, int srcHeight, int srcStride, int dstStride)

  1610. {

  1611. 	int x,y;

  1612. 	

  1613. 	// first line

  1614. 	for(x=0; x<srcWidth; x++){

  1615. 		dst[2*x+0]=

  1616. 		dst[2*x+1]= src[x];

  1617. 	}

  1618. 	dst+= dstStride;

  1619. 

  1620. 	for(y=1; y<srcHeight; y++){

  1621. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1622. 		const int mmxSize= srcWidth;

  1623. 		asm volatile(

  1624. 			"movl %4, %%eax			\n\t"

  1625. 			"1:				\n\t"

  1626. 			"movq (%0, %%eax), %%mm0	\n\t"

  1627. 			"movq (%1, %%eax), %%mm1	\n\t"

  1628. 			"movq 1(%0, %%eax), %%mm2	\n\t"

  1629. 			"movq 1(%1, %%eax), %%mm3	\n\t"

  1630. 			"movq %%mm0, %%mm4		\n\t"

  1631. 			"movq %%mm1, %%mm5		\n\t"

  1632. 			PAVGB" %%mm3, %%mm0		\n\t"

  1633. 			PAVGB" %%mm3, %%mm0		\n\t"

  1634. 			PAVGB" %%mm4, %%mm3		\n\t"

  1635. 			PAVGB" %%mm4, %%mm3		\n\t"

  1636. 			PAVGB" %%mm2, %%mm1		\n\t"

  1637. 			PAVGB" %%mm2, %%mm1		\n\t"

  1638. 			PAVGB" %%mm5, %%mm2		\n\t"

  1639. 			PAVGB" %%mm5, %%mm2		\n\t"

  1640. 			"movq %%mm3, %%mm4		\n\t"

  1641. 			"movq %%mm2, %%mm5		\n\t"

  1642. 			"punpcklbw %%mm1, %%mm3		\n\t"

  1643. 			"punpckhbw %%mm1, %%mm4		\n\t"

  1644. 			"punpcklbw %%mm0, %%mm2		\n\t"

  1645. 			"punpckhbw %%mm0, %%mm5		\n\t"

  1646. #if 1

  1647. 			MOVNTQ" %%mm3, (%2, %%eax, 2)	\n\t"

  1648. 			MOVNTQ" %%mm4, 8(%2, %%eax, 2)	\n\t"

  1649. 			MOVNTQ" %%mm2, (%3, %%eax, 2)	\n\t"

  1650. 			MOVNTQ" %%mm5, 8(%3, %%eax, 2)	\n\t"

  1651. #else

  1652. 			"movq %%mm3, (%2, %%eax, 2)	\n\t"

  1653. 			"movq %%mm4, 8(%2, %%eax, 2)	\n\t"

  1654. 			"movq %%mm2, (%3, %%eax, 2)	\n\t"

  1655. 			"movq %%mm5, 8(%3, %%eax, 2)	\n\t"

  1656. #endif

  1657. 			"addl $8, %%eax			\n\t"

  1658. 			" js 1b				\n\t"

  1659. 			:: "r" (src + mmxSize-1), "r" (src + srcStride + mmxSize-1),

  1660. 			   "r" (dst + mmxSize*2), "r" (dst + dstStride + mmxSize*2),

  1661. 			   "g" (-mmxSize)

  1662. 			: "%eax"

  1663. 

  1664. 		);

  1665. 		dst[0]= 

  1666. 		dst[dstStride]= src[0];

  1667. #else

  1668. 		dst[0]= 

  1669. 		dst[dstStride]= src[0];

  1670. 

  1671. 		for(x=0; x<srcWidth-1; x++){

  1672. 			dst[2*x          +1]= (3*src[x+0] +   src[x+srcStride+1])>>2;

  1673. 			dst[2*x+dstStride+2]= (  src[x+0] + 3*src[x+srcStride+1])>>2;

  1674. 			dst[2*x+dstStride+1]= (  src[x+1] + 3*src[x+srcStride  ])>>2;

  1675. 			dst[2*x          +2]= (3*src[x+1] +   src[x+srcStride  ])>>2;

  1676. 		}

  1677. #endif

  1678. 		dst[srcWidth*2 -1]= 

  1679. 		dst[srcWidth*2 -1 + dstStride]= src[srcWidth-1];

  1680. 

  1681. 		dst+=dstStride*2;

  1682. 		src+=srcStride;

  1683. 	}

  1684. 	src-=srcStride;

  1685. 	

  1686. 	// last line

  1687. 	for(x=0; x<srcWidth; x++){

  1688. 		dst[2*x+0]=

  1689. 		dst[2*x+1]= src[x];

  1690. 	}

  1691. #ifdef HAVE_MMX

  1692. asm volatile(   EMMS" \n\t"

  1693.         	SFENCE" \n\t"

  1694.         	:::"memory");

  1695. #endif

  1696. }

  1697. 

  1698. /**

  1699.  *

  1700.  * height should be a multiple of 2 and width should be a multiple of 16 (if this is a

  1701.  * problem for anyone then tell me, and ill fix it)

  1702.  * chrominance data is only taken from every secound line others are ignored FIXME write HQ version

  1703.  */

  1704. static inline void RENAME(uyvytoyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,

  1705. 	unsigned int width, unsigned int height,

  1706. 	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)

  1707. {

  1708. 	unsigned y;

  1709. 	const unsigned chromWidth= width>>1;

  1710. 	for(y=0; y<height; y+=2)

  1711. 	{

  1712. #ifdef HAVE_MMX

  1713. 		asm volatile(

  1714. 			"xorl %%eax, %%eax		\n\t"

  1715. 			"pcmpeqw %%mm7, %%mm7		\n\t"

  1716. 			"psrlw $8, %%mm7		\n\t" // FF,00,FF,00...

  1717. 			".balign 16			\n\t"

  1718. 			"1:				\n\t"

  1719. 			PREFETCH" 64(%0, %%eax, 4)	\n\t"

  1720. 			"movq (%0, %%eax, 4), %%mm0	\n\t" // UYVY UYVY(0)

  1721. 			"movq 8(%0, %%eax, 4), %%mm1	\n\t" // UYVY UYVY(4)

  1722. 			"movq %%mm0, %%mm2		\n\t" // UYVY UYVY(0)

  1723. 			"movq %%mm1, %%mm3		\n\t" // UYVY UYVY(4)

  1724. 			"pand %%mm7, %%mm0		\n\t" // U0V0 U0V0(0)

  1725. 			"pand %%mm7, %%mm1		\n\t" // U0V0 U0V0(4)

  1726. 			"psrlw $8, %%mm2		\n\t" // Y0Y0 Y0Y0(0)

  1727. 			"psrlw $8, %%mm3		\n\t" // Y0Y0 Y0Y0(4)

  1728. 			"packuswb %%mm1, %%mm0		\n\t" // UVUV UVUV(0)

  1729. 			"packuswb %%mm3, %%mm2		\n\t" // YYYY YYYY(0)

  1730. 

  1731. 			MOVNTQ" %%mm2, (%1, %%eax, 2)	\n\t"

  1732. 

  1733. 			"movq 16(%0, %%eax, 4), %%mm1	\n\t" // UYVY UYVY(8)

  1734. 			"movq 24(%0, %%eax, 4), %%mm2	\n\t" // UYVY UYVY(12)

  1735. 			"movq %%mm1, %%mm3		\n\t" // UYVY UYVY(8)

  1736. 			"movq %%mm2, %%mm4		\n\t" // UYVY UYVY(12)

  1737. 			"pand %%mm7, %%mm1		\n\t" // U0V0 U0V0(8)

  1738. 			"pand %%mm7, %%mm2		\n\t" // U0V0 U0V0(12)

  1739. 			"psrlw $8, %%mm3		\n\t" // Y0Y0 Y0Y0(8)

  1740. 			"psrlw $8, %%mm4		\n\t" // Y0Y0 Y0Y0(12)

  1741. 			"packuswb %%mm2, %%mm1		\n\t" // UVUV UVUV(8)

  1742. 			"packuswb %%mm4, %%mm3		\n\t" // YYYY YYYY(8)

  1743. 

  1744. 			MOVNTQ" %%mm3, 8(%1, %%eax, 2)	\n\t"

  1745. 

  1746. 			"movq %%mm0, %%mm2		\n\t" // UVUV UVUV(0)

  1747. 			"movq %%mm1, %%mm3		\n\t" // UVUV UVUV(8)

  1748. 			"psrlw $8, %%mm0		\n\t" // V0V0 V0V0(0)

  1749. 			"psrlw $8, %%mm1		\n\t" // V0V0 V0V0(8)

  1750. 			"pand %%mm7, %%mm2		\n\t" // U0U0 U0U0(0)

  1751. 			"pand %%mm7, %%mm3		\n\t" // U0U0 U0U0(8)

  1752. 			"packuswb %%mm1, %%mm0		\n\t" // VVVV VVVV(0)

  1753. 			"packuswb %%mm3, %%mm2		\n\t" // UUUU UUUU(0)

  1754. 

  1755. 			MOVNTQ" %%mm0, (%3, %%eax)	\n\t"

  1756. 			MOVNTQ" %%mm2, (%2, %%eax)	\n\t"

  1757. 

  1758. 			"addl $8, %%eax			\n\t"

  1759. 			"cmpl %4, %%eax			\n\t"

  1760. 			" jb 1b				\n\t"

  1761. 			::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (chromWidth)

  1762. 			: "memory", "%eax"

  1763. 		);

  1764. 

  1765. 		ydst += lumStride;

  1766. 		src  += srcStride;

  1767. 

  1768. 		asm volatile(

  1769. 			"xorl %%eax, %%eax		\n\t"

  1770. 			".balign 16			\n\t"

  1771. 			"1:				\n\t"

  1772. 			PREFETCH" 64(%0, %%eax, 4)	\n\t"

  1773. 			"movq (%0, %%eax, 4), %%mm0	\n\t" // YUYV YUYV(0)

  1774. 			"movq 8(%0, %%eax, 4), %%mm1	\n\t" // YUYV YUYV(4)

  1775. 			"movq 16(%0, %%eax, 4), %%mm2	\n\t" // YUYV YUYV(8)

  1776. 			"movq 24(%0, %%eax, 4), %%mm3	\n\t" // YUYV YUYV(12)

  1777. 			"psrlw $8, %%mm0		\n\t" // Y0Y0 Y0Y0(0)

  1778. 			"psrlw $8, %%mm1		\n\t" // Y0Y0 Y0Y0(4)

  1779. 			"psrlw $8, %%mm2		\n\t" // Y0Y0 Y0Y0(8)

  1780. 			"psrlw $8, %%mm3		\n\t" // Y0Y0 Y0Y0(12)

  1781. 			"packuswb %%mm1, %%mm0		\n\t" // YYYY YYYY(0)

  1782. 			"packuswb %%mm3, %%mm2		\n\t" // YYYY YYYY(8)

  1783. 

  1784. 			MOVNTQ" %%mm0, (%1, %%eax, 2)	\n\t"

  1785. 			MOVNTQ" %%mm2, 8(%1, %%eax, 2)	\n\t"

  1786. 

  1787. 			"addl $8, %%eax			\n\t"

  1788. 			"cmpl %4, %%eax			\n\t"

  1789. 			" jb 1b				\n\t"

  1790. 

  1791. 			::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "r" (chromWidth)

  1792. 			: "memory", "%eax"

  1793. 		);

  1794. #else

  1795. 		unsigned i;

  1796. 		for(i=0; i<chromWidth; i++)

  1797. 		{

  1798. 			udst[i] 	= src[4*i+0];

  1799. 			ydst[2*i+0] 	= src[4*i+1];

  1800. 			vdst[i] 	= src[4*i+2];

  1801. 			ydst[2*i+1] 	= src[4*i+3];

  1802. 		}

  1803. 		ydst += lumStride;

  1804. 		src  += srcStride;

  1805. 

  1806. 		for(i=0; i<chromWidth; i++)

  1807. 		{

  1808. 			ydst[2*i+0] 	= src[4*i+1];

  1809. 			ydst[2*i+1] 	= src[4*i+3];

  1810. 		}

  1811. #endif

  1812. 		udst += chromStride;

  1813. 		vdst += chromStride;

  1814. 		ydst += lumStride;

  1815. 		src  += srcStride;

  1816. 	}

  1817. #ifdef HAVE_MMX

  1818. asm volatile(   EMMS" \n\t"

  1819.         	SFENCE" \n\t"

  1820.         	:::"memory");

  1821. #endif

  1822. }

  1823. 

  1824. /**

  1825.  *

  1826.  * height should be a multiple of 2 and width should be a multiple of 2 (if this is a

  1827.  * problem for anyone then tell me, and ill fix it)

  1828.  * chrominance data is only taken from every secound line others are ignored in the C version FIXME write HQ version

  1829.  */

  1830. static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,

  1831. 	unsigned int width, unsigned int height,

  1832. 	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride)

  1833. {

  1834. 	unsigned y;

  1835. 	const unsigned chromWidth= width>>1;

  1836. #ifdef HAVE_MMX

  1837. 	for(y=0; y<height-2; y+=2)

  1838. 	{

  1839. 		unsigned i;

  1840. 		for(i=0; i<2; i++)

  1841. 		{

  1842. 			asm volatile(

  1843. 				"movl %2, %%eax			\n\t"

  1844. 				"movq "MANGLE(bgr2YCoeff)", %%mm6		\n\t"

  1845. 				"movq "MANGLE(w1111)", %%mm5		\n\t"

  1846. 				"pxor %%mm7, %%mm7		\n\t"

  1847. 				"leal (%%eax, %%eax, 2), %%ebx	\n\t"

  1848. 				".balign 16			\n\t"

  1849. 				"1:				\n\t"

  1850. 				PREFETCH" 64(%0, %%ebx)		\n\t"

  1851. 				"movd (%0, %%ebx), %%mm0	\n\t"

  1852. 				"movd 3(%0, %%ebx), %%mm1	\n\t"

  1853. 				"punpcklbw %%mm7, %%mm0		\n\t"

  1854. 				"punpcklbw %%mm7, %%mm1		\n\t"

  1855. 				"movd 6(%0, %%ebx), %%mm2	\n\t"

  1856. 				"movd 9(%0, %%ebx), %%mm3	\n\t"

  1857. 				"punpcklbw %%mm7, %%mm2		\n\t"

  1858. 				"punpcklbw %%mm7, %%mm3		\n\t"

  1859. 				"pmaddwd %%mm6, %%mm0		\n\t"

  1860. 				"pmaddwd %%mm6, %%mm1		\n\t"

  1861. 				"pmaddwd %%mm6, %%mm2		\n\t"

  1862. 				"pmaddwd %%mm6, %%mm3		\n\t"

  1863. #ifndef FAST_BGR2YV12

  1864. 				"psrad $8, %%mm0		\n\t"

  1865. 				"psrad $8, %%mm1		\n\t"

  1866. 				"psrad $8, %%mm2		\n\t"

  1867. 				"psrad $8, %%mm3		\n\t"

  1868. #endif

  1869. 				"packssdw %%mm1, %%mm0		\n\t"

  1870. 				"packssdw %%mm3, %%mm2		\n\t"

  1871. 				"pmaddwd %%mm5, %%mm0		\n\t"

  1872. 				"pmaddwd %%mm5, %%mm2		\n\t"

  1873. 				"packssdw %%mm2, %%mm0		\n\t"

  1874. 				"psraw $7, %%mm0		\n\t"

  1875. 

  1876. 				"movd 12(%0, %%ebx), %%mm4	\n\t"

  1877. 				"movd 15(%0, %%ebx), %%mm1	\n\t"

  1878. 				"punpcklbw %%mm7, %%mm4		\n\t"

  1879. 				"punpcklbw %%mm7, %%mm1		\n\t"

  1880. 				"movd 18(%0, %%ebx), %%mm2	\n\t"

  1881. 				"movd 21(%0, %%ebx), %%mm3	\n\t"

  1882. 				"punpcklbw %%mm7, %%mm2		\n\t"

  1883. 				"punpcklbw %%mm7, %%mm3		\n\t"

  1884. 				"pmaddwd %%mm6, %%mm4		\n\t"

  1885. 				"pmaddwd %%mm6, %%mm1		\n\t"

  1886. 				"pmaddwd %%mm6, %%mm2		\n\t"

  1887. 				"pmaddwd %%mm6, %%mm3		\n\t"

  1888. #ifndef FAST_BGR2YV12

  1889. 				"psrad $8, %%mm4		\n\t"

  1890. 				"psrad $8, %%mm1		\n\t"

  1891. 				"psrad $8, %%mm2		\n\t"

  1892. 				"psrad $8, %%mm3		\n\t"

  1893. #endif

  1894. 				"packssdw %%mm1, %%mm4		\n\t"

  1895. 				"packssdw %%mm3, %%mm2		\n\t"

  1896. 				"pmaddwd %%mm5, %%mm4		\n\t"

  1897. 				"pmaddwd %%mm5, %%mm2		\n\t"

  1898. 				"addl $24, %%ebx		\n\t"

  1899. 				"packssdw %%mm2, %%mm4		\n\t"

  1900. 				"psraw $7, %%mm4		\n\t"

  1901. 

  1902. 				"packuswb %%mm4, %%mm0		\n\t"

  1903. 				"paddusb "MANGLE(bgr2YOffset)", %%mm0	\n\t"

  1904. 

  1905. 				MOVNTQ" %%mm0, (%1, %%eax)	\n\t"

  1906. 				"addl $8, %%eax			\n\t"

  1907. 				" js 1b				\n\t"

  1908. 				: : "r" (src+width*3), "r" (ydst+width), "g" (-width)

  1909. 				: "%eax", "%ebx"

  1910. 			);

  1911. 			ydst += lumStride;

  1912. 			src  += srcStride;

  1913. 		}

  1914. 		src -= srcStride*2;

  1915. 		asm volatile(

  1916. 			"movl %4, %%eax			\n\t"

  1917. 			"movq "MANGLE(w1111)", %%mm5		\n\t"

  1918. 			"movq "MANGLE(bgr2UCoeff)", %%mm6		\n\t"

  1919. 			"pxor %%mm7, %%mm7		\n\t"

  1920. 			"leal (%%eax, %%eax, 2), %%ebx	\n\t"

  1921. 			"addl %%ebx, %%ebx		\n\t"

  1922. 			".balign 16			\n\t"

  1923. 			"1:				\n\t"

  1924. 			PREFETCH" 64(%0, %%ebx)		\n\t"

  1925. 			PREFETCH" 64(%1, %%ebx)		\n\t"

  1926. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1927. 			"movq (%0, %%ebx), %%mm0	\n\t"

  1928. 			"movq (%1, %%ebx), %%mm1	\n\t"

  1929. 			"movq 6(%0, %%ebx), %%mm2	\n\t"

  1930. 			"movq 6(%1, %%ebx), %%mm3	\n\t"

  1931. 			PAVGB" %%mm1, %%mm0		\n\t"

  1932. 			PAVGB" %%mm3, %%mm2		\n\t"

  1933. 			"movq %%mm0, %%mm1		\n\t"

  1934. 			"movq %%mm2, %%mm3		\n\t"

  1935. 			"psrlq $24, %%mm0		\n\t"

  1936. 			"psrlq $24, %%mm2		\n\t"

  1937. 			PAVGB" %%mm1, %%mm0		\n\t"

  1938. 			PAVGB" %%mm3, %%mm2		\n\t"

  1939. 			"punpcklbw %%mm7, %%mm0		\n\t"

  1940. 			"punpcklbw %%mm7, %%mm2		\n\t"

  1941. #else

  1942. 			"movd (%0, %%ebx), %%mm0	\n\t"

  1943. 			"movd (%1, %%ebx), %%mm1	\n\t"

  1944. 			"movd 3(%0, %%ebx), %%mm2	\n\t"

  1945. 			"movd 3(%1, %%ebx), %%mm3	\n\t"

  1946. 			"punpcklbw %%mm7, %%mm0		\n\t"

  1947. 			"punpcklbw %%mm7, %%mm1		\n\t"

  1948. 			"punpcklbw %%mm7, %%mm2		\n\t"

  1949. 			"punpcklbw %%mm7, %%mm3		\n\t"

  1950. 			"paddw %%mm1, %%mm0		\n\t"

  1951. 			"paddw %%mm3, %%mm2		\n\t"

  1952. 			"paddw %%mm2, %%mm0		\n\t"

  1953. 			"movd 6(%0, %%ebx), %%mm4	\n\t"

  1954. 			"movd 6(%1, %%ebx), %%mm1	\n\t"

  1955. 			"movd 9(%0, %%ebx), %%mm2	\n\t"

  1956. 			"movd 9(%1, %%ebx), %%mm3	\n\t"

  1957. 			"punpcklbw %%mm7, %%mm4		\n\t"

  1958. 			"punpcklbw %%mm7, %%mm1		\n\t"

  1959. 			"punpcklbw %%mm7, %%mm2		\n\t"

  1960. 			"punpcklbw %%mm7, %%mm3		\n\t"

  1961. 			"paddw %%mm1, %%mm4		\n\t"

  1962. 			"paddw %%mm3, %%mm2		\n\t"

  1963. 			"paddw %%mm4, %%mm2		\n\t"

  1964. 			"psrlw $2, %%mm0		\n\t"

  1965. 			"psrlw $2, %%mm2		\n\t"

  1966. #endif

  1967. 			"movq "MANGLE(bgr2VCoeff)", %%mm1		\n\t"

  1968. 			"movq "MANGLE(bgr2VCoeff)", %%mm3		\n\t"

  1969. 

  1970. 			"pmaddwd %%mm0, %%mm1		\n\t"

  1971. 			"pmaddwd %%mm2, %%mm3		\n\t"

  1972. 			"pmaddwd %%mm6, %%mm0		\n\t"

  1973. 			"pmaddwd %%mm6, %%mm2		\n\t"

  1974. #ifndef FAST_BGR2YV12

  1975. 			"psrad $8, %%mm0		\n\t"

  1976. 			"psrad $8, %%mm1		\n\t"

  1977. 			"psrad $8, %%mm2		\n\t"

  1978. 			"psrad $8, %%mm3		\n\t"

  1979. #endif

  1980. 			"packssdw %%mm2, %%mm0		\n\t"

  1981. 			"packssdw %%mm3, %%mm1		\n\t"

  1982. 			"pmaddwd %%mm5, %%mm0		\n\t"

  1983. 			"pmaddwd %%mm5, %%mm1		\n\t"

  1984. 			"packssdw %%mm1, %%mm0		\n\t" // V1 V0 U1 U0

  1985. 			"psraw $7, %%mm0		\n\t"

  1986. 

  1987. #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)

  1988. 			"movq 12(%0, %%ebx), %%mm4	\n\t"

  1989. 			"movq 12(%1, %%ebx), %%mm1	\n\t"

  1990. 			"movq 18(%0, %%ebx), %%mm2	\n\t"

  1991. 			"movq 18(%1, %%ebx), %%mm3	\n\t"

  1992. 			PAVGB" %%mm1, %%mm4		\n\t"

  1993. 			PAVGB" %%mm3, %%mm2		\n\t"

  1994. 			"movq %%mm4, %%mm1		\n\t"

  1995. 			"movq %%mm2, %%mm3		\n\t"

  1996. 			"psrlq $24, %%mm4		\n\t"

  1997. 			"psrlq $24, %%mm2		\n\t"

  1998. 			PAVGB" %%mm1, %%mm4		\n\t"

  1999. 			PAVGB" %%mm3, %%mm2		\n\t"

  2000. 			"punpcklbw %%mm7, %%mm4		\n\t"

  2001. 			"punpcklbw %%mm7, %%mm2		\n\t"

  2002. #else

  2003. 			"movd 12(%0, %%ebx), %%mm4	\n\t"

  2004. 			"movd 12(%1, %%ebx), %%mm1	\n\t"

  2005. 			"movd 15(%0, %%ebx), %%mm2	\n\t"

  2006. 			"movd 15(%1, %%ebx), %%mm3	\n\t"

  2007. 			"punpcklbw %%mm7, %%mm4		\n\t"

  2008. 			"punpcklbw %%mm7, %%mm1		\n\t"

  2009. 			"punpcklbw %%mm7, %%mm2		\n\t"

  2010. 			"punpcklbw %%mm7, %%mm3		\n\t"

  2011. 			"paddw %%mm1, %%mm4		\n\t"

  2012. 			"paddw %%mm3, %%mm2		\n\t"

  2013. 			"paddw %%mm2, %%mm4		\n\t"

  2014. 			"movd 18(%0, %%ebx), %%mm5	\n\t"

  2015. 			"movd 18(%1, %%ebx), %%mm1	\n\t"

  2016. 			"movd 21(%0, %%ebx), %%mm2	\n\t"

  2017. 			"movd 21(%1, %%ebx), %%mm3	\n\t"

  2018. 			"punpcklbw %%mm7, %%mm5		\n\t"

  2019. 			"punpcklbw %%mm7, %%mm1		\n\t"

  2020. 			"punpcklbw %%mm7, %%mm2		\n\t"

  2021. 			"punpcklbw %%mm7, %%mm3		\n\t"

  2022. 			"paddw %%mm1, %%mm5		\n\t"

  2023. 			"paddw %%mm3, %%mm2		\n\t"

  2024. 			"paddw %%mm5, %%mm2		\n\t"

  2025. 			"movq "MANGLE(w1111)", %%mm5		\n\t"

  2026. 			"psrlw $2, %%mm4		\n\t"

  2027. 			"psrlw $2, %%mm2		\n\t"

  2028. #endif

  2029. 			"movq "MANGLE(bgr2VCoeff)", %%mm1		\n\t"

  2030. 			"movq "MANGLE(bgr2VCoeff)", %%mm3		\n\t"

  2031. 

  2032. 			"pmaddwd %%mm4, %%mm1		\n\t"

  2033. 			"pmaddwd %%mm2, %%mm3		\n\t"

  2034. 			"pmaddwd %%mm6, %%mm4		\n\t"

  2035. 			"pmaddwd %%mm6, %%mm2		\n\t"

  2036. #ifndef FAST_BGR2YV12

  2037. 			"psrad $8, %%mm4		\n\t"

  2038. 			"psrad $8, %%mm1		\n\t"

  2039. 			"psrad $8, %%mm2		\n\t"

  2040. 			"psrad $8, %%mm3		\n\t"

  2041. #endif

  2042. 			"packssdw %%mm2, %%mm4		\n\t"

  2043. 			"packssdw %%mm3, %%mm1		\n\t"

  2044. 			"pmaddwd %%mm5, %%mm4		\n\t"

  2045. 			"pmaddwd %%mm5, %%mm1		\n\t"

  2046. 			"addl $24, %%ebx		\n\t"

  2047. 			"packssdw %%mm1, %%mm4		\n\t" // V3 V2 U3 U2

  2048. 			"psraw $7, %%mm4		\n\t"

  2049. 

  2050. 			"movq %%mm0, %%mm1		\n\t"

  2051. 			"punpckldq %%mm4, %%mm0		\n\t"

  2052. 			"punpckhdq %%mm4, %%mm1		\n\t"

  2053. 			"packsswb %%mm1, %%mm0		\n\t"

  2054. 			"paddb "MANGLE(bgr2UVOffset)", %%mm0	\n\t"

  2055. 

  2056. 			"movd %%mm0, (%2, %%eax)	\n\t"

  2057. 			"punpckhdq %%mm0, %%mm0		\n\t"

  2058. 			"movd %%mm0, (%3, %%eax)	\n\t"

  2059. 			"addl $4, %%eax			\n\t"

  2060. 			" js 1b				\n\t"

  2061. 			: : "r" (src+width*6), "r" (src+srcStride+width*6), "r" (udst+width), "r" (vdst+width), "g" (-width)

  2062. 			: "%eax", "%ebx"

  2063. 		);

  2064. 

  2065. 		udst += chromStride;

  2066. 		vdst += chromStride;

  2067. 		src  += srcStride*2;

  2068. 	}

  2069. 

  2070. 	asm volatile(   EMMS" \n\t"

  2071. 			SFENCE" \n\t"

  2072. 			:::"memory");

  2073. #else

  2074. 	y=0;

  2075. #endif

  2076. 	for(; y<height; y+=2)

  2077. 	{

  2078. 		unsigned i;

  2079. 		for(i=0; i<chromWidth; i++)

  2080. 		{

  2081. 			unsigned int b= src[6*i+0];

  2082. 			unsigned int g= src[6*i+1];

  2083. 			unsigned int r= src[6*i+2];

  2084. 

  2085. 			unsigned int Y  =  ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;

  2086. 			unsigned int V  =  ((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128;

  2087. 			unsigned int U  =  ((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128;

  2088. 

  2089. 			udst[i] 	= U;

  2090. 			vdst[i] 	= V;

  2091. 			ydst[2*i] 	= Y;

  2092. 

  2093. 			b= src[6*i+3];

  2094. 			g= src[6*i+4];

  2095. 			r= src[6*i+5];

  2096. 

  2097. 			Y  =  ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;

  2098. 			ydst[2*i+1] 	= Y;

  2099. 		}

  2100. 		ydst += lumStride;

  2101. 		src  += srcStride;

  2102. 

  2103. 		for(i=0; i<chromWidth; i++)

  2104. 		{

  2105. 			unsigned int b= src[6*i+0];

  2106. 			unsigned int g= src[6*i+1];

  2107. 			unsigned int r= src[6*i+2];

  2108. 

  2109. 			unsigned int Y  =  ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;

  2110. 

  2111. 			ydst[2*i] 	= Y;

  2112. 

  2113. 			b= src[6*i+3];

  2114. 			g= src[6*i+4];

  2115. 			r= src[6*i+5];

  2116. 

  2117. 			Y  =  ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;

  2118. 			ydst[2*i+1] 	= Y;

  2119. 		}

  2120. 		udst += chromStride;

  2121. 		vdst += chromStride;

  2122. 		ydst += lumStride;

  2123. 		src  += srcStride;

  2124. 	}

  2125. }

  2126. 

  2127. void RENAME(interleaveBytes)(uint8_t *src1, uint8_t *src2, uint8_t *dest,

  2128. 			    unsigned width, unsigned height, unsigned src1Stride,

  2129. 			    unsigned src2Stride, unsigned dstStride){

  2130. 	unsigned h;

  2131. 

  2132. 	for(h=0; h < height; h++)

  2133. 	{

  2134. 		unsigned w;

  2135. 

  2136. #ifdef HAVE_MMX

  2137. #ifdef HAVE_SSE2

  2138. 		asm(

  2139. 			"xorl %%eax, %%eax		\n\t"

  2140. 			"1:				\n\t"

  2141. 			PREFETCH" 64(%1, %%eax)		\n\t"

  2142. 			PREFETCH" 64(%2, %%eax)		\n\t"

  2143. 			"movdqa (%1, %%eax), %%xmm0	\n\t"

  2144. 			"movdqa (%1, %%eax), %%xmm1	\n\t"

  2145. 			"movdqa (%2, %%eax), %%xmm2	\n\t"

  2146. 			"punpcklbw %%xmm2, %%xmm0	\n\t"

  2147. 			"punpckhbw %%xmm2, %%xmm1	\n\t"

  2148. 			"movntdq %%xmm0, (%0, %%eax, 2)	\n\t"

  2149. 			"movntdq %%xmm1, 16(%0, %%eax, 2)\n\t"

  2150. 			"addl $16, %%eax			\n\t"

  2151. 			"cmpl %3, %%eax			\n\t"

  2152. 			" jb 1b				\n\t"

  2153. 			::"r"(dest), "r"(src1), "r"(src2), "r" (width-15)

  2154. 			: "memory", "%eax"

  2155. 		);

  2156. #else

  2157. 		asm(

  2158. 			"xorl %%eax, %%eax		\n\t"

  2159. 			"1:				\n\t"

  2160. 			PREFETCH" 64(%1, %%eax)		\n\t"

  2161. 			PREFETCH" 64(%2, %%eax)		\n\t"

  2162. 			"movq (%1, %%eax), %%mm0	\n\t"

  2163. 			"movq 8(%1, %%eax), %%mm2	\n\t"

  2164. 			"movq %%mm0, %%mm1		\n\t"

  2165. 			"movq %%mm2, %%mm3		\n\t"

  2166. 			"movq (%2, %%eax), %%mm4	\n\t"

  2167. 			"movq 8(%2, %%eax), %%mm5	\n\t"

  2168. 			"punpcklbw %%mm4, %%mm0		\n\t"

  2169. 			"punpckhbw %%mm4, %%mm1		\n\t"

  2170. 			"punpcklbw %%mm5, %%mm2		\n\t"

  2171. 			"punpckhbw %%mm5, %%mm3		\n\t"

  2172. 			MOVNTQ" %%mm0, (%0, %%eax, 2)	\n\t"

  2173. 			MOVNTQ" %%mm1, 8(%0, %%eax, 2)	\n\t"

  2174. 			MOVNTQ" %%mm2, 16(%0, %%eax, 2)	\n\t"

  2175. 			MOVNTQ" %%mm3, 24(%0, %%eax, 2)	\n\t"

  2176. 			"addl $16, %%eax			\n\t"

  2177. 			"cmpl %3, %%eax			\n\t"

  2178. 			" jb 1b				\n\t"

  2179. 			::"r"(dest), "r"(src1), "r"(src2), "r" (width-15)

  2180. 			: "memory", "%eax"

  2181. 		);

  2182. #endif

  2183. 		for(w= (width&(~15)); w < width; w++)

  2184. 		{

  2185. 			dest[2*w+0] = src1[w];

  2186. 			dest[2*w+1] = src2[w];

  2187. 		}

  2188. #else

  2189. 		for(w=0; w < width; w++)

  2190. 		{

  2191. 			dest[2*w+0] = src1[w];

  2192. 			dest[2*w+1] = src2[w];

  2193. 		}

  2194. #endif

  2195. 		dest += dstStride;

  2196.                 src1 += src1Stride;

  2197.                 src2 += src2Stride;

  2198. 	}

  2199. #ifdef HAVE_MMX

  2200. 	asm(

  2201. 		EMMS" \n\t"

  2202. 		SFENCE" \n\t"

  2203. 		::: "memory"

  2204. 		);

  2205. #endif

  2206. }

  2207. 

  2208. static inline void RENAME(vu9_to_vu12)(const uint8_t *src1, const uint8_t *src2,

  2209. 			uint8_t *dst1, uint8_t *dst2,

  2210. 			unsigned width, unsigned height,

  2211. 			unsigned srcStride1, unsigned srcStride2,

  2212. 			unsigned dstStride1, unsigned dstStride2)

  2213. {

  2214.     unsigned y,x,w,h;

  2215.     w=width/2; h=height/2;

  2216. #ifdef HAVE_MMX

  2217.     asm volatile(

  2218. 	PREFETCH" %0\n\t"

  2219. 	PREFETCH" %1\n\t"

  2220. 	::"m"(*(src1+srcStride1)),"m"(*(src2+srcStride2)):"memory");

  2221. #endif

  2222.     for(y=0;y<h;y++){

  2223. 	const uint8_t* s1=src1+srcStride1*(y>>1);

  2224. 	uint8_t* d=dst1+dstStride1*y;

  2225. 	x=0;

  2226. #ifdef HAVE_MMX

  2227. 	if(w > 32)

  2228. 	for(;x<w;x+=32)

  2229. 	{

  2230. 	    asm volatile(

  2231. 		PREFETCH" 32%1\n\t"

  2232. 	        "movq	%1, %%mm0\n\t"

  2233. 	        "movq	8%1, %%mm2\n\t"

  2234. 	        "movq	16%1, %%mm4\n\t"

  2235. 	        "movq	24%1, %%mm6\n\t"

  2236. 	        "movq	%%mm0, %%mm1\n\t"

  2237. 	        "movq	%%mm2, %%mm3\n\t"

  2238. 	        "movq	%%mm4, %%mm5\n\t"

  2239. 	        "movq	%%mm6, %%mm7\n\t"

  2240. 		"punpcklbw %%mm0, %%mm0\n\t"

  2241. 		"punpckhbw %%mm1, %%mm1\n\t"

  2242. 		"punpcklbw %%mm2, %%mm2\n\t"

  2243. 		"punpckhbw %%mm3, %%mm3\n\t"

  2244. 		"punpcklbw %%mm4, %%mm4\n\t"

  2245. 		"punpckhbw %%mm5, %%mm5\n\t"

  2246. 		"punpcklbw %%mm6, %%mm6\n\t"

  2247. 		"punpckhbw %%mm7, %%mm7\n\t"

  2248. 		MOVNTQ"	%%mm0, %0\n\t"

  2249. 		MOVNTQ"	%%mm1, 8%0\n\t"

  2250. 		MOVNTQ"	%%mm2, 16%0\n\t"

  2251. 		MOVNTQ"	%%mm3, 24%0\n\t"

  2252. 		MOVNTQ"	%%mm4, 32%0\n\t"

  2253. 		MOVNTQ"	%%mm5, 40%0\n\t"

  2254. 		MOVNTQ"	%%mm6, 48%0\n\t"

  2255. 		MOVNTQ"	%%mm7, 56%0"

  2256. 		:"=m"(d[2*x])

  2257. 		:"m"(s1[x])

  2258. 		:"memory");

  2259. 	}

  2260. #endif

  2261. 	for(;x<w;x++) d[2*x]=d[2*x+1]=s1[x];

  2262.     }

  2263.     for(y=0;y<h;y++){

  2264. 	const uint8_t* s2=src2+srcStride2*(y>>1);

  2265. 	uint8_t* d=dst2+dstStride2*y;

  2266. 	x=0;

  2267. #ifdef HAVE_MMX

  2268. 	if(w > 32)

  2269. 	for(;x<w;x+=32)

  2270. 	{

  2271. 	    asm volatile(

  2272. 		PREFETCH" 32%1\n\t"

  2273. 	        "movq	%1, %%mm0\n\t"

  2274. 	        "movq	8%1, %%mm2\n\t"

  2275. 	        "movq	16%1, %%mm4\n\t"

  2276. 	        "movq	24%1, %%mm6\n\t"

  2277. 	        "movq	%%mm0, %%mm1\n\t"

  2278. 	        "movq	%%mm2, %%mm3\n\t"

  2279. 	        "movq	%%mm4, %%mm5\n\t"

  2280. 	        "movq	%%mm6, %%mm7\n\t"

  2281. 		"punpcklbw %%mm0, %%mm0\n\t"

  2282. 		"punpckhbw %%mm1, %%mm1\n\t"

  2283. 		"punpcklbw %%mm2, %%mm2\n\t"

  2284. 		"punpckhbw %%mm3, %%mm3\n\t"

  2285. 		"punpcklbw %%mm4, %%mm4\n\t"

  2286. 		"punpckhbw %%mm5, %%mm5\n\t"

  2287. 		"punpcklbw %%mm6, %%mm6\n\t"

  2288. 		"punpckhbw %%mm7, %%mm7\n\t"

  2289. 		MOVNTQ"	%%mm0, %0\n\t"

  2290. 		MOVNTQ"	%%mm1, 8%0\n\t"

  2291. 		MOVNTQ"	%%mm2, 16%0\n\t"

  2292. 		MOVNTQ"	%%mm3, 24%0\n\t"

  2293. 		MOVNTQ"	%%mm4, 32%0\n\t"

  2294. 		MOVNTQ"	%%mm5, 40%0\n\t"

  2295. 		MOVNTQ"	%%mm6, 48%0\n\t"

  2296. 		MOVNTQ"	%%mm7, 56%0"

  2297. 		:"=m"(d[2*x])

  2298. 		:"m"(s2[x])

  2299. 		:"memory");

  2300. 	}

  2301. #endif

  2302. 	for(;x<w;x++) d[2*x]=d[2*x+1]=s2[x];

  2303.     }

  2304. #ifdef HAVE_MMX

  2305. 	asm(

  2306. 		EMMS" \n\t"

  2307. 		SFENCE" \n\t"

  2308. 		::: "memory"

  2309. 		);

  2310. #endif

  2311. }

  2312. 

  2313. static inline void RENAME(yvu9_to_yuy2)(const uint8_t *src1, const uint8_t *src2, const uint8_t *src3,

  2314. 			uint8_t *dst,

  2315. 			unsigned width, unsigned height,

  2316. 			unsigned srcStride1, unsigned srcStride2,

  2317. 			unsigned srcStride3, unsigned dstStride)

  2318. {

  2319.     unsigned y,x,x2,w,h;

  2320.     w=width/2; h=height;

  2321. #ifdef HAVE_MMX

  2322.     asm volatile(

  2323. 	PREFETCH" %0\n\t"

  2324. 	PREFETCH" %1\n\t"

  2325. 	PREFETCH" %2\n\t"

  2326. 	::"m"(*(src1+srcStride1)),"m"(*(src2+srcStride2)),"m"(*(src3+srcStride3)):"memory");

  2327. #endif

  2328.     for(y=0;y<h;y++){

  2329. 	const uint8_t* yp=src1+srcStride1*y;

  2330. 	const uint8_t* up=src2+srcStride2*(y>>2);

  2331. 	const uint8_t* vp=src3+srcStride3*(y>>2);

  2332. 	uint8_t* d=dst+dstStride*y;

  2333. 	x2=0;

  2334. 	x=0;

  2335. #ifdef HAVE_MMX

  2336. 	for(;x<w;x+=8,x2+=32)

  2337. 	{

  2338. 	    asm volatile(

  2339. 		PREFETCH" 32%1\n\t"

  2340. 		PREFETCH" 32%2\n\t"

  2341. 		PREFETCH" 32%3\n\t"

  2342. 		"movq	%1, %%mm0\n\t"       /* Y0Y1Y2Y3Y4Y5Y6Y7 */

  2343. 		"movq	%2, %%mm1\n\t"       /* U0U1U2U3U4U5U6U7 */

  2344. 		"movq	%3, %%mm2\n\t"	     /* V0V1V2V3V4V5V6V7 */

  2345. 		"movq	%%mm0, %%mm3\n\t"    /* Y0Y1Y2Y3Y4Y5Y6Y7 */

  2346. 		"movq	%%mm1, %%mm4\n\t"    /* U0U1U2U3U4U5U6U7 */

  2347. 		"movq	%%mm2, %%mm5\n\t"    /* V0V1V2V3V4V5V6V7 */

  2348. 		"punpcklbw %%mm1, %%mm1\n\t" /* U0U0 U1U1 U2U2 U3U3 */

  2349. 		"punpcklbw %%mm2, %%mm2\n\t" /* V0V0 V1V1 V2V2 V3V3 */

  2350. 		"punpckhbw %%mm4, %%mm4\n\t" /* U4U4 U5U5 U6U6 U7U7 */

  2351. 		"punpckhbw %%mm5, %%mm5\n\t" /* V4V4 V5V5 V6V6 V7V7 */

  2352. 

  2353. 		"movq	%%mm1, %%mm6\n\t"

  2354. 		"punpcklbw %%mm2, %%mm1\n\t" /* U0V0 U0V0 U1V1 U1V1*/

  2355. 		"punpcklbw %%mm1, %%mm0\n\t" /* Y0U0 Y1V0 Y2U0 Y3V0*/

  2356. 		"punpckhbw %%mm1, %%mm3\n\t" /* Y4U1 Y5V1 Y6U1 Y7V1*/

  2357. 		MOVNTQ"	%%mm0, %0\n\t"

  2358. 		MOVNTQ"	%%mm3, 8%0\n\t"

  2359. 		

  2360. 		"punpckhbw %%mm2, %%mm6\n\t" /* U2V2 U2V2 U3V3 U3V3*/

  2361. 		"movq	8%1, %%mm0\n\t"

  2362. 		"movq	%%mm0, %%mm3\n\t"

  2363. 		"punpcklbw %%mm6, %%mm0\n\t" /* Y U2 Y V2 Y U2 Y V2*/

  2364. 		"punpckhbw %%mm6, %%mm3\n\t" /* Y U3 Y V3 Y U3 Y V3*/

  2365. 		MOVNTQ"	%%mm0, 16%0\n\t"

  2366. 		MOVNTQ"	%%mm3, 24%0\n\t"

  2367. 

  2368. 		"movq	%%mm4, %%mm6\n\t"

  2369. 		"movq	16%1, %%mm0\n\t"

  2370. 		"movq	%%mm0, %%mm3\n\t"

  2371. 		"punpcklbw %%mm5, %%mm4\n\t"

  2372. 		"punpcklbw %%mm4, %%mm0\n\t" /* Y U4 Y V4 Y U4 Y V4*/

  2373. 		"punpckhbw %%mm4, %%mm3\n\t" /* Y U5 Y V5 Y U5 Y V5*/

  2374. 		MOVNTQ"	%%mm0, 32%0\n\t"

  2375. 		MOVNTQ"	%%mm3, 40%0\n\t"

  2376. 		

  2377. 		"punpckhbw %%mm5, %%mm6\n\t"

  2378. 		"movq	24%1, %%mm0\n\t"

  2379. 		"movq	%%mm0, %%mm3\n\t"

  2380. 		"punpcklbw %%mm6, %%mm0\n\t" /* Y U6 Y V6 Y U6 Y V6*/

  2381. 		"punpckhbw %%mm6, %%mm3\n\t" /* Y U7 Y V7 Y U7 Y V7*/

  2382. 		MOVNTQ"	%%mm0, 48%0\n\t"

  2383. 		MOVNTQ"	%%mm3, 56%0\n\t"

  2384. 

  2385. 		:"=m"(d[8*x])

  2386. 		:"m"(yp[x2]),"m"(up[x]),"m"(vp[x])

  2387. 		:"memory");

  2388. 	}

  2389. #endif

  2390. 	for(;x<w;x++,x2+=4)

  2391. 	{

  2392. 	    d[8*x+0]=yp[x2];

  2393. 	    d[8*x+1]=up[x];

  2394. 	    d[8*x+2]=yp[x2+1];

  2395. 	    d[8*x+3]=vp[x];

  2396. 	    d[8*x+4]=yp[x2+2];

  2397. 	    d[8*x+5]=up[x];

  2398. 	    d[8*x+6]=yp[x2+3];

  2399. 	    d[8*x+7]=vp[x];

  2400. 	}

  2401.     }

  2402. #ifdef HAVE_MMX

  2403. 	asm(

  2404. 		EMMS" \n\t"

  2405. 		SFENCE" \n\t"

  2406. 		::: "memory"

  2407. 		);

  2408. #endif

  2409. }