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.
410 Commits
32 Branches
325MB
Tree: 370799068d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '370799068d'
${ noResults }
FFmpeg/postproc/swscale.c

1729 lines
47KB
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.   supported Input formats: YV12, I420, IYUV, YUY2, BGR32, BGR24, Y8, Y800

  21.   supported output formats: YV12, I420, IYUV, BGR15, BGR16, BGR24, BGR32 (grayscale soon too)

  22.   BGR15/16 support dithering

  23. */

  24. 

  25. #include <inttypes.h>

  26. #include <string.h>

  27. #include <math.h>

  28. #include <stdio.h>

  29. #include "../config.h"

  30. #include "../mangle.h"

  31. #ifdef HAVE_MALLOC_H

  32. #include <malloc.h>

  33. #endif

  34. #include "swscale.h"

  35. #include "../cpudetect.h"

  36. #include "../libvo/img_format.h"

  37. #include "rgb2rgb.h"

  38. #undef MOVNTQ

  39. #undef PAVGB

  40. 

  41. //#undef HAVE_MMX2

  42. //#define HAVE_3DNOW

  43. //#undef HAVE_MMX

  44. //#undef ARCH_X86

  45. #define DITHER1XBPP

  46. 

  47. #define RET 0xC3 //near return opcode for X86

  48. 

  49. #ifdef MP_DEBUG

  50. #define ASSERT(x) if(!(x)) { printf("ASSERT " #x " failed\n"); *((int*)0)=0; }

  51. #else

  52. #define ASSERT(x) ;

  53. #endif

  54. 

  55. #ifdef M_PI

  56. #define PI M_PI

  57. #else

  58. #define PI 3.14159265358979323846

  59. #endif

  60. 

  61. //FIXME replace this with something faster

  62. #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420)

  63. #define isYUV(x)       ((x)==IMGFMT_YUY2 || isPlanarYUV(x))

  64. #define isHalfChrV(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420)

  65. #define isHalfChrH(x)  ((x)==IMGFMT_YUY2 || (x)==IMGFMT_YV12 || (x)==IMGFMT_I420)

  66. #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24)

  67. #define isGray(x)      ((x)==IMGFMT_Y800)

  68. #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YUY2 \

  69. 			|| (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24\

  70. 			|| (x)==IMGFMT_Y800)

  71. #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 \

  72. 			|| (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15)

  73. #define isBGR(x)       ((x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15)

  74. 

  75. #define RGB2YUV_SHIFT 16

  76. #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))

  77. #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))

  78. #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))

  79. #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))

  80. #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))

  81. #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))

  82. #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))

  83. #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))

  84. #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))

  85. 

  86. extern int verbose; // defined in mplayer.c

  87. /*

  88. NOTES

  89. 

  90. known BUGS with known cause (no bugreports please!, but patches are welcome :) )

  91. horizontal fast_bilinear MMX2 scaler reads 1-7 samples too much (might cause a sig11)

  92. 

  93. Special versions: fast Y 1:1 scaling (no interpolation in y direction)

  94. 

  95. TODO

  96. more intelligent missalignment avoidance for the horizontal scaler

  97. write special vertical cubic upscale version

  98. Optimize C code (yv12 / minmax)

  99. add support for packed pixel yuv input & output

  100. add support for Y8 output

  101. optimize bgr24 & bgr32

  102. add BGR4 output support

  103. write special BGR->BGR scaler

  104. deglobalize yuv2rgb*.c

  105. */

  106. 

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

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

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

  110. 

  111. #ifdef ARCH_X86

  112. #define CAN_COMPILE_X86_ASM

  113. #endif

  114. 

  115. #ifdef CAN_COMPILE_X86_ASM

  116. static uint64_t __attribute__((aligned(8))) yCoeff=    0x2568256825682568LL;

  117. static uint64_t __attribute__((aligned(8))) vrCoeff=   0x3343334333433343LL;

  118. static uint64_t __attribute__((aligned(8))) ubCoeff=   0x40cf40cf40cf40cfLL;

  119. static uint64_t __attribute__((aligned(8))) vgCoeff=   0xE5E2E5E2E5E2E5E2LL;

  120. static uint64_t __attribute__((aligned(8))) ugCoeff=   0xF36EF36EF36EF36ELL;

  121. static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;

  122. static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;

  123. static uint64_t __attribute__((aligned(8))) w400=      0x0400040004000400LL;

  124. static uint64_t __attribute__((aligned(8))) w80=       0x0080008000800080LL;

  125. static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;

  126. static uint64_t __attribute__((aligned(8))) w02=       0x0002000200020002LL;

  127. static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;

  128. static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;

  129. static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;

  130. static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;

  131. 

  132. static volatile uint64_t __attribute__((aligned(8))) b5Dither;

  133. static volatile uint64_t __attribute__((aligned(8))) g5Dither;

  134. static volatile uint64_t __attribute__((aligned(8))) g6Dither;

  135. static volatile uint64_t __attribute__((aligned(8))) r5Dither;

  136. 

  137. static uint64_t __attribute__((aligned(8))) dither4[2]={

  138. 	0x0103010301030103LL,

  139. 	0x0200020002000200LL,};

  140. 

  141. static uint64_t __attribute__((aligned(8))) dither8[2]={

  142. 	0x0602060206020602LL,

  143. 	0x0004000400040004LL,};

  144. 

  145. static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;

  146. static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;

  147. static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;

  148. static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;

  149. static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;

  150. static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;

  151. 

  152. static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;

  153. static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;

  154. static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;

  155. 

  156. // FIXME remove

  157. static uint64_t __attribute__((aligned(8))) asm_yalpha1;

  158. static uint64_t __attribute__((aligned(8))) asm_uvalpha1;

  159. #endif

  160. 

  161. // clipping helper table for C implementations:

  162. static unsigned char clip_table[768];

  163. 

  164. static unsigned short clip_table16b[768];

  165. static unsigned short clip_table16g[768];

  166. static unsigned short clip_table16r[768];

  167. static unsigned short clip_table15b[768];

  168. static unsigned short clip_table15g[768];

  169. static unsigned short clip_table15r[768];

  170. 

  171. // yuv->rgb conversion tables:

  172. static    int yuvtab_2568[256];

  173. static    int yuvtab_3343[256];

  174. static    int yuvtab_0c92[256];

  175. static    int yuvtab_1a1e[256];

  176. static    int yuvtab_40cf[256];

  177. // Needed for cubic scaler to catch overflows

  178. static    int clip_yuvtab_2568[768];

  179. static    int clip_yuvtab_3343[768];

  180. static    int clip_yuvtab_0c92[768];

  181. static    int clip_yuvtab_1a1e[768];

  182. static    int clip_yuvtab_40cf[768];

  183. 

  184. //global sws_flags from the command line

  185. int sws_flags=2;

  186. 

  187. //global srcFilter

  188. SwsFilter src_filter= {NULL, NULL, NULL, NULL};

  189. 

  190. float sws_lum_gblur= 0.0;

  191. float sws_chr_gblur= 0.0;

  192. int sws_chr_vshift= 0;

  193. int sws_chr_hshift= 0;

  194. float sws_chr_sharpen= 0.0;

  195. float sws_lum_sharpen= 0.0;

  196. 

  197. /* cpuCaps combined from cpudetect and whats actually compiled in

  198.    (if there is no support for something compiled in it wont appear here) */

  199. static CpuCaps cpuCaps;

  200. 

  201. void (*swScale)(SwsContext *context, uint8_t* src[], int srcStride[], int srcSliceY,

  202.              int srcSliceH, uint8_t* dst[], int dstStride[])=NULL;

  203. 

  204. static SwsVector *getConvVec(SwsVector *a, SwsVector *b);

  205. 

  206. #ifdef CAN_COMPILE_X86_ASM

  207. void in_asm_used_var_warning_killer()

  208. {

  209.  volatile int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+

  210.  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+asm_yalpha1+ asm_uvalpha1+

  211.  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;

  212.  if(i) i=0;

  213. }

  214. #endif

  215. 

  216. static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,

  217. 				    int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,

  218. 				    uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW)

  219. {

  220. 	//FIXME Optimize (just quickly writen not opti..)

  221. 	int i;

  222. 	for(i=0; i<dstW; i++)

  223. 	{

  224. 		int val=0;

  225. 		int j;

  226. 		for(j=0; j<lumFilterSize; j++)

  227. 			val += lumSrc[j][i] * lumFilter[j];

  228. 

  229. 		dest[i]= MIN(MAX(val>>19, 0), 255);

  230. 	}

  231. 

  232. 	if(uDest != NULL)

  233. 		for(i=0; i<(dstW>>1); i++)

  234. 		{

  235. 			int u=0;

  236. 			int v=0;

  237. 			int j;

  238. 			for(j=0; j<chrFilterSize; j++)

  239. 			{

  240. 				u += chrSrc[j][i] * chrFilter[j];

  241. 				v += chrSrc[j][i + 2048] * chrFilter[j];

  242. 			}

  243. 

  244. 			uDest[i]= MIN(MAX(u>>19, 0), 255);

  245. 			vDest[i]= MIN(MAX(v>>19, 0), 255);

  246. 		}

  247. }

  248. 

  249. static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,

  250. 				    int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,

  251. 				    uint8_t *dest, int dstW, int dstFormat)

  252. {

  253. 	if(dstFormat==IMGFMT_BGR32)

  254. 	{

  255. 		int i;

  256. 		for(i=0; i<(dstW>>1); i++){

  257. 			int j;

  258. 			int Y1=0;

  259. 			int Y2=0;

  260. 			int U=0;

  261. 			int V=0;

  262. 			int Cb, Cr, Cg;

  263. 			for(j=0; j<lumFilterSize; j++)

  264. 			{

  265. 				Y1 += lumSrc[j][2*i] * lumFilter[j];

  266. 				Y2 += lumSrc[j][2*i+1] * lumFilter[j];

  267. 			}

  268. 			for(j=0; j<chrFilterSize; j++)

  269. 			{

  270. 				U += chrSrc[j][i] * chrFilter[j];

  271. 				V += chrSrc[j][i+2048] * chrFilter[j];

  272. 			}

  273. 			Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];

  274. 			Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];

  275. 			U >>= 19;

  276. 			V >>= 19;

  277. 

  278. 			Cb= clip_yuvtab_40cf[U+ 256];

  279. 			Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];

  280. 			Cr= clip_yuvtab_3343[V+ 256];

  281. 

  282. 			dest[8*i+0]=clip_table[((Y1 + Cb) >>13)];

  283. 			dest[8*i+1]=clip_table[((Y1 + Cg) >>13)];

  284. 			dest[8*i+2]=clip_table[((Y1 + Cr) >>13)];

  285. 

  286. 			dest[8*i+4]=clip_table[((Y2 + Cb) >>13)];

  287. 			dest[8*i+5]=clip_table[((Y2 + Cg) >>13)];

  288. 			dest[8*i+6]=clip_table[((Y2 + Cr) >>13)];

  289. 		}

  290. 	}

  291. 	else if(dstFormat==IMGFMT_BGR24)

  292. 	{

  293. 		int i;

  294. 		for(i=0; i<(dstW>>1); i++){

  295. 			int j;

  296. 			int Y1=0;

  297. 			int Y2=0;

  298. 			int U=0;

  299. 			int V=0;

  300. 			int Cb, Cr, Cg;

  301. 			for(j=0; j<lumFilterSize; j++)

  302. 			{

  303. 				Y1 += lumSrc[j][2*i] * lumFilter[j];

  304. 				Y2 += lumSrc[j][2*i+1] * lumFilter[j];

  305. 			}

  306. 			for(j=0; j<chrFilterSize; j++)

  307. 			{

  308. 				U += chrSrc[j][i] * chrFilter[j];

  309. 				V += chrSrc[j][i+2048] * chrFilter[j];

  310. 			}

  311. 			Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];

  312. 			Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];

  313. 			U >>= 19;

  314. 			V >>= 19;

  315. 

  316. 			Cb= clip_yuvtab_40cf[U+ 256];

  317. 			Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];

  318. 			Cr= clip_yuvtab_3343[V+ 256];

  319. 

  320. 			dest[0]=clip_table[((Y1 + Cb) >>13)];

  321. 			dest[1]=clip_table[((Y1 + Cg) >>13)];

  322. 			dest[2]=clip_table[((Y1 + Cr) >>13)];

  323. 

  324. 			dest[3]=clip_table[((Y2 + Cb) >>13)];

  325. 			dest[4]=clip_table[((Y2 + Cg) >>13)];

  326. 			dest[5]=clip_table[((Y2 + Cr) >>13)];

  327. 			dest+=6;

  328. 		}

  329. 	}

  330. 	else if(dstFormat==IMGFMT_BGR16)

  331. 	{

  332. 		int i;

  333. #ifdef DITHER1XBPP

  334. 		static int ditherb1=1<<14;

  335. 		static int ditherg1=1<<13;

  336. 		static int ditherr1=2<<14;

  337. 		static int ditherb2=3<<14;

  338. 		static int ditherg2=3<<13;

  339. 		static int ditherr2=0<<14;

  340. 

  341. 		ditherb1 ^= (1^2)<<14;

  342. 		ditherg1 ^= (1^2)<<13;

  343. 		ditherr1 ^= (1^2)<<14;

  344. 		ditherb2 ^= (3^0)<<14;

  345. 		ditherg2 ^= (3^0)<<13;

  346. 		ditherr2 ^= (3^0)<<14;

  347. #else

  348. 		const int ditherb1=0;

  349. 		const int ditherg1=0;

  350. 		const int ditherr1=0;

  351. 		const int ditherb2=0;

  352. 		const int ditherg2=0;

  353. 		const int ditherr2=0;

  354. #endif

  355. 		for(i=0; i<(dstW>>1); i++){

  356. 			int j;

  357. 			int Y1=0;

  358. 			int Y2=0;

  359. 			int U=0;

  360. 			int V=0;

  361. 			int Cb, Cr, Cg;

  362. 			for(j=0; j<lumFilterSize; j++)

  363. 			{

  364. 				Y1 += lumSrc[j][2*i] * lumFilter[j];

  365. 				Y2 += lumSrc[j][2*i+1] * lumFilter[j];

  366. 			}

  367. 			for(j=0; j<chrFilterSize; j++)

  368. 			{

  369. 				U += chrSrc[j][i] * chrFilter[j];

  370. 				V += chrSrc[j][i+2048] * chrFilter[j];

  371. 			}

  372. 			Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];

  373. 			Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];

  374. 			U >>= 19;

  375. 			V >>= 19;

  376. 

  377. 			Cb= clip_yuvtab_40cf[U+ 256];

  378. 			Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];

  379. 			Cr= clip_yuvtab_3343[V+ 256];

  380. 

  381. 			((uint16_t*)dest)[2*i] =

  382. 				clip_table16b[(Y1 + Cb + ditherb1) >>13] |

  383. 				clip_table16g[(Y1 + Cg + ditherg1) >>13] |

  384. 				clip_table16r[(Y1 + Cr + ditherr1) >>13];

  385. 

  386. 			((uint16_t*)dest)[2*i+1] =

  387. 				clip_table16b[(Y2 + Cb + ditherb2) >>13] |

  388. 				clip_table16g[(Y2 + Cg + ditherg2) >>13] |

  389. 				clip_table16r[(Y2 + Cr + ditherr2) >>13];

  390. 		}

  391. 	}

  392. 	else if(dstFormat==IMGFMT_BGR15)

  393. 	{

  394. 		int i;

  395. #ifdef DITHER1XBPP

  396. 		static int ditherb1=1<<14;

  397. 		static int ditherg1=1<<14;

  398. 		static int ditherr1=2<<14;

  399. 		static int ditherb2=3<<14;

  400. 		static int ditherg2=3<<14;

  401. 		static int ditherr2=0<<14;

  402. 

  403. 		ditherb1 ^= (1^2)<<14;

  404. 		ditherg1 ^= (1^2)<<14;

  405. 		ditherr1 ^= (1^2)<<14;

  406. 		ditherb2 ^= (3^0)<<14;

  407. 		ditherg2 ^= (3^0)<<14;

  408. 		ditherr2 ^= (3^0)<<14;

  409. #else

  410. 		const int ditherb1=0;

  411. 		const int ditherg1=0;

  412. 		const int ditherr1=0;

  413. 		const int ditherb2=0;

  414. 		const int ditherg2=0;

  415. 		const int ditherr2=0;

  416. #endif

  417. 		for(i=0; i<(dstW>>1); i++){

  418. 			int j;

  419. 			int Y1=0;

  420. 			int Y2=0;

  421. 			int U=0;

  422. 			int V=0;

  423. 			int Cb, Cr, Cg;

  424. 			for(j=0; j<lumFilterSize; j++)

  425. 			{

  426. 				Y1 += lumSrc[j][2*i] * lumFilter[j];

  427. 				Y2 += lumSrc[j][2*i+1] * lumFilter[j];

  428. 			}

  429. 			for(j=0; j<chrFilterSize; j++)

  430. 			{

  431. 				U += chrSrc[j][i] * chrFilter[j];

  432. 				V += chrSrc[j][i+2048] * chrFilter[j];

  433. 			}

  434. 			Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];

  435. 			Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];

  436. 			U >>= 19;

  437. 			V >>= 19;

  438. 

  439. 			Cb= clip_yuvtab_40cf[U+ 256];

  440. 			Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];

  441. 			Cr= clip_yuvtab_3343[V+ 256];

  442. 

  443. 			((uint16_t*)dest)[2*i] =

  444. 				clip_table15b[(Y1 + Cb + ditherb1) >>13] |

  445. 				clip_table15g[(Y1 + Cg + ditherg1) >>13] |

  446. 				clip_table15r[(Y1 + Cr + ditherr1) >>13];

  447. 

  448. 			((uint16_t*)dest)[2*i+1] =

  449. 				clip_table15b[(Y2 + Cb + ditherb2) >>13] |

  450. 				clip_table15g[(Y2 + Cg + ditherg2) >>13] |

  451. 				clip_table15r[(Y2 + Cr + ditherr2) >>13];

  452. 		}

  453. 	}

  454. }

  455. 

  456. 

  457. //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one

  458. //Plain C versions

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

  460. #define COMPILE_C

  461. #endif

  462. 

  463. #ifdef CAN_COMPILE_X86_ASM

  464. 

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

  466. #define COMPILE_MMX

  467. #endif

  468. 

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

  470. #define COMPILE_MMX2

  471. #endif

  472. 

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

  474. #define COMPILE_3DNOW

  475. #endif

  476. #endif //CAN_COMPILE_X86_ASM

  477. 

  478. #undef HAVE_MMX

  479. #undef HAVE_MMX2

  480. #undef HAVE_3DNOW

  481. 

  482. #ifdef COMPILE_C

  483. #undef HAVE_MMX

  484. #undef HAVE_MMX2

  485. #undef HAVE_3DNOW

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

  487. #include "swscale_template.c"

  488. #endif

  489. 

  490. #ifdef CAN_COMPILE_X86_ASM

  491. 

  492. //X86 versions

  493. /*

  494. #undef RENAME

  495. #undef HAVE_MMX

  496. #undef HAVE_MMX2

  497. #undef HAVE_3DNOW

  498. #define ARCH_X86

  499. #define RENAME(a) a ## _X86

  500. #include "swscale_template.c"

  501. */

  502. //MMX versions

  503. #ifdef COMPILE_MMX

  504. #undef RENAME

  505. #define HAVE_MMX

  506. #undef HAVE_MMX2

  507. #undef HAVE_3DNOW

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

  509. #include "swscale_template.c"

  510. #endif

  511. 

  512. //MMX2 versions

  513. #ifdef COMPILE_MMX2

  514. #undef RENAME

  515. #define HAVE_MMX

  516. #define HAVE_MMX2

  517. #undef HAVE_3DNOW

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

  519. #include "swscale_template.c"

  520. #endif

  521. 

  522. //3DNOW versions

  523. #ifdef COMPILE_3DNOW

  524. #undef RENAME

  525. #define HAVE_MMX

  526. #undef HAVE_MMX2

  527. #define HAVE_3DNOW

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

  529. #include "swscale_template.c"

  530. #endif

  531. 

  532. #endif //CAN_COMPILE_X86_ASM

  533. 

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

  535. 

  536. 

  537. // old global scaler, dont use for new code

  538. // will use sws_flags from the command line

  539. void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,

  540. 			     int srcSliceH, uint8_t* dst[], int dstStride, int dstbpp,

  541. 			     int srcW, int srcH, int dstW, int dstH){

  542. 

  543. 	static SwsContext *context=NULL;

  544. 	int dstFormat;

  545. 	int dstStride3[3]= {dstStride, dstStride>>1, dstStride>>1};

  546. 

  547. 	switch(dstbpp)

  548. 	{

  549. 		case 8 : dstFormat= IMGFMT_Y8;		break;

  550. 		case 12: dstFormat= IMGFMT_YV12;	break;

  551. 		case 15: dstFormat= IMGFMT_BGR15;	break;

  552. 		case 16: dstFormat= IMGFMT_BGR16;	break;

  553. 		case 24: dstFormat= IMGFMT_BGR24;	break;

  554. 		case 32: dstFormat= IMGFMT_BGR32;	break;

  555. 		default: return;

  556. 	}

  557. 

  558. 	if(!context) context=getSwsContextFromCmdLine(srcW, srcH, IMGFMT_YV12, dstW, dstH, dstFormat);

  559. 

  560. 	swScale(context, src, srcStride, srcSliceY, srcSliceH, dst, dstStride3);

  561. }

  562. 

  563. // will use sws_flags & src_filter (from cmd line)

  564. SwsContext *getSwsContextFromCmdLine(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat)

  565. {

  566. 	int flags=0;

  567. 	static int firstTime=1;

  568. 

  569. #ifdef ARCH_X86

  570. 	if(gCpuCaps.hasMMX)

  571. 		asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)

  572. #endif

  573. 	if(firstTime)

  574. 	{

  575. 		firstTime=0;

  576. 		flags= SWS_PRINT_INFO;

  577. 	}

  578. 	else if(verbose>1) flags= SWS_PRINT_INFO;

  579. 

  580. 	if(src_filter.lumH) freeVec(src_filter.lumH);

  581. 	if(src_filter.lumV) freeVec(src_filter.lumV);

  582. 	if(src_filter.chrH) freeVec(src_filter.chrH);

  583. 	if(src_filter.chrV) freeVec(src_filter.chrV);

  584. 

  585. 	if(sws_lum_gblur!=0.0){

  586. 		src_filter.lumH= getGaussianVec(sws_lum_gblur, 3.0);

  587. 		src_filter.lumV= getGaussianVec(sws_lum_gblur, 3.0);

  588. 	}else{

  589. 		src_filter.lumH= getIdentityVec();

  590. 		src_filter.lumV= getIdentityVec();

  591. 	}

  592. 

  593. 	if(sws_chr_gblur!=0.0){

  594. 		src_filter.chrH= getGaussianVec(sws_chr_gblur, 3.0);

  595. 		src_filter.chrV= getGaussianVec(sws_chr_gblur, 3.0);

  596. 	}else{

  597. 		src_filter.chrH= getIdentityVec();

  598. 		src_filter.chrV= getIdentityVec();

  599. 	}

  600. 

  601. 	if(sws_chr_sharpen!=0.0){

  602. 		SwsVector *g= getConstVec(-1.0, 3);

  603. 		SwsVector *id= getConstVec(10.0/sws_chr_sharpen, 1);

  604. 		g->coeff[1]=2.0;

  605. 		addVec(id, g);

  606. 		convVec(src_filter.chrH, id);

  607. 		convVec(src_filter.chrV, id);

  608. 		freeVec(g);

  609. 		freeVec(id);

  610. 	}

  611. 

  612. 	if(sws_lum_sharpen!=0.0){

  613. 		SwsVector *g= getConstVec(-1.0, 3);

  614. 		SwsVector *id= getConstVec(10.0/sws_lum_sharpen, 1);

  615. 		g->coeff[1]=2.0;

  616. 		addVec(id, g);

  617. 		convVec(src_filter.lumH, id);

  618. 		convVec(src_filter.lumV, id);

  619. 		freeVec(g);

  620. 		freeVec(id);

  621. 	}

  622. 

  623. 	if(sws_chr_hshift)

  624. 		shiftVec(src_filter.chrH, sws_chr_hshift);

  625. 

  626. 	if(sws_chr_vshift)

  627. 		shiftVec(src_filter.chrV, sws_chr_vshift);

  628. 

  629. 	normalizeVec(src_filter.chrH, 1.0);

  630. 	normalizeVec(src_filter.chrV, 1.0);

  631. 	normalizeVec(src_filter.lumH, 1.0);

  632. 	normalizeVec(src_filter.lumV, 1.0);

  633. 

  634. 	if(verbose > 1) printVec(src_filter.chrH);

  635. 	if(verbose > 1) printVec(src_filter.lumH);

  636. 

  637. 	switch(sws_flags)

  638. 	{

  639. 		case 0: flags|= SWS_FAST_BILINEAR; break;

  640. 		case 1: flags|= SWS_BILINEAR; break;

  641. 		case 2: flags|= SWS_BICUBIC; break;

  642. 		case 3: flags|= SWS_X; break;

  643. 		case 4: flags|= SWS_POINT; break;

  644. 		case 5: flags|= SWS_AREA; break;

  645. 		default:flags|= SWS_BILINEAR; break;

  646. 	}

  647. 

  648. 	return getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, &src_filter, NULL);

  649. }

  650. 

  651. 

  652. static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,

  653. 			      int srcW, int dstW, int filterAlign, int one, int flags,

  654. 			      SwsVector *srcFilter, SwsVector *dstFilter)

  655. {

  656. 	int i;

  657. 	int filterSize;

  658. 	int filter2Size;

  659. 	int minFilterSize;

  660. 	double *filter=NULL;

  661. 	double *filter2=NULL;

  662. #ifdef ARCH_X86

  663. 	if(gCpuCaps.hasMMX)

  664. 		asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)

  665. #endif

  666. 

  667. 	*filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));

  668. 	(*filterPos)[dstW]=0; // the MMX scaler will read over the end 

  669. 

  670. 	if(ABS(xInc - 0x10000) <10) // unscaled

  671. 	{

  672. 		int i;

  673. 		filterSize= 1;

  674. 		filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);

  675. 		for(i=0; i<dstW*filterSize; i++) filter[i]=0;

  676. 

  677. 		for(i=0; i<dstW; i++)

  678. 		{

  679. 			filter[i*filterSize]=1;

  680. 			(*filterPos)[i]=i;

  681. 		}

  682. 

  683. 	}

  684. 	else if(flags&SWS_POINT) // lame looking point sampling mode

  685. 	{

  686. 		int i;

  687. 		int xDstInSrc;

  688. 		filterSize= 1;

  689. 		filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);

  690. 		

  691. 		xDstInSrc= xInc/2 - 0x8000;

  692. 		for(i=0; i<dstW; i++)

  693. 		{

  694. 			int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;

  695. 

  696. 			(*filterPos)[i]= xx;

  697. 			filter[i]= 1.0;

  698. 			xDstInSrc+= xInc;

  699. 		}

  700. 	}

  701. 	else if(xInc <= (1<<16) || (flags&SWS_FAST_BILINEAR)) // upscale

  702. 	{

  703. 		int i;

  704. 		int xDstInSrc;

  705. 		if     (flags&SWS_BICUBIC) filterSize= 4;

  706. 		else if(flags&SWS_X      ) filterSize= 4;

  707. 		else			   filterSize= 2; // SWS_BILINEAR / SWS_AREA 

  708. //		printf("%d %d %d\n", filterSize, srcW, dstW);

  709. 		filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);

  710. 

  711. 		xDstInSrc= xInc/2 - 0x8000;

  712. 		for(i=0; i<dstW; i++)

  713. 		{

  714. 			int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;

  715. 			int j;

  716. 

  717. 			(*filterPos)[i]= xx;

  718. 			if((flags & SWS_BICUBIC) || (flags & SWS_X))

  719. 			{

  720. 				double d= ABS(((xx+1)<<16) - xDstInSrc)/(double)(1<<16);

  721. 				double y1,y2,y3,y4;

  722. 				double A= -0.6;

  723. 				if(flags & SWS_BICUBIC){

  724. 						// Equation is from VirtualDub

  725. 					y1 = (        +     A*d -       2.0*A*d*d +       A*d*d*d);

  726. 					y2 = (+ 1.0             -     (A+3.0)*d*d + (A+2.0)*d*d*d);

  727. 					y3 = (        -     A*d + (2.0*A+3.0)*d*d - (A+2.0)*d*d*d);

  728. 					y4 = (                  +           A*d*d -       A*d*d*d);

  729. 				}else{

  730. 						// cubic interpolation (derived it myself)

  731. 					y1 = (    -2.0*d + 3.0*d*d - 1.0*d*d*d)/6.0;

  732. 					y2 = (6.0 -3.0*d - 6.0*d*d + 3.0*d*d*d)/6.0;

  733. 					y3 = (    +6.0*d + 3.0*d*d - 3.0*d*d*d)/6.0;

  734. 					y4 = (    -1.0*d           + 1.0*d*d*d)/6.0;

  735. 				}

  736. 

  737. //				printf("%d %d %d \n", coeff, (int)d, xDstInSrc);

  738. 				filter[i*filterSize + 0]= y1;

  739. 				filter[i*filterSize + 1]= y2;

  740. 				filter[i*filterSize + 2]= y3;

  741. 				filter[i*filterSize + 3]= y4;

  742. //				printf("%1.3f %1.3f %1.3f %1.3f %1.3f\n",d , y1, y2, y3, y4);

  743. 			}

  744. 			else

  745. 			{

  746. 				//Bilinear upscale / linear interpolate / Area averaging

  747. 				for(j=0; j<filterSize; j++)

  748. 				{

  749. 					double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);

  750. 					double coeff= 1.0 - d;

  751. 					if(coeff<0) coeff=0;

  752. 	//				printf("%d %d %d \n", coeff, (int)d, xDstInSrc);

  753. 					filter[i*filterSize + j]= coeff;

  754. 					xx++;

  755. 				}

  756. 			}

  757. 			xDstInSrc+= xInc;

  758. 		}

  759. 	}

  760. 	else // downscale

  761. 	{

  762. 		int xDstInSrc;

  763. 		if(flags&SWS_BICUBIC)	filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);

  764. 		else if(flags&SWS_X)	filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);

  765. 		else if(flags&SWS_AREA)	filterSize= (int)ceil(1 + 1.0*srcW / (double)dstW);

  766. 		else /* BILINEAR */	filterSize= (int)ceil(1 + 2.0*srcW / (double)dstW);

  767. //		printf("%d %d %d\n", *filterSize, srcW, dstW);

  768. 		filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);

  769. 

  770. 		xDstInSrc= xInc/2 - 0x8000;

  771. 		for(i=0; i<dstW; i++)

  772. 		{

  773. 			int xx= (int)((double)xDstInSrc/(double)(1<<16) - (filterSize-1)*0.5 + 0.5);

  774. 			int j;

  775. 			(*filterPos)[i]= xx;

  776. 			for(j=0; j<filterSize; j++)

  777. 			{

  778. 				double d= ABS((xx<<16) - xDstInSrc)/(double)xInc;

  779. 				double coeff;

  780. 				if((flags & SWS_BICUBIC) || (flags & SWS_X))

  781. 				{

  782. 					double A= -0.75;

  783. //					d*=2;

  784. 					// Equation is from VirtualDub

  785. 					if(d<1.0)

  786. 						coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);

  787. 					else if(d<2.0)

  788. 						coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);

  789. 					else

  790. 						coeff=0.0;

  791. 				}

  792. 				else if(flags & SWS_AREA)

  793. 				{

  794. 					double srcPixelSize= (1<<16)/(double)xInc;

  795. 					if(d + srcPixelSize/2 < 0.5) coeff= 1.0;

  796. 					else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;

  797. 					else coeff=0.0;

  798. 				}

  799. 				else

  800. 				{

  801. 					coeff= 1.0 - d;

  802. 					if(coeff<0) coeff=0;

  803. 				}

  804. //				printf("%1.3f %2.3f %d \n", coeff, d, xDstInSrc);

  805. 				filter[i*filterSize + j]= coeff;

  806. 				xx++;

  807. 			}

  808. 			xDstInSrc+= xInc;

  809. 		}

  810. 	}

  811. 

  812. 	/* apply src & dst Filter to filter -> filter2

  813. 	   free(filter);

  814. 	*/

  815. 	filter2Size= filterSize;

  816. 	if(srcFilter) filter2Size+= srcFilter->length - 1;

  817. 	if(dstFilter) filter2Size+= dstFilter->length - 1;

  818. 	filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));

  819. 

  820. 	for(i=0; i<dstW; i++)

  821. 	{

  822. 		int j;

  823. 		SwsVector scaleFilter;

  824. 		SwsVector *outVec;

  825. 

  826. 		scaleFilter.coeff= filter + i*filterSize;

  827. 		scaleFilter.length= filterSize;

  828. 

  829. 		if(srcFilter) outVec= getConvVec(srcFilter, &scaleFilter);

  830. 		else	      outVec= &scaleFilter;

  831. 

  832. 		ASSERT(outVec->length == filter2Size)

  833. 		//FIXME dstFilter

  834. 

  835. 		for(j=0; j<outVec->length; j++)

  836. 		{

  837. 			filter2[i*filter2Size + j]= outVec->coeff[j];

  838. 		}

  839. 

  840. 		(*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;

  841. 

  842. 		if(outVec != &scaleFilter) freeVec(outVec);

  843. 	}

  844. 	free(filter); filter=NULL;

  845. 

  846. 	/* try to reduce the filter-size (step1 find size and shift left) */

  847. 	// Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)

  848. 	minFilterSize= 0;

  849. 	for(i=dstW-1; i>=0; i--)

  850. 	{

  851. 		int min= filter2Size;

  852. 		int j;

  853. 		double cutOff=0.0;

  854. 

  855. 		/* get rid off near zero elements on the left by shifting left */

  856. 		for(j=0; j<filter2Size; j++)

  857. 		{

  858. 			int k;

  859. 			cutOff += ABS(filter2[i*filter2Size]);

  860. 

  861. 			if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;

  862. 

  863. 			/* preserve Monotonicity because the core cant handle the filter otherwise */

  864. 			if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;

  865. 

  866. 			// Move filter coeffs left

  867. 			for(k=1; k<filter2Size; k++)

  868. 				filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];

  869. 			filter2[i*filter2Size + k - 1]= 0.0;

  870. 			(*filterPos)[i]++;

  871. 		}

  872. 

  873. 		cutOff=0.0;

  874. 		/* count near zeros on the right */

  875. 		for(j=filter2Size-1; j>0; j--)

  876. 		{

  877. 			cutOff += ABS(filter2[i*filter2Size + j]);

  878. 

  879. 			if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;

  880. 			min--;

  881. 		}

  882. 

  883. 		if(min>minFilterSize) minFilterSize= min;

  884. 	}

  885. 

  886. 	filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));

  887. 	filter= (double*)memalign(8, filterSize*dstW*sizeof(double));

  888. 	*outFilterSize= filterSize;

  889. 

  890. 	if((flags&SWS_PRINT_INFO) && verbose)

  891. 		printf("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);

  892. 	/* try to reduce the filter-size (step2 reduce it) */

  893. 	for(i=0; i<dstW; i++)

  894. 	{

  895. 		int j;

  896. 

  897. 		for(j=0; j<filterSize; j++)

  898. 		{

  899. 			if(j>=filter2Size) filter[i*filterSize + j]= 0.0;

  900. 			else		   filter[i*filterSize + j]= filter2[i*filter2Size + j];

  901. 		}

  902. 	}

  903. 	free(filter2); filter2=NULL;

  904. 	

  905. 	ASSERT(filterSize > 0)

  906. 

  907. 	//FIXME try to align filterpos if possible

  908. 

  909. 	//fix borders

  910. 	for(i=0; i<dstW; i++)

  911. 	{

  912. 		int j;

  913. 		if((*filterPos)[i] < 0)

  914. 		{

  915. 			// Move filter coeffs left to compensate for filterPos

  916. 			for(j=1; j<filterSize; j++)

  917. 			{

  918. 				int left= MAX(j + (*filterPos)[i], 0);

  919. 				filter[i*filterSize + left] += filter[i*filterSize + j];

  920. 				filter[i*filterSize + j]=0;

  921. 			}

  922. 			(*filterPos)[i]= 0;

  923. 		}

  924. 

  925. 		if((*filterPos)[i] + filterSize > srcW)

  926. 		{

  927. 			int shift= (*filterPos)[i] + filterSize - srcW;

  928. 			// Move filter coeffs right to compensate for filterPos

  929. 			for(j=filterSize-2; j>=0; j--)

  930. 			{

  931. 				int right= MIN(j + shift, filterSize-1);

  932. 				filter[i*filterSize +right] += filter[i*filterSize +j];

  933. 				filter[i*filterSize +j]=0;

  934. 			}

  935. 			(*filterPos)[i]= srcW - filterSize;

  936. 		}

  937. 	}

  938. 

  939. 	// Note the +1 is for the MMXscaler which reads over the end

  940. 	*outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));

  941. 	memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));

  942. 

  943. 	/* Normalize & Store in outFilter */

  944. 	for(i=0; i<dstW; i++)

  945. 	{

  946. 		int j;

  947. 		double sum=0;

  948. 		double scale= one;

  949. 		for(j=0; j<filterSize; j++)

  950. 		{

  951. 			sum+= filter[i*filterSize + j];

  952. 		}

  953. 		scale/= sum;

  954. 		for(j=0; j<filterSize; j++)

  955. 		{

  956. 			(*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);

  957. 		}

  958. 	}

  959. 

  960. 	free(filter);

  961. }

  962. 

  963. #ifdef ARCH_X86

  964. static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode)

  965. {

  966. 	uint8_t *fragment;

  967. 	int imm8OfPShufW1;

  968. 	int imm8OfPShufW2;

  969. 	int fragmentLength;

  970. 

  971. 	int xpos, i;

  972. 

  973. 	// create an optimized horizontal scaling routine

  974. 

  975. 	//code fragment

  976. 

  977. 	asm volatile(

  978. 		"jmp 9f				\n\t"

  979. 	// Begin

  980. 		"0:				\n\t"

  981. 		"movq (%%esi), %%mm0		\n\t" //FIXME Alignment

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

  983. 		"psrlq $8, %%mm0		\n\t"

  984. 		"punpcklbw %%mm7, %%mm1	\n\t"

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

  986. 		"punpcklbw %%mm7, %%mm0	\n\t"

  987. 		"addw %%bx, %%cx		\n\t" //2*xalpha += (4*lumXInc)&0xFFFF

  988. 		"pshufw $0xFF, %%mm1, %%mm1	\n\t"

  989. 		"1:				\n\t"

  990. 		"adcl %%edx, %%esi		\n\t" //xx+= (4*lumXInc)>>16 + carry

  991. 		"pshufw $0xFF, %%mm0, %%mm0	\n\t"

  992. 		"2:				\n\t"

  993. 		"psrlw $9, %%mm3		\n\t"

  994. 		"psubw %%mm1, %%mm0		\n\t"

  995. 		"pmullw %%mm3, %%mm0		\n\t"

  996. 		"paddw %%mm6, %%mm2		\n\t" // 2*alpha += xpos&0xFFFF

  997. 		"psllw $7, %%mm1		\n\t"

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

  999. 

  1000. 		"movq %%mm0, (%%edi, %%eax)	\n\t"

  1001. 

  1002. 		"addl $8, %%eax			\n\t"

  1003. 	// End

  1004. 		"9:				\n\t"

  1005. //		"int $3\n\t"

  1006. 		"leal 0b, %0			\n\t"

  1007. 		"leal 1b, %1			\n\t"

  1008. 		"leal 2b, %2			\n\t"

  1009. 		"decl %1			\n\t"

  1010. 		"decl %2			\n\t"

  1011. 		"subl %0, %1			\n\t"

  1012. 		"subl %0, %2			\n\t"

  1013. 		"leal 9b, %3			\n\t"

  1014. 		"subl %0, %3			\n\t"

  1015. 		:"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),

  1016. 		"=r" (fragmentLength)

  1017. 	);

  1018. 

  1019. 	xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers

  1020. 

  1021. 	for(i=0; i<dstW/8; i++)

  1022. 	{

  1023. 		int xx=xpos>>16;

  1024. 

  1025. 		if((i&3) == 0)

  1026. 		{

  1027. 			int a=0;

  1028. 			int b=((xpos+xInc)>>16) - xx;

  1029. 			int c=((xpos+xInc*2)>>16) - xx;

  1030. 			int d=((xpos+xInc*3)>>16) - xx;

  1031. 

  1032. 			memcpy(funnyCode + fragmentLength*i/4, fragment, fragmentLength);

  1033. 

  1034. 			funnyCode[fragmentLength*i/4 + imm8OfPShufW1]=

  1035. 			funnyCode[fragmentLength*i/4 + imm8OfPShufW2]=

  1036. 				a | (b<<2) | (c<<4) | (d<<6);

  1037. 

  1038. 			// if we dont need to read 8 bytes than dont :), reduces the chance of

  1039. 			// crossing a cache line

  1040. 			if(d<3) funnyCode[fragmentLength*i/4 + 1]= 0x6E;

  1041. 

  1042. 			funnyCode[fragmentLength*(i+4)/4]= RET;

  1043. 		}

  1044. 		xpos+=xInc;

  1045. 	}

  1046. }

  1047. #endif // ARCH_X86

  1048. 

  1049. //FIXME remove

  1050. void SwScale_Init(){

  1051. }

  1052. 

  1053. static void globalInit(){

  1054.     // generating tables:

  1055.     int i;

  1056.     for(i=0; i<768; i++){

  1057. 	int c= MIN(MAX(i-256, 0), 255);

  1058. 	clip_table[i]=c;

  1059. 	yuvtab_2568[c]= clip_yuvtab_2568[i]=(0x2568*(c-16))+(256<<13);

  1060. 	yuvtab_3343[c]= clip_yuvtab_3343[i]=0x3343*(c-128);

  1061. 	yuvtab_0c92[c]= clip_yuvtab_0c92[i]=-0x0c92*(c-128);

  1062. 	yuvtab_1a1e[c]= clip_yuvtab_1a1e[i]=-0x1a1e*(c-128);

  1063. 	yuvtab_40cf[c]= clip_yuvtab_40cf[i]=0x40cf*(c-128);

  1064.     }

  1065. 

  1066.     for(i=0; i<768; i++)

  1067.     {

  1068. 	int v= clip_table[i];

  1069. 	clip_table16b[i]= v>>3;

  1070. 	clip_table16g[i]= (v<<3)&0x07E0;

  1071. 	clip_table16r[i]= (v<<8)&0xF800;

  1072. 	clip_table15b[i]= v>>3;

  1073. 	clip_table15g[i]= (v<<2)&0x03E0;

  1074. 	clip_table15r[i]= (v<<7)&0x7C00;

  1075.     }

  1076. 

  1077. cpuCaps= gCpuCaps;

  1078. 

  1079. #ifdef RUNTIME_CPUDETECT

  1080. #ifdef CAN_COMPILE_X86_ASM

  1081. 	// ordered per speed fasterst first

  1082. 	if(gCpuCaps.hasMMX2)

  1083. 		swScale= swScale_MMX2;

  1084. 	else if(gCpuCaps.has3DNow)

  1085. 		swScale= swScale_3DNow;

  1086. 	else if(gCpuCaps.hasMMX)

  1087. 		swScale= swScale_MMX;

  1088. 	else

  1089. 		swScale= swScale_C;

  1090. 

  1091. #else

  1092. 	swScale= swScale_C;

  1093. 	cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;

  1094. #endif

  1095. #else //RUNTIME_CPUDETECT

  1096. #ifdef HAVE_MMX2

  1097. 	swScale= swScale_MMX2;

  1098. 	cpuCaps.has3DNow = 0;

  1099. #elif defined (HAVE_3DNOW)

  1100. 	swScale= swScale_3DNow;

  1101. 	cpuCaps.hasMMX2 = 0;

  1102. #elif defined (HAVE_MMX)

  1103. 	swScale= swScale_MMX;

  1104. 	cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;

  1105. #else

  1106. 	swScale= swScale_C;

  1107. 	cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;

  1108. #endif

  1109. #endif //!RUNTIME_CPUDETECT

  1110. }

  1111. 

  1112. /* Warper functions for yuv2bgr */

  1113. static void planarYuvToBgr(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1114.              int srcSliceH, uint8_t* dst[], int dstStride[]){

  1115. 

  1116. 	if(c->srcFormat==IMGFMT_YV12)

  1117. 		yuv2rgb( dst[0],src[0],src[1],src[2],c->srcW,c->srcH,dstStride[0],srcStride[0],srcStride[1] );

  1118. 	else /* I420 & IYUV */

  1119. 		yuv2rgb( dst[0],src[0],src[2],src[1],c->srcW,c->srcH,dstStride[0],srcStride[0],srcStride[1] );

  1120. }

  1121. 

  1122. SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,

  1123.                          SwsFilter *srcFilter, SwsFilter *dstFilter){

  1124. 

  1125. 	SwsContext *c;

  1126. 	int i;

  1127. 	int usesFilter;

  1128. 	SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};

  1129. 

  1130. #ifdef ARCH_X86

  1131. 	if(gCpuCaps.hasMMX)

  1132. 		asm volatile("emms\n\t"::: "memory");

  1133. #endif

  1134. 

  1135. 	if(swScale==NULL) globalInit();

  1136. 

  1137. 	/* avoid dupplicate Formats, so we dont need to check to much */

  1138. 	if(srcFormat==IMGFMT_IYUV) srcFormat=IMGFMT_I420;

  1139. 	if(srcFormat==IMGFMT_Y8)   srcFormat=IMGFMT_Y800;

  1140. 	if(dstFormat==IMGFMT_Y8)   dstFormat=IMGFMT_Y800;

  1141. 

  1142. 	if(!isSupportedIn(srcFormat)) 

  1143. 	{

  1144. 		fprintf(stderr, "swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));

  1145. 		return NULL;

  1146. 	}

  1147. 	if(!isSupportedOut(dstFormat))

  1148. 	{

  1149. 		fprintf(stderr, "swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));

  1150. 		return NULL;

  1151. 	}

  1152. 

  1153. 	/* sanity check */

  1154. 	if(srcW<4 || srcH<1 || dstW<8 || dstH<1) //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code

  1155. 	{

  1156. 		fprintf(stderr, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 

  1157. 			srcW, srcH, dstW, dstH);

  1158. 		return NULL;

  1159. 	}

  1160. 

  1161. 	if(!dstFilter) dstFilter= &dummyFilter;

  1162. 	if(!srcFilter) srcFilter= &dummyFilter;

  1163. 

  1164. 	c= memalign(64, sizeof(SwsContext));

  1165. 	memset(c, 0, sizeof(SwsContext));

  1166. 

  1167. 	c->srcW= srcW;

  1168. 	c->srcH= srcH;

  1169. 	c->dstW= dstW;

  1170. 	c->dstH= dstH;

  1171. 	c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;

  1172. 	c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;

  1173. 	c->flags= flags;

  1174. 	c->dstFormat= dstFormat;

  1175. 	c->srcFormat= srcFormat;

  1176. 

  1177. 	usesFilter=0;

  1178. 	if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;

  1179. 	if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;

  1180. 	if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;

  1181. 	if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;

  1182. 	if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;

  1183. 	if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;

  1184. 	if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;

  1185. 	if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;

  1186. 	

  1187. 	/* special Cases */

  1188. 	if(srcW==dstW && srcH==dstH && !usesFilter)

  1189. 	{

  1190. 		/* yuv2bgr */

  1191. 		if(isPlanarYUV(srcFormat) && isBGR(dstFormat))

  1192. 		{

  1193. 			// FIXME multiple yuv2rgb converters wont work that way cuz that thing is full of globals&statics

  1194. 			yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_BGR);

  1195. 			c->swScale= planarYuvToBgr;

  1196. 			

  1197. 			if(flags&SWS_PRINT_INFO)

  1198. 				printf("SwScaler: using unscaled %s -> %s special converter\n", 

  1199. 					vo_format_name(srcFormat), vo_format_name(dstFormat));

  1200. 			return c;

  1201. 		}

  1202. 	}

  1203. 

  1204. 	if(cpuCaps.hasMMX2)

  1205. 	{

  1206. 		c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;

  1207. 		if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))

  1208. 		{

  1209. 			if(flags&SWS_PRINT_INFO)

  1210. 				fprintf(stderr, "SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");

  1211. 		}

  1212. 	}

  1213. 	else

  1214. 		c->canMMX2BeUsed=0;

  1215. 

  1216. 

  1217. 	/* dont use full vertical UV input/internaly if the source doesnt even have it */

  1218. 	if(isHalfChrV(srcFormat)) c->flags= flags= flags&(~SWS_FULL_CHR_V);

  1219. 	/* dont use full horizontal UV input if the source doesnt even have it */

  1220. 	if(isHalfChrH(srcFormat)) c->flags= flags= flags&(~SWS_FULL_CHR_H_INP);

  1221. 	/* dont use full horizontal UV internally if the destination doesnt even have it */

  1222. 	if(isHalfChrH(dstFormat)) c->flags= flags= flags&(~SWS_FULL_CHR_H_INT);

  1223. 

  1224. 	if(flags&SWS_FULL_CHR_H_INP)	c->chrSrcW= srcW;

  1225. 	else				c->chrSrcW= (srcW+1)>>1;

  1226. 

  1227. 	if(flags&SWS_FULL_CHR_H_INT)	c->chrDstW= dstW;

  1228. 	else				c->chrDstW= (dstW+1)>>1;

  1229. 

  1230. 	if(flags&SWS_FULL_CHR_V)	c->chrSrcH= srcH;

  1231. 	else				c->chrSrcH= (srcH+1)>>1;

  1232. 

  1233. 	if(isHalfChrV(dstFormat))	c->chrDstH= (dstH+1)>>1;

  1234. 	else				c->chrDstH= dstH;

  1235. 

  1236. 	c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;

  1237. 	c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;

  1238. 

  1239. 

  1240. 	// match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst

  1241. 	// but only for the FAST_BILINEAR mode otherwise do correct scaling

  1242. 	// n-2 is the last chrominance sample available

  1243. 	// this is not perfect, but noone shuld notice the difference, the more correct variant

  1244. 	// would be like the vertical one, but that would require some special code for the

  1245. 	// first and last pixel

  1246. 	if(flags&SWS_FAST_BILINEAR)

  1247. 	{

  1248. 		if(c->canMMX2BeUsed)

  1249. 		{

  1250. 			c->lumXInc+= 20;

  1251. 			c->chrXInc+= 20;

  1252. 		}

  1253. 		//we dont use the x86asm scaler if mmx is available

  1254. 		else if(cpuCaps.hasMMX)

  1255. 		{

  1256. 			c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;

  1257. 			c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;

  1258. 		}

  1259. 	}

  1260. 

  1261. 	/* precalculate horizontal scaler filter coefficients */

  1262. 	{

  1263. 		const int filterAlign= cpuCaps.hasMMX ? 4 : 1;

  1264. 

  1265. 		initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,

  1266. 				 srcW      ,       dstW, filterAlign, 1<<14, flags,

  1267. 				 srcFilter->lumH, dstFilter->lumH);

  1268. 		initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,

  1269. 				(srcW+1)>>1, c->chrDstW, filterAlign, 1<<14, flags,

  1270. 				 srcFilter->chrH, dstFilter->chrH);

  1271. 

  1272. #ifdef ARCH_X86

  1273. // cant downscale !!!

  1274. 		if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))

  1275. 		{

  1276. 			initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode);

  1277. 			initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode);

  1278. 		}

  1279. #endif

  1280. 	} // Init Horizontal stuff

  1281. 

  1282. 

  1283. 

  1284. 	/* precalculate vertical scaler filter coefficients */

  1285. 	initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,

  1286. 			srcH      ,        dstH, 1, (1<<12)-4, flags,

  1287. 			srcFilter->lumV, dstFilter->lumV);

  1288. 	initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,

  1289. 			(srcH+1)>>1, c->chrDstH, 1, (1<<12)-4, flags,

  1290. 			 srcFilter->chrV, dstFilter->chrV);

  1291. 

  1292. 	// Calculate Buffer Sizes so that they wont run out while handling these damn slices

  1293. 	c->vLumBufSize= c->vLumFilterSize;

  1294. 	c->vChrBufSize= c->vChrFilterSize;

  1295. 	for(i=0; i<dstH; i++)

  1296. 	{

  1297. 		int chrI= i*c->chrDstH / dstH;

  1298. 		int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1299. 				 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<1));

  1300. 		nextSlice&= ~1; // Slices start at even boundaries

  1301. 		if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1302. 			c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];

  1303. 		if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>1))

  1304. 			c->vChrBufSize= (nextSlice>>1) - c->vChrFilterPos[chrI];

  1305. 	}

  1306. 

  1307. 	// allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1308. 	c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));

  1309. 	c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));

  1310. 	//Note we need at least one pixel more at the end because of the mmx code (just in case someone wanna replace the 4000/8000)

  1311. 	for(i=0; i<c->vLumBufSize; i++)

  1312. 		c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);

  1313. 	for(i=0; i<c->vChrBufSize; i++)

  1314. 		c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);

  1315. 

  1316. 	//try to avoid drawing green stuff between the right end and the stride end

  1317. 	for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);

  1318. 	for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);

  1319. 

  1320. 	ASSERT(c->chrDstH <= dstH)

  1321. 

  1322. 	// pack filter data for mmx code

  1323. 	if(cpuCaps.hasMMX)

  1324. 	{

  1325. 		c->lumMmxFilter= (int16_t*)memalign(8, c->vLumFilterSize*      dstH*4*sizeof(int16_t));

  1326. 		c->chrMmxFilter= (int16_t*)memalign(8, c->vChrFilterSize*c->chrDstH*4*sizeof(int16_t));

  1327. 		for(i=0; i<c->vLumFilterSize*dstH; i++)

  1328. 			c->lumMmxFilter[4*i]=c->lumMmxFilter[4*i+1]=c->lumMmxFilter[4*i+2]=c->lumMmxFilter[4*i+3]=

  1329. 				c->vLumFilter[i];

  1330. 		for(i=0; i<c->vChrFilterSize*c->chrDstH; i++)

  1331. 			c->chrMmxFilter[4*i]=c->chrMmxFilter[4*i+1]=c->chrMmxFilter[4*i+2]=c->chrMmxFilter[4*i+3]=

  1332. 				c->vChrFilter[i];

  1333. 	}

  1334. 

  1335. 	if(flags&SWS_PRINT_INFO)

  1336. 	{

  1337. #ifdef DITHER1XBPP

  1338. 		char *dither= " dithered";

  1339. #else

  1340. 		char *dither= "";

  1341. #endif

  1342. 		if(flags&SWS_FAST_BILINEAR)

  1343. 			fprintf(stderr, "\nSwScaler: FAST_BILINEAR scaler, ");

  1344. 		else if(flags&SWS_BILINEAR)

  1345. 			fprintf(stderr, "\nSwScaler: BILINEAR scaler, ");

  1346. 		else if(flags&SWS_BICUBIC)

  1347. 			fprintf(stderr, "\nSwScaler: BICUBIC scaler, ");

  1348. 		else if(flags&SWS_X)

  1349. 			fprintf(stderr, "\nSwScaler: Experimental scaler, ");

  1350. 		else if(flags&SWS_POINT)

  1351. 			fprintf(stderr, "\nSwScaler: Nearest Neighbor / POINT scaler, ");

  1352. 		else if(flags&SWS_AREA)

  1353. 			fprintf(stderr, "\nSwScaler: Area Averageing scaler, ");

  1354. 		else

  1355. 			fprintf(stderr, "\nSwScaler: ehh flags invalid?! ");

  1356. 

  1357. 		if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)

  1358. 			fprintf(stderr, "from %s to%s %s ", 

  1359. 				vo_format_name(srcFormat), dither, vo_format_name(dstFormat));

  1360. 		else

  1361. 			fprintf(stderr, "from %s to %s ", 

  1362. 				vo_format_name(srcFormat), vo_format_name(dstFormat));

  1363. 

  1364. 		if(cpuCaps.hasMMX2)

  1365. 			fprintf(stderr, "using MMX2\n");

  1366. 		else if(cpuCaps.has3DNow)

  1367. 			fprintf(stderr, "using 3DNOW\n");

  1368. 		else if(cpuCaps.hasMMX)

  1369. 			fprintf(stderr, "using MMX\n");

  1370. 		else

  1371. 			fprintf(stderr, "using C\n");

  1372. 	}

  1373. 

  1374. 	if((flags & SWS_PRINT_INFO) && verbose)

  1375. 	{

  1376. 		if(cpuCaps.hasMMX)

  1377. 		{

  1378. 			if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1379. 				printf("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1380. 			else

  1381. 			{

  1382. 				if(c->hLumFilterSize==4)

  1383. 					printf("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");

  1384. 				else if(c->hLumFilterSize==8)

  1385. 					printf("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");

  1386. 				else

  1387. 					printf("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");

  1388. 

  1389. 				if(c->hChrFilterSize==4)

  1390. 					printf("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");

  1391. 				else if(c->hChrFilterSize==8)

  1392. 					printf("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");

  1393. 				else

  1394. 					printf("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");

  1395. 			}

  1396. 		}

  1397. 		else

  1398. 		{

  1399. #ifdef ARCH_X86

  1400. 			printf("SwScaler: using X86-Asm scaler for horizontal scaling\n");

  1401. #else

  1402. 			if(flags & SWS_FAST_BILINEAR)

  1403. 				printf("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");

  1404. 			else

  1405. 				printf("SwScaler: using C scaler for horizontal scaling\n");

  1406. #endif

  1407. 		}

  1408. 		if(isPlanarYUV(dstFormat))

  1409. 		{

  1410. 			if(c->vLumFilterSize==1)

  1411. 				printf("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");

  1412. 			else

  1413. 				printf("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");

  1414. 		}

  1415. 		else

  1416. 		{

  1417. 			if(c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1418. 				printf("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1419. 				       "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");

  1420. 			else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)

  1421. 				printf("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");

  1422. 			else

  1423. 				printf("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");

  1424. 		}

  1425. 

  1426. 		if(dstFormat==IMGFMT_BGR24)

  1427. 			printf("SwScaler: using %s YV12->BGR24 Converter\n",

  1428. 				cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));

  1429. 		else if(dstFormat==IMGFMT_BGR32)

  1430. 			printf("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");

  1431. 		else if(dstFormat==IMGFMT_BGR16)

  1432. 			printf("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");

  1433. 		else if(dstFormat==IMGFMT_BGR15)

  1434. 			printf("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");

  1435. 

  1436. 		printf("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1437. 	}

  1438. 	if((flags & SWS_PRINT_INFO) && verbose>1)

  1439. 	{

  1440. 		printf("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1441. 			c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1442. 		printf("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1443. 			c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1444. 	}

  1445. 

  1446. 	c->swScale= swScale;

  1447. 	return c;

  1448. }

  1449. 

  1450. /**

  1451.  * returns a normalized gaussian curve used to filter stuff

  1452.  * quality=3 is high quality, lowwer is lowwer quality

  1453.  */

  1454. 

  1455. SwsVector *getGaussianVec(double variance, double quality){

  1456. 	const int length= (int)(variance*quality + 0.5) | 1;

  1457. 	int i;

  1458. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1459. 	double middle= (length-1)*0.5;

  1460. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1461. 

  1462. 	vec->coeff= coeff;

  1463. 	vec->length= length;

  1464. 

  1465. 	for(i=0; i<length; i++)

  1466. 	{

  1467. 		double dist= i-middle;

  1468. 		coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);

  1469. 	}

  1470. 

  1471. 	normalizeVec(vec, 1.0);

  1472. 

  1473. 	return vec;

  1474. }

  1475. 

  1476. SwsVector *getConstVec(double c, int length){

  1477. 	int i;

  1478. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1479. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1480. 

  1481. 	vec->coeff= coeff;

  1482. 	vec->length= length;

  1483. 

  1484. 	for(i=0; i<length; i++)

  1485. 		coeff[i]= c;

  1486. 

  1487. 	return vec;

  1488. }

  1489. 

  1490. 

  1491. SwsVector *getIdentityVec(void){

  1492. 	double *coeff= memalign(sizeof(double), sizeof(double));

  1493. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1494. 	coeff[0]= 1.0;

  1495. 

  1496. 	vec->coeff= coeff;

  1497. 	vec->length= 1;

  1498. 

  1499. 	return vec;

  1500. }

  1501. 

  1502. void normalizeVec(SwsVector *a, double height){

  1503. 	int i;

  1504. 	double sum=0;

  1505. 	double inv;

  1506. 

  1507. 	for(i=0; i<a->length; i++)

  1508. 		sum+= a->coeff[i];

  1509. 

  1510. 	inv= height/sum;

  1511. 

  1512. 	for(i=0; i<a->length; i++)

  1513. 		a->coeff[i]*= height;

  1514. }

  1515. 

  1516. void scaleVec(SwsVector *a, double scalar){

  1517. 	int i;

  1518. 

  1519. 	for(i=0; i<a->length; i++)

  1520. 		a->coeff[i]*= scalar;

  1521. }

  1522. 

  1523. static SwsVector *getConvVec(SwsVector *a, SwsVector *b){

  1524. 	int length= a->length + b->length - 1;

  1525. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1526. 	int i, j;

  1527. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1528. 

  1529. 	vec->coeff= coeff;

  1530. 	vec->length= length;

  1531. 

  1532. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  1533. 

  1534. 	for(i=0; i<a->length; i++)

  1535. 	{

  1536. 		for(j=0; j<b->length; j++)

  1537. 		{

  1538. 			coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1539. 		}

  1540. 	}

  1541. 

  1542. 	return vec;

  1543. }

  1544. 

  1545. static SwsVector *sumVec(SwsVector *a, SwsVector *b){

  1546. 	int length= MAX(a->length, b->length);

  1547. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1548. 	int i;

  1549. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1550. 

  1551. 	vec->coeff= coeff;

  1552. 	vec->length= length;

  1553. 

  1554. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  1555. 

  1556. 	for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1557. 	for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1558. 

  1559. 	return vec;

  1560. }

  1561. 

  1562. static SwsVector *diffVec(SwsVector *a, SwsVector *b){

  1563. 	int length= MAX(a->length, b->length);

  1564. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1565. 	int i;

  1566. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1567. 

  1568. 	vec->coeff= coeff;

  1569. 	vec->length= length;

  1570. 

  1571. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  1572. 

  1573. 	for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1574. 	for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1575. 

  1576. 	return vec;

  1577. }

  1578. 

  1579. /* shift left / or right if "shift" is negative */

  1580. static SwsVector *getShiftedVec(SwsVector *a, int shift){

  1581. 	int length= a->length + ABS(shift)*2;

  1582. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1583. 	int i;

  1584. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1585. 

  1586. 	vec->coeff= coeff;

  1587. 	vec->length= length;

  1588. 

  1589. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  1590. 

  1591. 	for(i=0; i<a->length; i++)

  1592. 	{

  1593. 		coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1594. 	}

  1595. 

  1596. 	return vec;

  1597. }

  1598. 

  1599. void shiftVec(SwsVector *a, int shift){

  1600. 	SwsVector *shifted= getShiftedVec(a, shift);

  1601. 	free(a->coeff);

  1602. 	a->coeff= shifted->coeff;

  1603. 	a->length= shifted->length;

  1604. 	free(shifted);

  1605. }

  1606. 

  1607. void addVec(SwsVector *a, SwsVector *b){

  1608. 	SwsVector *sum= sumVec(a, b);

  1609. 	free(a->coeff);

  1610. 	a->coeff= sum->coeff;

  1611. 	a->length= sum->length;

  1612. 	free(sum);

  1613. }

  1614. 

  1615. void subVec(SwsVector *a, SwsVector *b){

  1616. 	SwsVector *diff= diffVec(a, b);

  1617. 	free(a->coeff);

  1618. 	a->coeff= diff->coeff;

  1619. 	a->length= diff->length;

  1620. 	free(diff);

  1621. }

  1622. 

  1623. void convVec(SwsVector *a, SwsVector *b){

  1624. 	SwsVector *conv= getConvVec(a, b);

  1625. 	free(a->coeff);

  1626. 	a->coeff= conv->coeff;

  1627. 	a->length= conv->length;

  1628. 	free(conv);

  1629. }

  1630. 

  1631. SwsVector *cloneVec(SwsVector *a){

  1632. 	double *coeff= memalign(sizeof(double), a->length*sizeof(double));

  1633. 	int i;

  1634. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1635. 

  1636. 	vec->coeff= coeff;

  1637. 	vec->length= a->length;

  1638. 

  1639. 	for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];

  1640. 

  1641. 	return vec;

  1642. }

  1643. 

  1644. void printVec(SwsVector *a){

  1645. 	int i;

  1646. 	double max=0;

  1647. 	double min=0;

  1648. 	double range;

  1649. 

  1650. 	for(i=0; i<a->length; i++)

  1651. 		if(a->coeff[i]>max) max= a->coeff[i];

  1652. 

  1653. 	for(i=0; i<a->length; i++)

  1654. 		if(a->coeff[i]<min) min= a->coeff[i];

  1655. 

  1656. 	range= max - min;

  1657. 

  1658. 	for(i=0; i<a->length; i++)

  1659. 	{

  1660. 		int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1661. 		printf("%1.3f ", a->coeff[i]);

  1662. 		for(;x>0; x--) printf(" ");

  1663. 		printf("|\n");

  1664. 	}

  1665. }

  1666. 

  1667. void freeVec(SwsVector *a){

  1668. 	if(!a) return;

  1669. 	if(a->coeff) free(a->coeff);

  1670. 	a->coeff=NULL;

  1671. 	a->length=0;

  1672. 	free(a);

  1673. }

  1674. 

  1675. void freeSwsContext(SwsContext *c){

  1676. 	int i;

  1677. 

  1678. 	if(!c) return;

  1679. 

  1680. 	if(c->lumPixBuf)

  1681. 	{

  1682. 		for(i=0; i<c->vLumBufSize; i++)

  1683. 		{

  1684. 			if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);

  1685. 			c->lumPixBuf[i]=NULL;

  1686. 		}

  1687. 		free(c->lumPixBuf);

  1688. 		c->lumPixBuf=NULL;

  1689. 	}

  1690. 

  1691. 	if(c->chrPixBuf)

  1692. 	{

  1693. 		for(i=0; i<c->vChrBufSize; i++)

  1694. 		{

  1695. 			if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);

  1696. 			c->chrPixBuf[i]=NULL;

  1697. 		}

  1698. 		free(c->chrPixBuf);

  1699. 		c->chrPixBuf=NULL;

  1700. 	}

  1701. 

  1702. 	if(c->vLumFilter) free(c->vLumFilter);

  1703. 	c->vLumFilter = NULL;

  1704. 	if(c->vChrFilter) free(c->vChrFilter);

  1705. 	c->vChrFilter = NULL;

  1706. 	if(c->hLumFilter) free(c->hLumFilter);

  1707. 	c->hLumFilter = NULL;

  1708. 	if(c->hChrFilter) free(c->hChrFilter);

  1709. 	c->hChrFilter = NULL;

  1710. 

  1711. 	if(c->vLumFilterPos) free(c->vLumFilterPos);

  1712. 	c->vLumFilterPos = NULL;

  1713. 	if(c->vChrFilterPos) free(c->vChrFilterPos);

  1714. 	c->vChrFilterPos = NULL;

  1715. 	if(c->hLumFilterPos) free(c->hLumFilterPos);

  1716. 	c->hLumFilterPos = NULL;

  1717. 	if(c->hChrFilterPos) free(c->hChrFilterPos);

  1718. 	c->hChrFilterPos = NULL;

  1719. 

  1720. 	if(c->lumMmxFilter) free(c->lumMmxFilter);

  1721. 	c->lumMmxFilter = NULL;

  1722. 	if(c->chrMmxFilter) free(c->chrMmxFilter);

  1723. 	c->chrMmxFilter = NULL;

  1724. 

  1725. 	free(c);

  1726. }

  1727.