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.
414 Commits
32 Branches
444MB
Tree: adeaecb9ae
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'adeaecb9ae'
${ noResults }
FFmpeg/postproc/swscale.c

1850 lines
51KB
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, RGB32, RGB24, Y8, Y800

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

  22.   BGR15/16 support dithering

  23.   

  24.   unscaled special converters

  25.   YV12/I420/IYUV -> BGR15/BGR16/BGR24/BGR32

  26.   YV12/I420/IYUV -> YV12/I420/IYUV

  27.   YUY2/BGR15/BGR16/BGR24/BGR32/RGB24/RGB32 -> same format

  28. */

  29. 

  30. #include <inttypes.h>

  31. #include <string.h>

  32. #include <math.h>

  33. #include <stdio.h>

  34. #include "../config.h"

  35. #include "../mangle.h"

  36. #ifdef HAVE_MALLOC_H

  37. #include <malloc.h>

  38. #endif

  39. #include "swscale.h"

  40. #include "../cpudetect.h"

  41. #include "../bswap.h"

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

  43. #include "rgb2rgb.h"

  44. #undef MOVNTQ

  45. #undef PAVGB

  46. 

  47. //#undef HAVE_MMX2

  48. //#define HAVE_3DNOW

  49. //#undef HAVE_MMX

  50. //#undef ARCH_X86

  51. #define DITHER1XBPP

  52. 

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

  54. 

  55. #ifdef MP_DEBUG

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

  57. #else

  58. #define ASSERT(x) ;

  59. #endif

  60. 

  61. #ifdef M_PI

  62. #define PI M_PI

  63. #else

  64. #define PI 3.14159265358979323846

  65. #endif

  66. 

  67. //FIXME replace this with something faster

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

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

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

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

  72. #define isPacked(x)    ((x)==IMGFMT_YUY2 || ((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR || ((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)

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

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

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

  76. 			|| (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\

  77. 			|| (x)==IMGFMT_Y800)

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

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

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

  81. 

  82. #define RGB2YUV_SHIFT 16

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

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

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

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

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

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

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

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

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

  92. 

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

  94. /*

  95. NOTES

  96. 

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

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

  99. 

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

  101. 

  102. TODO

  103. more intelligent missalignment avoidance for the horizontal scaler

  104. write special vertical cubic upscale version

  105. Optimize C code (yv12 / minmax)

  106. add support for packed pixel yuv input & output

  107. add support for Y8 output

  108. optimize bgr24 & bgr32

  109. add BGR4 output support

  110. write special BGR->BGR scaler

  111. deglobalize yuv2rgb*.c

  112. */

  113. 

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

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

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

  117. 

  118. #ifdef ARCH_X86

  119. #define CAN_COMPILE_X86_ASM

  120. #endif

  121. 

  122. #ifdef CAN_COMPILE_X86_ASM

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  138. 

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

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

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

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

  143. 

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

  145. 	0x0103010301030103LL,

  146. 	0x0200020002000200LL,};

  147. 

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

  149. 	0x0602060206020602LL,

  150. 	0x0004000400040004LL,};

  151. 

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

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

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

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

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

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

  158. 

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

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

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

  162. 

  163. // FIXME remove

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

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

  166. #endif

  167. 

  168. // clipping helper table for C implementations:

  169. static unsigned char clip_table[768];

  170. 

  171. static unsigned short clip_table16b[768];

  172. static unsigned short clip_table16g[768];

  173. static unsigned short clip_table16r[768];

  174. static unsigned short clip_table15b[768];

  175. static unsigned short clip_table15g[768];

  176. static unsigned short clip_table15r[768];

  177. 

  178. // yuv->rgb conversion tables:

  179. static    int yuvtab_2568[256];

  180. static    int yuvtab_3343[256];

  181. static    int yuvtab_0c92[256];

  182. static    int yuvtab_1a1e[256];

  183. static    int yuvtab_40cf[256];

  184. // Needed for cubic scaler to catch overflows

  185. static    int clip_yuvtab_2568[768];

  186. static    int clip_yuvtab_3343[768];

  187. static    int clip_yuvtab_0c92[768];

  188. static    int clip_yuvtab_1a1e[768];

  189. static    int clip_yuvtab_40cf[768];

  190. 

  191. //global sws_flags from the command line

  192. int sws_flags=2;

  193. 

  194. //global srcFilter

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

  196. 

  197. float sws_lum_gblur= 0.0;

  198. float sws_chr_gblur= 0.0;

  199. int sws_chr_vshift= 0;

  200. int sws_chr_hshift= 0;

  201. float sws_chr_sharpen= 0.0;

  202. float sws_lum_sharpen= 0.0;

  203. 

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

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

  206. static CpuCaps cpuCaps;

  207. 

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

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

  210. 

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

  212. 

  213. #ifdef CAN_COMPILE_X86_ASM

  214. void in_asm_used_var_warning_killer()

  215. {

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

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

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

  219.  if(i) i=0;

  220. }

  221. #endif

  222. 

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

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

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

  226. {

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

  228. 	int i;

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

  230. 	{

  231. 		int val=0;

  232. 		int j;

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

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

  235. 

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

  237. 	}

  238. 

  239. 	if(uDest != NULL)

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

  241. 		{

  242. 			int u=0;

  243. 			int v=0;

  244. 			int j;

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

  246. 			{

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

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

  249. 			}

  250. 

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

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

  253. 		}

  254. }

  255. 

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

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

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

  259. {

  260. 	if(dstFormat==IMGFMT_BGR32)

  261. 	{

  262. 		int i;

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

  264. 			int j;

  265. 			int Y1=0;

  266. 			int Y2=0;

  267. 			int U=0;

  268. 			int V=0;

  269. 			int Cb, Cr, Cg;

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

  271. 			{

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

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

  274. 			}

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

  276. 			{

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

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

  279. 			}

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

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

  282. 			U >>= 19;

  283. 			V >>= 19;

  284. 

  285. 			Cb= clip_yuvtab_40cf[U+ 256];

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

  287. 			Cr= clip_yuvtab_3343[V+ 256];

  288. 

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

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

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

  292. 

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

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

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

  296. 		}

  297. 	}

  298. 	else if(dstFormat==IMGFMT_BGR24)

  299. 	{

  300. 		int i;

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

  302. 			int j;

  303. 			int Y1=0;

  304. 			int Y2=0;

  305. 			int U=0;

  306. 			int V=0;

  307. 			int Cb, Cr, Cg;

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

  309. 			{

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

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

  312. 			}

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

  314. 			{

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

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

  317. 			}

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

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

  320. 			U >>= 19;

  321. 			V >>= 19;

  322. 

  323. 			Cb= clip_yuvtab_40cf[U+ 256];

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

  325. 			Cr= clip_yuvtab_3343[V+ 256];

  326. 

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

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

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

  330. 

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

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

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

  334. 			dest+=6;

  335. 		}

  336. 	}

  337. 	else if(dstFormat==IMGFMT_BGR16)

  338. 	{

  339. 		int i;

  340. #ifdef DITHER1XBPP

  341. 		static int ditherb1=1<<14;

  342. 		static int ditherg1=1<<13;

  343. 		static int ditherr1=2<<14;

  344. 		static int ditherb2=3<<14;

  345. 		static int ditherg2=3<<13;

  346. 		static int ditherr2=0<<14;

  347. 

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

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

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

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

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

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

  354. #else

  355. 		const int ditherb1=0;

  356. 		const int ditherg1=0;

  357. 		const int ditherr1=0;

  358. 		const int ditherb2=0;

  359. 		const int ditherg2=0;

  360. 		const int ditherr2=0;

  361. #endif

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

  363. 			int j;

  364. 			int Y1=0;

  365. 			int Y2=0;

  366. 			int U=0;

  367. 			int V=0;

  368. 			int Cb, Cr, Cg;

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

  370. 			{

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

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

  373. 			}

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

  375. 			{

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

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

  378. 			}

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

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

  381. 			U >>= 19;

  382. 			V >>= 19;

  383. 

  384. 			Cb= clip_yuvtab_40cf[U+ 256];

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

  386. 			Cr= clip_yuvtab_3343[V+ 256];

  387. 

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

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

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

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

  392. 

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

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

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

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

  397. 		}

  398. 	}

  399. 	else if(dstFormat==IMGFMT_BGR15)

  400. 	{

  401. 		int i;

  402. #ifdef DITHER1XBPP

  403. 		static int ditherb1=1<<14;

  404. 		static int ditherg1=1<<14;

  405. 		static int ditherr1=2<<14;

  406. 		static int ditherb2=3<<14;

  407. 		static int ditherg2=3<<14;

  408. 		static int ditherr2=0<<14;

  409. 

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

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

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

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

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

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

  416. #else

  417. 		const int ditherb1=0;

  418. 		const int ditherg1=0;

  419. 		const int ditherr1=0;

  420. 		const int ditherb2=0;

  421. 		const int ditherg2=0;

  422. 		const int ditherr2=0;

  423. #endif

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

  425. 			int j;

  426. 			int Y1=0;

  427. 			int Y2=0;

  428. 			int U=0;

  429. 			int V=0;

  430. 			int Cb, Cr, Cg;

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

  432. 			{

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

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

  435. 			}

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

  437. 			{

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

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

  440. 			}

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

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

  443. 			U >>= 19;

  444. 			V >>= 19;

  445. 

  446. 			Cb= clip_yuvtab_40cf[U+ 256];

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

  448. 			Cr= clip_yuvtab_3343[V+ 256];

  449. 

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

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

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

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

  454. 

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

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

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

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

  459. 		}

  460. 	}

  461. }

  462. 

  463. 

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

  465. //Plain C versions

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

  467. #define COMPILE_C

  468. #endif

  469. 

  470. #ifdef CAN_COMPILE_X86_ASM

  471. 

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

  473. #define COMPILE_MMX

  474. #endif

  475. 

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

  477. #define COMPILE_MMX2

  478. #endif

  479. 

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

  481. #define COMPILE_3DNOW

  482. #endif

  483. #endif //CAN_COMPILE_X86_ASM

  484. 

  485. #undef HAVE_MMX

  486. #undef HAVE_MMX2

  487. #undef HAVE_3DNOW

  488. 

  489. #ifdef COMPILE_C

  490. #undef HAVE_MMX

  491. #undef HAVE_MMX2

  492. #undef HAVE_3DNOW

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

  494. #include "swscale_template.c"

  495. #endif

  496. 

  497. #ifdef CAN_COMPILE_X86_ASM

  498. 

  499. //X86 versions

  500. /*

  501. #undef RENAME

  502. #undef HAVE_MMX

  503. #undef HAVE_MMX2

  504. #undef HAVE_3DNOW

  505. #define ARCH_X86

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

  507. #include "swscale_template.c"

  508. */

  509. //MMX versions

  510. #ifdef COMPILE_MMX

  511. #undef RENAME

  512. #define HAVE_MMX

  513. #undef HAVE_MMX2

  514. #undef HAVE_3DNOW

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

  516. #include "swscale_template.c"

  517. #endif

  518. 

  519. //MMX2 versions

  520. #ifdef COMPILE_MMX2

  521. #undef RENAME

  522. #define HAVE_MMX

  523. #define HAVE_MMX2

  524. #undef HAVE_3DNOW

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

  526. #include "swscale_template.c"

  527. #endif

  528. 

  529. //3DNOW versions

  530. #ifdef COMPILE_3DNOW

  531. #undef RENAME

  532. #define HAVE_MMX

  533. #undef HAVE_MMX2

  534. #define HAVE_3DNOW

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

  536. #include "swscale_template.c"

  537. #endif

  538. 

  539. #endif //CAN_COMPILE_X86_ASM

  540. 

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

  542. 

  543. 

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

  545. // will use sws_flags from the command line

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

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

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

  549. 

  550. 	static SwsContext *context=NULL;

  551. 	int dstFormat;

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

  553. 

  554. 	switch(dstbpp)

  555. 	{

  556. 		case 8 : dstFormat= IMGFMT_Y8;		break;

  557. 		case 12: dstFormat= IMGFMT_YV12;	break;

  558. 		case 15: dstFormat= IMGFMT_BGR15;	break;

  559. 		case 16: dstFormat= IMGFMT_BGR16;	break;

  560. 		case 24: dstFormat= IMGFMT_BGR24;	break;

  561. 		case 32: dstFormat= IMGFMT_BGR32;	break;

  562. 		default: return;

  563. 	}

  564. 

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

  566. 

  567. 	context->swScale(context, src, srcStride, srcSliceY, srcSliceH, dst, dstStride3);

  568. }

  569. 

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

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

  572. {

  573. 	int flags=0;

  574. 	static int firstTime=1;

  575. 

  576. #ifdef ARCH_X86

  577. 	if(gCpuCaps.hasMMX)

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

  579. #endif

  580. 	if(firstTime)

  581. 	{

  582. 		firstTime=0;

  583. 		flags= SWS_PRINT_INFO;

  584. 	}

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

  586. 

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

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

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

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

  591. 

  592. 	if(sws_lum_gblur!=0.0){

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

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

  595. 	}else{

  596. 		src_filter.lumH= getIdentityVec();

  597. 		src_filter.lumV= getIdentityVec();

  598. 	}

  599. 

  600. 	if(sws_chr_gblur!=0.0){

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

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

  603. 	}else{

  604. 		src_filter.chrH= getIdentityVec();

  605. 		src_filter.chrV= getIdentityVec();

  606. 	}

  607. 

  608. 	if(sws_chr_sharpen!=0.0){

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

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

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

  612. 		addVec(id, g);

  613. 		convVec(src_filter.chrH, id);

  614. 		convVec(src_filter.chrV, id);

  615. 		freeVec(g);

  616. 		freeVec(id);

  617. 	}

  618. 

  619. 	if(sws_lum_sharpen!=0.0){

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

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

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

  623. 		addVec(id, g);

  624. 		convVec(src_filter.lumH, id);

  625. 		convVec(src_filter.lumV, id);

  626. 		freeVec(g);

  627. 		freeVec(id);

  628. 	}

  629. 

  630. 	if(sws_chr_hshift)

  631. 		shiftVec(src_filter.chrH, sws_chr_hshift);

  632. 

  633. 	if(sws_chr_vshift)

  634. 		shiftVec(src_filter.chrV, sws_chr_vshift);

  635. 

  636. 	normalizeVec(src_filter.chrH, 1.0);

  637. 	normalizeVec(src_filter.chrV, 1.0);

  638. 	normalizeVec(src_filter.lumH, 1.0);

  639. 	normalizeVec(src_filter.lumV, 1.0);

  640. 

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

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

  643. 

  644. 	switch(sws_flags)

  645. 	{

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

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

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

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

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

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

  652. 		default:flags|= SWS_BILINEAR; break;

  653. 	}

  654. 

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

  656. }

  657. 

  658. 

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

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

  661. 			      SwsVector *srcFilter, SwsVector *dstFilter)

  662. {

  663. 	int i;

  664. 	int filterSize;

  665. 	int filter2Size;

  666. 	int minFilterSize;

  667. 	double *filter=NULL;

  668. 	double *filter2=NULL;

  669. #ifdef ARCH_X86

  670. 	if(gCpuCaps.hasMMX)

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

  672. #endif

  673. 

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

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

  676. 

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

  678. 	{

  679. 		int i;

  680. 		filterSize= 1;

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

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

  683. 

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

  685. 		{

  686. 			filter[i*filterSize]=1;

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

  688. 		}

  689. 

  690. 	}

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

  692. 	{

  693. 		int i;

  694. 		int xDstInSrc;

  695. 		filterSize= 1;

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

  697. 		

  698. 		xDstInSrc= xInc/2 - 0x8000;

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

  700. 		{

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

  702. 

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

  704. 			filter[i]= 1.0;

  705. 			xDstInSrc+= xInc;

  706. 		}

  707. 	}

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

  709. 	{

  710. 		int i;

  711. 		int xDstInSrc;

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

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

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

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

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

  717. 

  718. 		xDstInSrc= xInc/2 - 0x8000;

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

  720. 		{

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

  722. 			int j;

  723. 

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

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

  726. 			{

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

  728. 				double y1,y2,y3,y4;

  729. 				double A= -0.6;

  730. 				if(flags & SWS_BICUBIC){

  731. 						// Equation is from VirtualDub

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

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

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

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

  736. 				}else{

  737. 						// cubic interpolation (derived it myself)

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

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

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

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

  742. 				}

  743. 

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

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

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

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

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

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

  750. 			}

  751. 			else

  752. 			{

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

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

  755. 				{

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

  757. 					double coeff= 1.0 - d;

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

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

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

  761. 					xx++;

  762. 				}

  763. 			}

  764. 			xDstInSrc+= xInc;

  765. 		}

  766. 	}

  767. 	else // downscale

  768. 	{

  769. 		int xDstInSrc;

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

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

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

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

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

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

  776. 

  777. 		xDstInSrc= xInc/2 - 0x8000;

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

  779. 		{

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

  781. 			int j;

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

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

  784. 			{

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

  786. 				double coeff;

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

  788. 				{

  789. 					double A= -0.75;

  790. //					d*=2;

  791. 					// Equation is from VirtualDub

  792. 					if(d<1.0)

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

  794. 					else if(d<2.0)

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

  796. 					else

  797. 						coeff=0.0;

  798. 				}

  799. 				else if(flags & SWS_AREA)

  800. 				{

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

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

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

  804. 					else coeff=0.0;

  805. 				}

  806. 				else

  807. 				{

  808. 					coeff= 1.0 - d;

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

  810. 				}

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

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

  813. 				xx++;

  814. 			}

  815. 			xDstInSrc+= xInc;

  816. 		}

  817. 	}

  818. 

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

  820. 	   free(filter);

  821. 	*/

  822. 	filter2Size= filterSize;

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

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

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

  826. 

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

  828. 	{

  829. 		int j;

  830. 		SwsVector scaleFilter;

  831. 		SwsVector *outVec;

  832. 

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

  834. 		scaleFilter.length= filterSize;

  835. 

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

  837. 		else	      outVec= &scaleFilter;

  838. 

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

  840. 		//FIXME dstFilter

  841. 

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

  843. 		{

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

  845. 		}

  846. 

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

  848. 

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

  850. 	}

  851. 	free(filter); filter=NULL;

  852. 

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

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

  855. 	minFilterSize= 0;

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

  857. 	{

  858. 		int min= filter2Size;

  859. 		int j;

  860. 		double cutOff=0.0;

  861. 

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

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

  864. 		{

  865. 			int k;

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

  867. 

  868. 			if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;

  869. 

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

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

  872. 

  873. 			// Move filter coeffs left

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

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

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

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

  878. 		}

  879. 

  880. 		cutOff=0.0;

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

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

  883. 		{

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

  885. 

  886. 			if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;

  887. 			min--;

  888. 		}

  889. 

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

  891. 	}

  892. 

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

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

  895. 	*outFilterSize= filterSize;

  896. 

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

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

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

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

  901. 	{

  902. 		int j;

  903. 

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

  905. 		{

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

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

  908. 		}

  909. 	}

  910. 	free(filter2); filter2=NULL;

  911. 	

  912. 	ASSERT(filterSize > 0)

  913. 

  914. 	//FIXME try to align filterpos if possible

  915. 

  916. 	//fix borders

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

  918. 	{

  919. 		int j;

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

  921. 		{

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

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

  924. 			{

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

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

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

  928. 			}

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

  930. 		}

  931. 

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

  933. 		{

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

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

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

  937. 			{

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

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

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

  941. 			}

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

  943. 		}

  944. 	}

  945. 

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

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

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

  949. 

  950. 	/* Normalize & Store in outFilter */

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

  952. 	{

  953. 		int j;

  954. 		double sum=0;

  955. 		double scale= one;

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

  957. 		{

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

  959. 		}

  960. 		scale/= sum;

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

  962. 		{

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

  964. 		}

  965. 	}

  966. 	

  967. 	(*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end

  968. 	for(i=0; i<*outFilterSize; i++)

  969. 	{

  970. 		int j= dstW*(*outFilterSize);

  971. 		(*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];

  972. 	}

  973. 

  974. 	free(filter);

  975. }

  976. 

  977. #ifdef ARCH_X86

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

  979. {

  980. 	uint8_t *fragment;

  981. 	int imm8OfPShufW1;

  982. 	int imm8OfPShufW2;

  983. 	int fragmentLength;

  984. 

  985. 	int xpos, i;

  986. 

  987. 	// create an optimized horizontal scaling routine

  988. 

  989. 	//code fragment

  990. 

  991. 	asm volatile(

  992. 		"jmp 9f				\n\t"

  993. 	// Begin

  994. 		"0:				\n\t"

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

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

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

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

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

  1000. 		"punpcklbw %%mm7, %%mm0	\n\t"

  1001. 		"addw %%bx, %%cx		\n\t" //2*xalpha += (4*lumXInc)&0xFFFF

  1002. 		"pshufw $0xFF, %%mm1, %%mm1	\n\t"

  1003. 		"1:				\n\t"

  1004. 		"adcl %%edx, %%esi		\n\t" //xx+= (4*lumXInc)>>16 + carry

  1005. 		"pshufw $0xFF, %%mm0, %%mm0	\n\t"

  1006. 		"2:				\n\t"

  1007. 		"psrlw $9, %%mm3		\n\t"

  1008. 		"psubw %%mm1, %%mm0		\n\t"

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

  1010. 		"paddw %%mm6, %%mm2		\n\t" // 2*alpha += xpos&0xFFFF

  1011. 		"psllw $7, %%mm1		\n\t"

  1012. 		"paddw %%mm1, %%mm0		\n\t"

  1013. 

  1014. 		"movq %%mm0, (%%edi, %%eax)	\n\t"

  1015. 

  1016. 		"addl $8, %%eax			\n\t"

  1017. 	// End

  1018. 		"9:				\n\t"

  1019. //		"int $3\n\t"

  1020. 		"leal 0b, %0			\n\t"

  1021. 		"leal 1b, %1			\n\t"

  1022. 		"leal 2b, %2			\n\t"

  1023. 		"decl %1			\n\t"

  1024. 		"decl %2			\n\t"

  1025. 		"subl %0, %1			\n\t"

  1026. 		"subl %0, %2			\n\t"

  1027. 		"leal 9b, %3			\n\t"

  1028. 		"subl %0, %3			\n\t"

  1029. 		:"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),

  1030. 		"=r" (fragmentLength)

  1031. 	);

  1032. 

  1033. 	xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers

  1034. 

  1035. 	for(i=0; i<dstW/8; i++)

  1036. 	{

  1037. 		int xx=xpos>>16;

  1038. 

  1039. 		if((i&3) == 0)

  1040. 		{

  1041. 			int a=0;

  1042. 			int b=((xpos+xInc)>>16) - xx;

  1043. 			int c=((xpos+xInc*2)>>16) - xx;

  1044. 			int d=((xpos+xInc*3)>>16) - xx;

  1045. 

  1046. 			memcpy(funnyCode + fragmentLength*i/4, fragment, fragmentLength);

  1047. 

  1048. 			funnyCode[fragmentLength*i/4 + imm8OfPShufW1]=

  1049. 			funnyCode[fragmentLength*i/4 + imm8OfPShufW2]=

  1050. 				a | (b<<2) | (c<<4) | (d<<6);

  1051. 

  1052. 			// if we dont need to read 8 bytes than dont :), reduces the chance of

  1053. 			// crossing a cache line

  1054. 			if(d<3) funnyCode[fragmentLength*i/4 + 1]= 0x6E;

  1055. 

  1056. 			funnyCode[fragmentLength*(i+4)/4]= RET;

  1057. 		}

  1058. 		xpos+=xInc;

  1059. 	}

  1060. }

  1061. #endif // ARCH_X86

  1062. 

  1063. //FIXME remove

  1064. void SwScale_Init(){

  1065. }

  1066. 

  1067. static void globalInit(){

  1068.     // generating tables:

  1069.     int i;

  1070.     for(i=0; i<768; i++){

  1071. 	int c= MIN(MAX(i-256, 0), 255);

  1072. 	clip_table[i]=c;

  1073. 	yuvtab_2568[c]= clip_yuvtab_2568[i]=(0x2568*(c-16))+(256<<13);

  1074. 	yuvtab_3343[c]= clip_yuvtab_3343[i]=0x3343*(c-128);

  1075. 	yuvtab_0c92[c]= clip_yuvtab_0c92[i]=-0x0c92*(c-128);

  1076. 	yuvtab_1a1e[c]= clip_yuvtab_1a1e[i]=-0x1a1e*(c-128);

  1077. 	yuvtab_40cf[c]= clip_yuvtab_40cf[i]=0x40cf*(c-128);

  1078.     }

  1079. 

  1080.     for(i=0; i<768; i++)

  1081.     {

  1082. 	int v= clip_table[i];

  1083. 	clip_table16b[i]= le2me_16( v>>3);

  1084. 	clip_table16g[i]= le2me_16((v<<3)&0x07E0);

  1085. 	clip_table16r[i]= le2me_16((v<<8)&0xF800);

  1086. 	clip_table15b[i]= le2me_16( v>>3);

  1087. 	clip_table15g[i]= le2me_16((v<<2)&0x03E0);

  1088. 	clip_table15r[i]= le2me_16((v<<7)&0x7C00);

  1089.     }

  1090. 

  1091. cpuCaps= gCpuCaps;

  1092. 

  1093. #ifdef RUNTIME_CPUDETECT

  1094. #ifdef CAN_COMPILE_X86_ASM

  1095. 	// ordered per speed fasterst first

  1096. 	if(gCpuCaps.hasMMX2)

  1097. 		swScale= swScale_MMX2;

  1098. 	else if(gCpuCaps.has3DNow)

  1099. 		swScale= swScale_3DNow;

  1100. 	else if(gCpuCaps.hasMMX)

  1101. 		swScale= swScale_MMX;

  1102. 	else

  1103. 		swScale= swScale_C;

  1104. 

  1105. #else

  1106. 	swScale= swScale_C;

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

  1108. #endif

  1109. #else //RUNTIME_CPUDETECT

  1110. #ifdef HAVE_MMX2

  1111. 	swScale= swScale_MMX2;

  1112. 	cpuCaps.has3DNow = 0;

  1113. #elif defined (HAVE_3DNOW)

  1114. 	swScale= swScale_3DNow;

  1115. 	cpuCaps.hasMMX2 = 0;

  1116. #elif defined (HAVE_MMX)

  1117. 	swScale= swScale_MMX;

  1118. 	cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;

  1119. #else

  1120. 	swScale= swScale_C;

  1121. 	cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;

  1122. #endif

  1123. #endif //!RUNTIME_CPUDETECT

  1124. }

  1125. 

  1126. /* Warper functions for yuv2bgr */

  1127. static void planarYuvToBgr(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1128.              int srcSliceH, uint8_t* dstParam[], int dstStride[]){

  1129. 	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;

  1130. 

  1131. 	if(c->srcFormat==IMGFMT_YV12)

  1132. 		yuv2rgb( dst,src[0],src[1],src[2],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );

  1133. 	else /* I420 & IYUV */

  1134. 		yuv2rgb( dst,src[0],src[2],src[1],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );

  1135. }

  1136. 

  1137. /* unscaled copy like stuff (assumes nearly identical formats) */

  1138. static void simpleCopy(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,

  1139.              int srcSliceH, uint8_t* dstParam[], int dstStride[]){

  1140. 

  1141. 	int srcStride[3];

  1142. 	uint8_t *src[3];

  1143. 	uint8_t *dst[3];

  1144. 

  1145. 	if(c->srcFormat == IMGFMT_I420){

  1146. 		src[0]= srcParam[0];

  1147. 		src[1]= srcParam[2];

  1148. 		src[2]= srcParam[1];

  1149. 		srcStride[0]= srcStrideParam[0];

  1150. 		srcStride[1]= srcStrideParam[2];

  1151. 		srcStride[2]= srcStrideParam[1];

  1152. 	}

  1153. 	else if(c->srcFormat==IMGFMT_YV12){

  1154. 		src[0]= srcParam[0];

  1155. 		src[1]= srcParam[1];

  1156. 		src[2]= srcParam[2];

  1157. 		srcStride[0]= srcStrideParam[0];

  1158. 		srcStride[1]= srcStrideParam[1];

  1159. 		srcStride[2]= srcStrideParam[2];

  1160. 	}

  1161. 	else if(isPacked(c->srcFormat) || isGray(c->srcFormat)){

  1162. 		src[0]= srcParam[0];

  1163. 		src[1]=

  1164. 		src[2]= NULL;

  1165. 		srcStride[0]= srcStrideParam[0];

  1166. 		srcStride[1]=

  1167. 		srcStride[2]= 0;

  1168. 	}

  1169. 

  1170. 	if(c->dstFormat == IMGFMT_I420){

  1171. 		dst[0]= dstParam[0];

  1172. 		dst[1]= dstParam[2];

  1173. 		dst[2]= dstParam[1];

  1174. 		

  1175. 	}else{

  1176. 		dst[0]= dstParam[0];

  1177. 		dst[1]= dstParam[1];

  1178. 		dst[2]= dstParam[2];

  1179. 	}

  1180. 

  1181. 	if(isPacked(c->srcFormat))

  1182. 	{

  1183. 		if(dstStride[0]==srcStride[0])

  1184. 			memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);

  1185. 		else

  1186. 		{

  1187. 			int i;

  1188. 			uint8_t *srcPtr= src[0];

  1189. 			uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;

  1190. 			int length=0;

  1191. 

  1192. 			/* universal length finder */

  1193. 			while(length+c->srcW <= dstStride[0] 

  1194. 			   && length+c->srcW <= srcStride[0]) length+= c->srcW;

  1195. 			ASSERT(length!=0);

  1196. 

  1197. 			for(i=0; i<srcSliceH; i++)

  1198. 			{

  1199. 				memcpy(dstPtr, srcPtr, length);

  1200. 				srcPtr+= srcStride[0];

  1201. 				dstPtr+= dstStride[0];

  1202. 			}

  1203. 		}

  1204. 	}

  1205. 	else 

  1206. 	{ /* Planar YUV */

  1207. 		int plane;

  1208. 		for(plane=0; plane<3; plane++)

  1209. 		{

  1210. 			int length= plane==0 ? c->srcW  : ((c->srcW+1)>>1);

  1211. 			int y=      plane==0 ? srcSliceY: ((srcSliceY+1)>>1);

  1212. 			int height= plane==0 ? srcSliceH: ((srcSliceH+1)>>1);

  1213. 

  1214. 			if(dstStride[plane]==srcStride[plane])

  1215. 				memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);

  1216. 			else

  1217. 			{

  1218. 				int i;

  1219. 				uint8_t *srcPtr= src[plane];

  1220. 				uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;

  1221. 				for(i=0; i<height; i++)

  1222. 				{

  1223. 					memcpy(dstPtr, srcPtr, length);

  1224. 					srcPtr+= srcStride[plane];

  1225. 					dstPtr+= dstStride[plane];

  1226. 				}

  1227. 			}

  1228. 		}

  1229. 	}

  1230. }

  1231. 

  1232. SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,

  1233.                          SwsFilter *srcFilter, SwsFilter *dstFilter){

  1234. 

  1235. 	SwsContext *c;

  1236. 	int i;

  1237. 	int usesFilter;

  1238. 	SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};

  1239. 

  1240. #ifdef ARCH_X86

  1241. 	if(gCpuCaps.hasMMX)

  1242. 		asm volatile("emms\n\t"::: "memory");

  1243. #endif

  1244. 

  1245. 	if(swScale==NULL) globalInit();

  1246. 

  1247. 	/* avoid dupplicate Formats, so we dont need to check to much */

  1248. 	if(srcFormat==IMGFMT_IYUV) srcFormat=IMGFMT_I420;

  1249. 	if(srcFormat==IMGFMT_Y8)   srcFormat=IMGFMT_Y800;

  1250. 	if(dstFormat==IMGFMT_Y8)   dstFormat=IMGFMT_Y800;

  1251. 

  1252. 	if(!isSupportedIn(srcFormat)) 

  1253. 	{

  1254. 		fprintf(stderr, "swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));

  1255. 		return NULL;

  1256. 	}

  1257. 	if(!isSupportedOut(dstFormat))

  1258. 	{

  1259. 		fprintf(stderr, "swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));

  1260. 		return NULL;

  1261. 	}

  1262. 

  1263. 	/* sanity check */

  1264. 	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

  1265. 	{

  1266. 		fprintf(stderr, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 

  1267. 			srcW, srcH, dstW, dstH);

  1268. 		return NULL;

  1269. 	}

  1270. 

  1271. 	if(!dstFilter) dstFilter= &dummyFilter;

  1272. 	if(!srcFilter) srcFilter= &dummyFilter;

  1273. 

  1274. 	c= memalign(64, sizeof(SwsContext));

  1275. 	memset(c, 0, sizeof(SwsContext));

  1276. 

  1277. 	c->srcW= srcW;

  1278. 	c->srcH= srcH;

  1279. 	c->dstW= dstW;

  1280. 	c->dstH= dstH;

  1281. 	c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;

  1282. 	c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;

  1283. 	c->flags= flags;

  1284. 	c->dstFormat= dstFormat;

  1285. 	c->srcFormat= srcFormat;

  1286. 

  1287. 	usesFilter=0;

  1288. 	if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;

  1289. 	if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;

  1290. 	if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;

  1291. 	if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;

  1292. 	if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;

  1293. 	if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;

  1294. 	if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;

  1295. 	if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;

  1296. 	

  1297. 	/* special Cases */

  1298. 	if(srcW==dstW && srcH==dstH && !usesFilter)

  1299. 	{

  1300. 		/* yuv2bgr */

  1301. 		if(isPlanarYUV(srcFormat) && isBGR(dstFormat))

  1302. 		{

  1303. 			// FIXME multiple yuv2rgb converters wont work that way cuz that thing is full of globals&statics

  1304. 			yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);

  1305. 			c->swScale= planarYuvToBgr;

  1306. 

  1307. 			if(flags&SWS_PRINT_INFO)

  1308. 				printf("SwScaler: using unscaled %s -> %s special converter\n", 

  1309. 					vo_format_name(srcFormat), vo_format_name(dstFormat));

  1310. 			return c;

  1311. 		}

  1312. 

  1313. 		/* simple copy */

  1314. 		if(srcFormat == dstFormat || (isPlanarYUV(srcFormat) && isPlanarYUV(dstFormat)))

  1315. 		{

  1316. 			c->swScale= simpleCopy;

  1317. 

  1318. 			if(flags&SWS_PRINT_INFO)

  1319. 				printf("SwScaler: using unscaled %s -> %s special converter\n", 

  1320. 					vo_format_name(srcFormat), vo_format_name(dstFormat));

  1321. 			return c;

  1322. 		}

  1323. 	}

  1324. 

  1325. 	if(cpuCaps.hasMMX2)

  1326. 	{

  1327. 		c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;

  1328. 		if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))

  1329. 		{

  1330. 			if(flags&SWS_PRINT_INFO)

  1331. 				fprintf(stderr, "SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");

  1332. 		}

  1333. 	}

  1334. 	else

  1335. 		c->canMMX2BeUsed=0;

  1336. 

  1337. 

  1338. 	/* dont use full vertical UV input/internaly if the source doesnt even have it */

  1339. 	if(isHalfChrV(srcFormat)) c->flags= flags= flags&(~SWS_FULL_CHR_V);

  1340. 	/* dont use full horizontal UV input if the source doesnt even have it */

  1341. 	if(isHalfChrH(srcFormat)) c->flags= flags= flags&(~SWS_FULL_CHR_H_INP);

  1342. 	/* dont use full horizontal UV internally if the destination doesnt even have it */

  1343. 	if(isHalfChrH(dstFormat)) c->flags= flags= flags&(~SWS_FULL_CHR_H_INT);

  1344. 

  1345. 	if(flags&SWS_FULL_CHR_H_INP)	c->chrSrcW= srcW;

  1346. 	else				c->chrSrcW= (srcW+1)>>1;

  1347. 

  1348. 	if(flags&SWS_FULL_CHR_H_INT)	c->chrDstW= dstW;

  1349. 	else				c->chrDstW= (dstW+1)>>1;

  1350. 

  1351. 	if(flags&SWS_FULL_CHR_V)	c->chrSrcH= srcH;

  1352. 	else				c->chrSrcH= (srcH+1)>>1;

  1353. 

  1354. 	if(isHalfChrV(dstFormat))	c->chrDstH= (dstH+1)>>1;

  1355. 	else				c->chrDstH= dstH;

  1356. 

  1357. 	c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;

  1358. 	c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;

  1359. 

  1360. 

  1361. 	// match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst

  1362. 	// but only for the FAST_BILINEAR mode otherwise do correct scaling

  1363. 	// n-2 is the last chrominance sample available

  1364. 	// this is not perfect, but noone shuld notice the difference, the more correct variant

  1365. 	// would be like the vertical one, but that would require some special code for the

  1366. 	// first and last pixel

  1367. 	if(flags&SWS_FAST_BILINEAR)

  1368. 	{

  1369. 		if(c->canMMX2BeUsed)

  1370. 		{

  1371. 			c->lumXInc+= 20;

  1372. 			c->chrXInc+= 20;

  1373. 		}

  1374. 		//we dont use the x86asm scaler if mmx is available

  1375. 		else if(cpuCaps.hasMMX)

  1376. 		{

  1377. 			c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;

  1378. 			c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;

  1379. 		}

  1380. 	}

  1381. 

  1382. 	/* precalculate horizontal scaler filter coefficients */

  1383. 	{

  1384. 		const int filterAlign= cpuCaps.hasMMX ? 4 : 1;

  1385. 

  1386. 		initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,

  1387. 				 srcW      ,       dstW, filterAlign, 1<<14, flags,

  1388. 				 srcFilter->lumH, dstFilter->lumH);

  1389. 		initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,

  1390. 				(srcW+1)>>1, c->chrDstW, filterAlign, 1<<14, flags,

  1391. 				 srcFilter->chrH, dstFilter->chrH);

  1392. 

  1393. #ifdef ARCH_X86

  1394. // cant downscale !!!

  1395. 		if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))

  1396. 		{

  1397. 			initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode);

  1398. 			initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode);

  1399. 		}

  1400. #endif

  1401. 	} // Init Horizontal stuff

  1402. 

  1403. 

  1404. 

  1405. 	/* precalculate vertical scaler filter coefficients */

  1406. 	initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,

  1407. 			srcH      ,        dstH, 1, (1<<12)-4, flags,

  1408. 			srcFilter->lumV, dstFilter->lumV);

  1409. 	initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,

  1410. 			(srcH+1)>>1, c->chrDstH, 1, (1<<12)-4, flags,

  1411. 			 srcFilter->chrV, dstFilter->chrV);

  1412. 

  1413. 	// Calculate Buffer Sizes so that they wont run out while handling these damn slices

  1414. 	c->vLumBufSize= c->vLumFilterSize;

  1415. 	c->vChrBufSize= c->vChrFilterSize;

  1416. 	for(i=0; i<dstH; i++)

  1417. 	{

  1418. 		int chrI= i*c->chrDstH / dstH;

  1419. 		int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1420. 				 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<1));

  1421. 		nextSlice&= ~1; // Slices start at even boundaries

  1422. 		if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1423. 			c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];

  1424. 		if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>1))

  1425. 			c->vChrBufSize= (nextSlice>>1) - c->vChrFilterPos[chrI];

  1426. 	}

  1427. 

  1428. 	// allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1429. 	c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));

  1430. 	c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));

  1431. 	//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)

  1432. 	for(i=0; i<c->vLumBufSize; i++)

  1433. 		c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);

  1434. 	for(i=0; i<c->vChrBufSize; i++)

  1435. 		c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);

  1436. 

  1437. 	//try to avoid drawing green stuff between the right end and the stride end

  1438. 	for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);

  1439. 	for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);

  1440. 

  1441. 	ASSERT(c->chrDstH <= dstH)

  1442. 

  1443. 	// pack filter data for mmx code

  1444. 	if(cpuCaps.hasMMX)

  1445. 	{

  1446. 		c->lumMmxFilter= (int16_t*)memalign(8, c->vLumFilterSize*      dstH*4*sizeof(int16_t));

  1447. 		c->chrMmxFilter= (int16_t*)memalign(8, c->vChrFilterSize*c->chrDstH*4*sizeof(int16_t));

  1448. 		for(i=0; i<c->vLumFilterSize*dstH; i++)

  1449. 			c->lumMmxFilter[4*i]=c->lumMmxFilter[4*i+1]=c->lumMmxFilter[4*i+2]=c->lumMmxFilter[4*i+3]=

  1450. 				c->vLumFilter[i];

  1451. 		for(i=0; i<c->vChrFilterSize*c->chrDstH; i++)

  1452. 			c->chrMmxFilter[4*i]=c->chrMmxFilter[4*i+1]=c->chrMmxFilter[4*i+2]=c->chrMmxFilter[4*i+3]=

  1453. 				c->vChrFilter[i];

  1454. 	}

  1455. 

  1456. 	if(flags&SWS_PRINT_INFO)

  1457. 	{

  1458. #ifdef DITHER1XBPP

  1459. 		char *dither= " dithered";

  1460. #else

  1461. 		char *dither= "";

  1462. #endif

  1463. 		if(flags&SWS_FAST_BILINEAR)

  1464. 			fprintf(stderr, "\nSwScaler: FAST_BILINEAR scaler, ");

  1465. 		else if(flags&SWS_BILINEAR)

  1466. 			fprintf(stderr, "\nSwScaler: BILINEAR scaler, ");

  1467. 		else if(flags&SWS_BICUBIC)

  1468. 			fprintf(stderr, "\nSwScaler: BICUBIC scaler, ");

  1469. 		else if(flags&SWS_X)

  1470. 			fprintf(stderr, "\nSwScaler: Experimental scaler, ");

  1471. 		else if(flags&SWS_POINT)

  1472. 			fprintf(stderr, "\nSwScaler: Nearest Neighbor / POINT scaler, ");

  1473. 		else if(flags&SWS_AREA)

  1474. 			fprintf(stderr, "\nSwScaler: Area Averageing scaler, ");

  1475. 		else

  1476. 			fprintf(stderr, "\nSwScaler: ehh flags invalid?! ");

  1477. 

  1478. 		if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)

  1479. 			fprintf(stderr, "from %s to%s %s ", 

  1480. 				vo_format_name(srcFormat), dither, vo_format_name(dstFormat));

  1481. 		else

  1482. 			fprintf(stderr, "from %s to %s ", 

  1483. 				vo_format_name(srcFormat), vo_format_name(dstFormat));

  1484. 

  1485. 		if(cpuCaps.hasMMX2)

  1486. 			fprintf(stderr, "using MMX2\n");

  1487. 		else if(cpuCaps.has3DNow)

  1488. 			fprintf(stderr, "using 3DNOW\n");

  1489. 		else if(cpuCaps.hasMMX)

  1490. 			fprintf(stderr, "using MMX\n");

  1491. 		else

  1492. 			fprintf(stderr, "using C\n");

  1493. 	}

  1494. 

  1495. 	if((flags & SWS_PRINT_INFO) && verbose)

  1496. 	{

  1497. 		if(cpuCaps.hasMMX)

  1498. 		{

  1499. 			if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1500. 				printf("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1501. 			else

  1502. 			{

  1503. 				if(c->hLumFilterSize==4)

  1504. 					printf("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");

  1505. 				else if(c->hLumFilterSize==8)

  1506. 					printf("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");

  1507. 				else

  1508. 					printf("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");

  1509. 

  1510. 				if(c->hChrFilterSize==4)

  1511. 					printf("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");

  1512. 				else if(c->hChrFilterSize==8)

  1513. 					printf("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");

  1514. 				else

  1515. 					printf("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");

  1516. 			}

  1517. 		}

  1518. 		else

  1519. 		{

  1520. #ifdef ARCH_X86

  1521. 			printf("SwScaler: using X86-Asm scaler for horizontal scaling\n");

  1522. #else

  1523. 			if(flags & SWS_FAST_BILINEAR)

  1524. 				printf("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");

  1525. 			else

  1526. 				printf("SwScaler: using C scaler for horizontal scaling\n");

  1527. #endif

  1528. 		}

  1529. 		if(isPlanarYUV(dstFormat))

  1530. 		{

  1531. 			if(c->vLumFilterSize==1)

  1532. 				printf("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");

  1533. 			else

  1534. 				printf("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");

  1535. 		}

  1536. 		else

  1537. 		{

  1538. 			if(c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1539. 				printf("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1540. 				       "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");

  1541. 			else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)

  1542. 				printf("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");

  1543. 			else

  1544. 				printf("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");

  1545. 		}

  1546. 

  1547. 		if(dstFormat==IMGFMT_BGR24)

  1548. 			printf("SwScaler: using %s YV12->BGR24 Converter\n",

  1549. 				cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));

  1550. 		else if(dstFormat==IMGFMT_BGR32)

  1551. 			printf("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");

  1552. 		else if(dstFormat==IMGFMT_BGR16)

  1553. 			printf("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");

  1554. 		else if(dstFormat==IMGFMT_BGR15)

  1555. 			printf("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");

  1556. 

  1557. 		printf("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1558. 	}

  1559. 	if((flags & SWS_PRINT_INFO) && verbose>1)

  1560. 	{

  1561. 		printf("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1562. 			c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1563. 		printf("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1564. 			c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1565. 	}

  1566. 

  1567. 	c->swScale= swScale;

  1568. 	return c;

  1569. }

  1570. 

  1571. /**

  1572.  * returns a normalized gaussian curve used to filter stuff

  1573.  * quality=3 is high quality, lowwer is lowwer quality

  1574.  */

  1575. 

  1576. SwsVector *getGaussianVec(double variance, double quality){

  1577. 	const int length= (int)(variance*quality + 0.5) | 1;

  1578. 	int i;

  1579. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1580. 	double middle= (length-1)*0.5;

  1581. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1582. 

  1583. 	vec->coeff= coeff;

  1584. 	vec->length= length;

  1585. 

  1586. 	for(i=0; i<length; i++)

  1587. 	{

  1588. 		double dist= i-middle;

  1589. 		coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);

  1590. 	}

  1591. 

  1592. 	normalizeVec(vec, 1.0);

  1593. 

  1594. 	return vec;

  1595. }

  1596. 

  1597. SwsVector *getConstVec(double c, int length){

  1598. 	int i;

  1599. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1600. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1601. 

  1602. 	vec->coeff= coeff;

  1603. 	vec->length= length;

  1604. 

  1605. 	for(i=0; i<length; i++)

  1606. 		coeff[i]= c;

  1607. 

  1608. 	return vec;

  1609. }

  1610. 

  1611. 

  1612. SwsVector *getIdentityVec(void){

  1613. 	double *coeff= memalign(sizeof(double), sizeof(double));

  1614. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1615. 	coeff[0]= 1.0;

  1616. 

  1617. 	vec->coeff= coeff;

  1618. 	vec->length= 1;

  1619. 

  1620. 	return vec;

  1621. }

  1622. 

  1623. void normalizeVec(SwsVector *a, double height){

  1624. 	int i;

  1625. 	double sum=0;

  1626. 	double inv;

  1627. 

  1628. 	for(i=0; i<a->length; i++)

  1629. 		sum+= a->coeff[i];

  1630. 

  1631. 	inv= height/sum;

  1632. 

  1633. 	for(i=0; i<a->length; i++)

  1634. 		a->coeff[i]*= height;

  1635. }

  1636. 

  1637. void scaleVec(SwsVector *a, double scalar){

  1638. 	int i;

  1639. 

  1640. 	for(i=0; i<a->length; i++)

  1641. 		a->coeff[i]*= scalar;

  1642. }

  1643. 

  1644. static SwsVector *getConvVec(SwsVector *a, SwsVector *b){

  1645. 	int length= a->length + b->length - 1;

  1646. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1647. 	int i, j;

  1648. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1649. 

  1650. 	vec->coeff= coeff;

  1651. 	vec->length= length;

  1652. 

  1653. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  1654. 

  1655. 	for(i=0; i<a->length; i++)

  1656. 	{

  1657. 		for(j=0; j<b->length; j++)

  1658. 		{

  1659. 			coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1660. 		}

  1661. 	}

  1662. 

  1663. 	return vec;

  1664. }

  1665. 

  1666. static SwsVector *sumVec(SwsVector *a, SwsVector *b){

  1667. 	int length= MAX(a->length, b->length);

  1668. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1669. 	int i;

  1670. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1671. 

  1672. 	vec->coeff= coeff;

  1673. 	vec->length= length;

  1674. 

  1675. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  1676. 

  1677. 	for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1678. 	for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1679. 

  1680. 	return vec;

  1681. }

  1682. 

  1683. static SwsVector *diffVec(SwsVector *a, SwsVector *b){

  1684. 	int length= MAX(a->length, b->length);

  1685. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1686. 	int i;

  1687. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1688. 

  1689. 	vec->coeff= coeff;

  1690. 	vec->length= length;

  1691. 

  1692. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  1693. 

  1694. 	for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1695. 	for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1696. 

  1697. 	return vec;

  1698. }

  1699. 

  1700. /* shift left / or right if "shift" is negative */

  1701. static SwsVector *getShiftedVec(SwsVector *a, int shift){

  1702. 	int length= a->length + ABS(shift)*2;

  1703. 	double *coeff= memalign(sizeof(double), length*sizeof(double));

  1704. 	int i;

  1705. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1706. 

  1707. 	vec->coeff= coeff;

  1708. 	vec->length= length;

  1709. 

  1710. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  1711. 

  1712. 	for(i=0; i<a->length; i++)

  1713. 	{

  1714. 		coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1715. 	}

  1716. 

  1717. 	return vec;

  1718. }

  1719. 

  1720. void shiftVec(SwsVector *a, int shift){

  1721. 	SwsVector *shifted= getShiftedVec(a, shift);

  1722. 	free(a->coeff);

  1723. 	a->coeff= shifted->coeff;

  1724. 	a->length= shifted->length;

  1725. 	free(shifted);

  1726. }

  1727. 

  1728. void addVec(SwsVector *a, SwsVector *b){

  1729. 	SwsVector *sum= sumVec(a, b);

  1730. 	free(a->coeff);

  1731. 	a->coeff= sum->coeff;

  1732. 	a->length= sum->length;

  1733. 	free(sum);

  1734. }

  1735. 

  1736. void subVec(SwsVector *a, SwsVector *b){

  1737. 	SwsVector *diff= diffVec(a, b);

  1738. 	free(a->coeff);

  1739. 	a->coeff= diff->coeff;

  1740. 	a->length= diff->length;

  1741. 	free(diff);

  1742. }

  1743. 

  1744. void convVec(SwsVector *a, SwsVector *b){

  1745. 	SwsVector *conv= getConvVec(a, b);

  1746. 	free(a->coeff);

  1747. 	a->coeff= conv->coeff;

  1748. 	a->length= conv->length;

  1749. 	free(conv);

  1750. }

  1751. 

  1752. SwsVector *cloneVec(SwsVector *a){

  1753. 	double *coeff= memalign(sizeof(double), a->length*sizeof(double));

  1754. 	int i;

  1755. 	SwsVector *vec= malloc(sizeof(SwsVector));

  1756. 

  1757. 	vec->coeff= coeff;

  1758. 	vec->length= a->length;

  1759. 

  1760. 	for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];

  1761. 

  1762. 	return vec;

  1763. }

  1764. 

  1765. void printVec(SwsVector *a){

  1766. 	int i;

  1767. 	double max=0;

  1768. 	double min=0;

  1769. 	double range;

  1770. 

  1771. 	for(i=0; i<a->length; i++)

  1772. 		if(a->coeff[i]>max) max= a->coeff[i];

  1773. 

  1774. 	for(i=0; i<a->length; i++)

  1775. 		if(a->coeff[i]<min) min= a->coeff[i];

  1776. 

  1777. 	range= max - min;

  1778. 

  1779. 	for(i=0; i<a->length; i++)

  1780. 	{

  1781. 		int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1782. 		printf("%1.3f ", a->coeff[i]);

  1783. 		for(;x>0; x--) printf(" ");

  1784. 		printf("|\n");

  1785. 	}

  1786. }

  1787. 

  1788. void freeVec(SwsVector *a){

  1789. 	if(!a) return;

  1790. 	if(a->coeff) free(a->coeff);

  1791. 	a->coeff=NULL;

  1792. 	a->length=0;

  1793. 	free(a);

  1794. }

  1795. 

  1796. void freeSwsContext(SwsContext *c){

  1797. 	int i;

  1798. 

  1799. 	if(!c) return;

  1800. 

  1801. 	if(c->lumPixBuf)

  1802. 	{

  1803. 		for(i=0; i<c->vLumBufSize; i++)

  1804. 		{

  1805. 			if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);

  1806. 			c->lumPixBuf[i]=NULL;

  1807. 		}

  1808. 		free(c->lumPixBuf);

  1809. 		c->lumPixBuf=NULL;

  1810. 	}

  1811. 

  1812. 	if(c->chrPixBuf)

  1813. 	{

  1814. 		for(i=0; i<c->vChrBufSize; i++)

  1815. 		{

  1816. 			if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);

  1817. 			c->chrPixBuf[i]=NULL;

  1818. 		}

  1819. 		free(c->chrPixBuf);

  1820. 		c->chrPixBuf=NULL;

  1821. 	}

  1822. 

  1823. 	if(c->vLumFilter) free(c->vLumFilter);

  1824. 	c->vLumFilter = NULL;

  1825. 	if(c->vChrFilter) free(c->vChrFilter);

  1826. 	c->vChrFilter = NULL;

  1827. 	if(c->hLumFilter) free(c->hLumFilter);

  1828. 	c->hLumFilter = NULL;

  1829. 	if(c->hChrFilter) free(c->hChrFilter);

  1830. 	c->hChrFilter = NULL;

  1831. 

  1832. 	if(c->vLumFilterPos) free(c->vLumFilterPos);

  1833. 	c->vLumFilterPos = NULL;

  1834. 	if(c->vChrFilterPos) free(c->vChrFilterPos);

  1835. 	c->vChrFilterPos = NULL;

  1836. 	if(c->hLumFilterPos) free(c->hLumFilterPos);

  1837. 	c->hLumFilterPos = NULL;

  1838. 	if(c->hChrFilterPos) free(c->hChrFilterPos);

  1839. 	c->hChrFilterPos = NULL;

  1840. 

  1841. 	if(c->lumMmxFilter) free(c->lumMmxFilter);

  1842. 	c->lumMmxFilter = NULL;

  1843. 	if(c->chrMmxFilter) free(c->chrMmxFilter);

  1844. 	c->chrMmxFilter = NULL;

  1845. 

  1846. 	free(c);

  1847. }

  1848.