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.
6220 Commits
32 Branches
303MB
Tree: e81e3ced6d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e81e3ced6d'
${ noResults }
FFmpeg/libswscale/swscale.c

2753 lines
78KB
Raw Blame History 

  1. /*

  2.     Copyright (C) 2001-2003 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., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA

  17. */

  18. 

  19. /*

  20.   supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09

  21.   supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09

  22.   {BGR,RGB}{1,4,8,15,16} support dithering

  23.   

  24.   unscaled special converters (YV12=I420=IYUV, Y800=Y8)

  25.   YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}

  26.   x -> x

  27.   YUV9 -> YV12

  28.   YUV9/YV12 -> Y800

  29.   Y800 -> YUV9/YV12

  30.   BGR24 -> BGR32 & RGB24 -> RGB32

  31.   BGR32 -> BGR24 & RGB32 -> RGB24

  32.   BGR15 -> BGR16

  33. */

  34. 

  35. /* 

  36. tested special converters (most are tested actually but i didnt write it down ...)

  37.  YV12 -> BGR16

  38.  YV12 -> YV12

  39.  BGR15 -> BGR16

  40.  BGR16 -> BGR16

  41.  YVU9 -> YV12

  42. 

  43. untested special converters

  44.   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)

  45.   YV12/I420 -> YV12/I420

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

  47.   BGR24 -> BGR32 & RGB24 -> RGB32

  48.   BGR32 -> BGR24 & RGB32 -> RGB24

  49.   BGR24 -> YV12

  50. */

  51. 

  52. #include <inttypes.h>

  53. #include <string.h>

  54. #include <math.h>

  55. #include <stdio.h>

  56. #include <unistd.h>

  57. #include "config.h"

  58. #include <assert.h>

  59. #ifdef HAVE_MALLOC_H

  60. #include <malloc.h>

  61. #else

  62. #include <stdlib.h>

  63. #endif

  64. #ifdef HAVE_SYS_MMAN_H

  65. #include <sys/mman.h>

  66. #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)

  67. #define MAP_ANONYMOUS MAP_ANON

  68. #endif

  69. #endif

  70. #include "swscale.h"

  71. #include "swscale_internal.h"

  72. #include "x86_cpu.h"

  73. #include "bswap.h"

  74. #include "img_format.h"

  75. #include "rgb2rgb.h"

  76. #ifdef USE_FASTMEMCPY

  77. #include "libvo/fastmemcpy.h"

  78. #endif

  79. 

  80. #undef MOVNTQ

  81. #undef PAVGB

  82. 

  83. //#undef HAVE_MMX2

  84. //#define HAVE_3DNOW

  85. //#undef HAVE_MMX

  86. //#undef ARCH_X86

  87. //#define WORDS_BIGENDIAN

  88. #define DITHER1XBPP

  89. 

  90. #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit

  91. 

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

  93. 

  94. #ifdef MP_DEBUG

  95. #define ASSERT(x) assert(x);

  96. #else

  97. #define ASSERT(x) ;

  98. #endif

  99. 

  100. #ifdef M_PI

  101. #define PI M_PI

  102. #else

  103. #define PI 3.14159265358979323846

  104. #endif

  105. 

  106. //FIXME replace this with something faster

  107. #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \

  108. 			|| (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21 \

  109. 			|| (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)

  110. #define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))

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

  112. #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)

  113. #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)

  114. #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\

  115. 			|| (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\

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

  117. 			|| (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\

  118. 			|| (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)

  119. #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\

  120. 			|| (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\

  121. 			|| isRGB(x) || isBGR(x)\

  122. 			|| (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21\

  123. 			|| (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)

  124. #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))

  125. 

  126. #define RGB2YUV_SHIFT 16

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

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

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

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

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

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

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

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

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

  136. 

  137. extern const int32_t Inverse_Table_6_9[8][4];

  138. 

  139. /*

  140. NOTES

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

  142. 

  143. TODO

  144. more intelligent missalignment avoidance for the horizontal scaler

  145. write special vertical cubic upscale version

  146. Optimize C code (yv12 / minmax)

  147. add support for packed pixel yuv input & output

  148. add support for Y8 output

  149. optimize bgr24 & bgr32

  150. add BGR4 output support

  151. write special BGR->BGR scaler

  152. */

  153. 

  154. #if defined(ARCH_X86) || defined(ARCH_X86_64)

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

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

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

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

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

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

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

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

  163. 

  164. static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;

  165. static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;

  166. static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;

  167. static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;

  168. 

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

  170. 	0x0103010301030103LL,

  171. 	0x0200020002000200LL,};

  172. 

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

  174. 	0x0602060206020602LL,

  175. 	0x0004000400040004LL,};

  176. 

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

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

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

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

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

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

  183. 

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

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

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

  187. 

  188. #ifdef FAST_BGR2YV12

  189. static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;

  190. static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;

  191. static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;

  192. #else

  193. static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;

  194. static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;

  195. static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;

  196. #endif /* FAST_BGR2YV12 */

  197. static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;

  198. static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;

  199. static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;

  200. #endif /* defined(ARCH_X86) || defined(ARCH_X86_64) */

  201. 

  202. // clipping helper table for C implementations:

  203. static unsigned char clip_table[768];

  204. 

  205. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);

  206. 		  

  207. extern const uint8_t dither_2x2_4[2][8];

  208. extern const uint8_t dither_2x2_8[2][8];

  209. extern const uint8_t dither_8x8_32[8][8];

  210. extern const uint8_t dither_8x8_73[8][8];

  211. extern const uint8_t dither_8x8_220[8][8];

  212. 

  213. /* Used for ffmpeg --> MPlayer format name conversion */

  214. static const int fmt_name[PIX_FMT_NB] = {

  215.     [PIX_FMT_YUV420P] = IMGFMT_I420,   ///< Planar YUV 4:2:0 (1 Cr & Cb sample per 2x2 Y samples)

  216.     [PIX_FMT_YUV422] = IMGFMT_Y422,    

  217.     [PIX_FMT_RGB24] = IMGFMT_RGB24,     ///< Packed pixel, 3 bytes per pixel, RGBRGB...

  218.     [PIX_FMT_BGR24] = IMGFMT_BGR24,     ///< Packed pixel, 3 bytes per pixel, BGRBGR...

  219.     [PIX_FMT_YUV422P] = IMGFMT_422P,   ///< Planar YUV 4:2:2 (1 Cr & Cb sample per 2x1 Y samples)

  220.     [PIX_FMT_YUV444P] = IMGFMT_444P,   ///< Planar YUV 4:4:4 (1 Cr & Cb sample per 1x1 Y samples)

  221.     [PIX_FMT_RGBA32] = IMGFMT_RGB32,    ///< Packed pixel, 4 bytes per pixel, BGRABGRA..., stored in cpu endianness

  222.     [PIX_FMT_YUV410P] = IMGFMT_YVU9,   ///< Planar YUV 4:1:0 (1 Cr & Cb sample per 4x4 Y samples)

  223.     [PIX_FMT_YUV411P] = IMGFMT_411P,   ///< Planar YUV 4:1:1 (1 Cr & Cb sample per 4x1 Y samples)

  224.     [PIX_FMT_RGB565] = IMGFMT_RGB16,    ///< always stored in cpu endianness 

  225.     [PIX_FMT_RGB555] = IMGFMT_RGB15,    ///< always stored in cpu endianness, most significant bit to 1 

  226.     [PIX_FMT_UYVY422] = IMGFMT_UYVY,   ///< Packed pixel, Cb Y0 Cr Y1 

  227. };

  228. 

  229. char *sws_format_name(int format)

  230. {

  231.     static char fmt_name[64];

  232.     char *res;

  233.     static int buffer;

  234. 

  235.     res = fmt_name + buffer * 32;

  236.     buffer = 1 - buffer;

  237.     snprintf(res, 32, "0x%x (%c%c%c%c)", format,

  238. 		    format >> 24, (format >> 16) & 0xFF,

  239. 		    (format >> 8) & 0xFF,

  240. 		    format & 0xFF);

  241. 

  242.     return res;

  243. }

  244. 

  245. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  246. void in_asm_used_var_warning_killer()

  247. {

  248.  volatile int i= bF8+bFC+w10+

  249.  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+

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

  251.  if(i) i=0;

  252. }

  253. #endif

  254. 

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

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

  257. 				    uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)

  258. {

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

  260. 	int i;

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

  262. 	{

  263. 		int val=1<<18;

  264. 		int j;

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

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

  267. 

  268. 		dest[i]= FFMIN(FFMAX(val>>19, 0), 255);

  269. 	}

  270. 

  271. 	if(uDest != NULL)

  272. 		for(i=0; i<chrDstW; i++)

  273. 		{

  274. 			int u=1<<18;

  275. 			int v=1<<18;

  276. 			int j;

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

  278. 			{

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

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

  281. 			}

  282. 

  283. 			uDest[i]= FFMIN(FFMAX(u>>19, 0), 255);

  284. 			vDest[i]= FFMIN(FFMAX(v>>19, 0), 255);

  285. 		}

  286. }

  287. 

  288. static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,

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

  290. 				uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)

  291. {

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

  293. 	int i;

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

  295. 	{

  296. 		int val=1<<18;

  297. 		int j;

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

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

  300. 

  301. 		dest[i]= FFMIN(FFMAX(val>>19, 0), 255);

  302. 	}

  303. 

  304. 	if(uDest == NULL)

  305. 		return;

  306. 

  307. 	if(dstFormat == IMGFMT_NV12)

  308. 		for(i=0; i<chrDstW; i++)

  309. 		{

  310. 			int u=1<<18;

  311. 			int v=1<<18;

  312. 			int j;

  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. 

  319. 			uDest[2*i]= FFMIN(FFMAX(u>>19, 0), 255);

  320. 			uDest[2*i+1]= FFMIN(FFMAX(v>>19, 0), 255);

  321. 		}

  322. 	else

  323. 		for(i=0; i<chrDstW; i++)

  324. 		{

  325. 			int u=1<<18;

  326. 			int v=1<<18;

  327. 			int j;

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

  329. 			{

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

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

  332. 			}

  333. 

  334. 			uDest[2*i]= FFMIN(FFMAX(v>>19, 0), 255);

  335. 			uDest[2*i+1]= FFMIN(FFMAX(u>>19, 0), 255);

  336. 		}

  337. }

  338. 

  339. #define YSCALE_YUV_2_PACKEDX_C(type) \

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

  341. 			int j;\

  342. 			int Y1=1<<18;\

  343. 			int Y2=1<<18;\

  344. 			int U=1<<18;\

  345. 			int V=1<<18;\

  346. 			type *r, *b, *g;\

  347. 			const int i2= 2*i;\

  348. 			\

  349. 			for(j=0; j<lumFilterSize; j++)\

  350. 			{\

  351. 				Y1 += lumSrc[j][i2] * lumFilter[j];\

  352. 				Y2 += lumSrc[j][i2+1] * lumFilter[j];\

  353. 			}\

  354. 			for(j=0; j<chrFilterSize; j++)\

  355. 			{\

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

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

  358. 			}\

  359. 			Y1>>=19;\

  360. 			Y2>>=19;\

  361. 			U >>=19;\

  362. 			V >>=19;\

  363. 			if((Y1|Y2|U|V)&256)\

  364. 			{\

  365. 				if(Y1>255)   Y1=255;\

  366. 				else if(Y1<0)Y1=0;\

  367. 				if(Y2>255)   Y2=255;\

  368. 				else if(Y2<0)Y2=0;\

  369. 				if(U>255)    U=255;\

  370. 				else if(U<0) U=0;\

  371. 				if(V>255)    V=255;\

  372. 				else if(V<0) V=0;\

  373. 			}

  374.                         

  375. #define YSCALE_YUV_2_RGBX_C(type) \

  376. 			YSCALE_YUV_2_PACKEDX_C(type)\

  377. 			r = c->table_rV[V];\

  378. 			g = c->table_gU[U] + c->table_gV[V];\

  379. 			b = c->table_bU[U];\

  380. 

  381. #define YSCALE_YUV_2_PACKED2_C \

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

  383. 			const int i2= 2*i;\

  384. 			int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\

  385. 			int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\

  386. 			int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\

  387. 			int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\

  388. 

  389. #define YSCALE_YUV_2_RGB2_C(type) \

  390. 			YSCALE_YUV_2_PACKED2_C\

  391. 			type *r, *b, *g;\

  392. 			r = c->table_rV[V];\

  393. 			g = c->table_gU[U] + c->table_gV[V];\

  394. 			b = c->table_bU[U];\

  395. 

  396. #define YSCALE_YUV_2_PACKED1_C \

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

  398. 			const int i2= 2*i;\

  399. 			int Y1= buf0[i2  ]>>7;\

  400. 			int Y2= buf0[i2+1]>>7;\

  401. 			int U= (uvbuf1[i     ])>>7;\

  402. 			int V= (uvbuf1[i+2048])>>7;\

  403. 

  404. #define YSCALE_YUV_2_RGB1_C(type) \

  405. 			YSCALE_YUV_2_PACKED1_C\

  406. 			type *r, *b, *g;\

  407. 			r = c->table_rV[V];\

  408. 			g = c->table_gU[U] + c->table_gV[V];\

  409. 			b = c->table_bU[U];\

  410. 

  411. #define YSCALE_YUV_2_PACKED1B_C \

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

  413. 			const int i2= 2*i;\

  414. 			int Y1= buf0[i2  ]>>7;\

  415. 			int Y2= buf0[i2+1]>>7;\

  416. 			int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\

  417. 			int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\

  418. 

  419. #define YSCALE_YUV_2_RGB1B_C(type) \

  420. 			YSCALE_YUV_2_PACKED1B_C\

  421. 			type *r, *b, *g;\

  422. 			r = c->table_rV[V];\

  423. 			g = c->table_gU[U] + c->table_gV[V];\

  424. 			b = c->table_bU[U];\

  425. 

  426. #define YSCALE_YUV_2_ANYRGB_C(func, func2)\

  427. 	switch(c->dstFormat)\

  428. 	{\

  429. 	case IMGFMT_BGR32:\

  430. 	case IMGFMT_RGB32:\

  431. 		func(uint32_t)\

  432. 			((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\

  433. 			((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\

  434. 		}		\

  435. 		break;\

  436. 	case IMGFMT_RGB24:\

  437. 		func(uint8_t)\

  438. 			((uint8_t*)dest)[0]= r[Y1];\

  439. 			((uint8_t*)dest)[1]= g[Y1];\

  440. 			((uint8_t*)dest)[2]= b[Y1];\

  441. 			((uint8_t*)dest)[3]= r[Y2];\

  442. 			((uint8_t*)dest)[4]= g[Y2];\

  443. 			((uint8_t*)dest)[5]= b[Y2];\

  444. 			dest+=6;\

  445. 		}\

  446. 		break;\

  447. 	case IMGFMT_BGR24:\

  448. 		func(uint8_t)\

  449. 			((uint8_t*)dest)[0]= b[Y1];\

  450. 			((uint8_t*)dest)[1]= g[Y1];\

  451. 			((uint8_t*)dest)[2]= r[Y1];\

  452. 			((uint8_t*)dest)[3]= b[Y2];\

  453. 			((uint8_t*)dest)[4]= g[Y2];\

  454. 			((uint8_t*)dest)[5]= r[Y2];\

  455. 			dest+=6;\

  456. 		}\

  457. 		break;\

  458. 	case IMGFMT_RGB16:\

  459. 	case IMGFMT_BGR16:\

  460. 		{\

  461. 			const int dr1= dither_2x2_8[y&1    ][0];\

  462. 			const int dg1= dither_2x2_4[y&1    ][0];\

  463. 			const int db1= dither_2x2_8[(y&1)^1][0];\

  464. 			const int dr2= dither_2x2_8[y&1    ][1];\

  465. 			const int dg2= dither_2x2_4[y&1    ][1];\

  466. 			const int db2= dither_2x2_8[(y&1)^1][1];\

  467. 			func(uint16_t)\

  468. 				((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\

  469. 				((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\

  470. 			}\

  471. 		}\

  472. 		break;\

  473. 	case IMGFMT_RGB15:\

  474. 	case IMGFMT_BGR15:\

  475. 		{\

  476. 			const int dr1= dither_2x2_8[y&1    ][0];\

  477. 			const int dg1= dither_2x2_8[y&1    ][1];\

  478. 			const int db1= dither_2x2_8[(y&1)^1][0];\

  479. 			const int dr2= dither_2x2_8[y&1    ][1];\

  480. 			const int dg2= dither_2x2_8[y&1    ][0];\

  481. 			const int db2= dither_2x2_8[(y&1)^1][1];\

  482. 			func(uint16_t)\

  483. 				((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\

  484. 				((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\

  485. 			}\

  486. 		}\

  487. 		break;\

  488. 	case IMGFMT_RGB8:\

  489. 	case IMGFMT_BGR8:\

  490. 		{\

  491. 			const uint8_t * const d64= dither_8x8_73[y&7];\

  492. 			const uint8_t * const d32= dither_8x8_32[y&7];\

  493. 			func(uint8_t)\

  494. 				((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\

  495. 				((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\

  496. 			}\

  497. 		}\

  498. 		break;\

  499. 	case IMGFMT_RGB4:\

  500. 	case IMGFMT_BGR4:\

  501. 		{\

  502. 			const uint8_t * const d64= dither_8x8_73 [y&7];\

  503. 			const uint8_t * const d128=dither_8x8_220[y&7];\

  504. 			func(uint8_t)\

  505. 				((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\

  506. 				                 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\

  507. 			}\

  508. 		}\

  509. 		break;\

  510. 	case IMGFMT_RG4B:\

  511. 	case IMGFMT_BG4B:\

  512. 		{\

  513. 			const uint8_t * const d64= dither_8x8_73 [y&7];\

  514. 			const uint8_t * const d128=dither_8x8_220[y&7];\

  515. 			func(uint8_t)\

  516. 				((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\

  517. 				((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\

  518. 			}\

  519. 		}\

  520. 		break;\

  521. 	case IMGFMT_RGB1:\

  522. 	case IMGFMT_BGR1:\

  523. 		{\

  524. 			const uint8_t * const d128=dither_8x8_220[y&7];\

  525. 			uint8_t *g= c->table_gU[128] + c->table_gV[128];\

  526. 			for(i=0; i<dstW-7; i+=8){\

  527. 				int acc;\

  528. 				acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\

  529. 				acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\

  530. 				acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\

  531. 				acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\

  532. 				acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\

  533. 				acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\

  534. 				acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\

  535. 				acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\

  536. 				((uint8_t*)dest)[0]= acc;\

  537. 				dest++;\

  538. 			}\

  539. \

  540. /*\

  541. ((uint8_t*)dest)-= dstW>>4;\

  542. {\

  543. 			int acc=0;\

  544. 			int left=0;\

  545. 			static int top[1024];\

  546. 			static int last_new[1024][1024];\

  547. 			static int last_in3[1024][1024];\

  548. 			static int drift[1024][1024];\

  549. 			int topLeft=0;\

  550. 			int shift=0;\

  551. 			int count=0;\

  552. 			const uint8_t * const d128=dither_8x8_220[y&7];\

  553. 			int error_new=0;\

  554. 			int error_in3=0;\

  555. 			int f=0;\

  556. 			\

  557. 			for(i=dstW>>1; i<dstW; i++){\

  558. 				int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\

  559. 				int in2 = (76309 * (in - 16) + 32768) >> 16;\

  560. 				int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\

  561. 				int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\

  562. 					+ (last_new[y][i] - in3)*f/256;\

  563. 				int new= old> 128 ? 255 : 0;\

  564. \

  565. 				error_new+= ABS(last_new[y][i] - new);\

  566. 				error_in3+= ABS(last_in3[y][i] - in3);\

  567. 				f= error_new - error_in3*4;\

  568. 				if(f<0) f=0;\

  569. 				if(f>256) f=256;\

  570. \

  571. 				topLeft= top[i];\

  572. 				left= top[i]= old - new;\

  573. 				last_new[y][i]= new;\

  574. 				last_in3[y][i]= in3;\

  575. \

  576. 				acc+= acc + (new&1);\

  577. 				if((i&7)==6){\

  578. 					((uint8_t*)dest)[0]= acc;\

  579. 					((uint8_t*)dest)++;\

  580. 				}\

  581. 			}\

  582. }\

  583. */\

  584. 		}\

  585. 		break;\

  586. 	case IMGFMT_YUY2:\

  587. 		func2\

  588. 			((uint8_t*)dest)[2*i2+0]= Y1;\

  589. 			((uint8_t*)dest)[2*i2+1]= U;\

  590. 			((uint8_t*)dest)[2*i2+2]= Y2;\

  591. 			((uint8_t*)dest)[2*i2+3]= V;\

  592. 		}		\

  593. 		break;\

  594. 	case IMGFMT_UYVY:\

  595. 		func2\

  596. 			((uint8_t*)dest)[2*i2+0]= U;\

  597. 			((uint8_t*)dest)[2*i2+1]= Y1;\

  598. 			((uint8_t*)dest)[2*i2+2]= V;\

  599. 			((uint8_t*)dest)[2*i2+3]= Y2;\

  600. 		}		\

  601. 		break;\

  602. 	}\

  603. 

  604. 

  605. static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,

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

  607. 				    uint8_t *dest, int dstW, int y)

  608. {

  609. 	int i;

  610. 	switch(c->dstFormat)

  611. 	{

  612. 	case IMGFMT_RGB32:

  613. 	case IMGFMT_BGR32:

  614. 		YSCALE_YUV_2_RGBX_C(uint32_t)

  615. 			((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];

  616. 			((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];

  617. 		}

  618. 		break;

  619. 	case IMGFMT_RGB24:

  620. 		YSCALE_YUV_2_RGBX_C(uint8_t)

  621. 			((uint8_t*)dest)[0]= r[Y1];

  622. 			((uint8_t*)dest)[1]= g[Y1];

  623. 			((uint8_t*)dest)[2]= b[Y1];

  624. 			((uint8_t*)dest)[3]= r[Y2];

  625. 			((uint8_t*)dest)[4]= g[Y2];

  626. 			((uint8_t*)dest)[5]= b[Y2];

  627. 			dest+=6;

  628. 		}

  629. 		break;

  630. 	case IMGFMT_BGR24:

  631. 		YSCALE_YUV_2_RGBX_C(uint8_t)

  632. 			((uint8_t*)dest)[0]= b[Y1];

  633. 			((uint8_t*)dest)[1]= g[Y1];

  634. 			((uint8_t*)dest)[2]= r[Y1];

  635. 			((uint8_t*)dest)[3]= b[Y2];

  636. 			((uint8_t*)dest)[4]= g[Y2];

  637. 			((uint8_t*)dest)[5]= r[Y2];

  638. 			dest+=6;

  639. 		}

  640. 		break;

  641. 	case IMGFMT_RGB16:

  642. 	case IMGFMT_BGR16:

  643. 		{

  644. 			const int dr1= dither_2x2_8[y&1    ][0];

  645. 			const int dg1= dither_2x2_4[y&1    ][0];

  646. 			const int db1= dither_2x2_8[(y&1)^1][0];

  647. 			const int dr2= dither_2x2_8[y&1    ][1];

  648. 			const int dg2= dither_2x2_4[y&1    ][1];

  649. 			const int db2= dither_2x2_8[(y&1)^1][1];

  650. 			YSCALE_YUV_2_RGBX_C(uint16_t)

  651. 				((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];

  652. 				((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];

  653. 			}

  654. 		}

  655. 		break;

  656. 	case IMGFMT_RGB15:

  657. 	case IMGFMT_BGR15:

  658. 		{

  659. 			const int dr1= dither_2x2_8[y&1    ][0];

  660. 			const int dg1= dither_2x2_8[y&1    ][1];

  661. 			const int db1= dither_2x2_8[(y&1)^1][0];

  662. 			const int dr2= dither_2x2_8[y&1    ][1];

  663. 			const int dg2= dither_2x2_8[y&1    ][0];

  664. 			const int db2= dither_2x2_8[(y&1)^1][1];

  665. 			YSCALE_YUV_2_RGBX_C(uint16_t)

  666. 				((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];

  667. 				((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];

  668. 			}

  669. 		}

  670. 		break;

  671. 	case IMGFMT_RGB8:

  672. 	case IMGFMT_BGR8:

  673. 		{

  674. 			const uint8_t * const d64= dither_8x8_73[y&7];

  675. 			const uint8_t * const d32= dither_8x8_32[y&7];

  676. 			YSCALE_YUV_2_RGBX_C(uint8_t)

  677. 				((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];

  678. 				((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];

  679. 			}

  680. 		}

  681. 		break;

  682. 	case IMGFMT_RGB4:

  683. 	case IMGFMT_BGR4:

  684. 		{

  685. 			const uint8_t * const d64= dither_8x8_73 [y&7];

  686. 			const uint8_t * const d128=dither_8x8_220[y&7];

  687. 			YSCALE_YUV_2_RGBX_C(uint8_t)

  688. 				((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]

  689. 				                  +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);

  690. 			}

  691. 		}

  692. 		break;

  693. 	case IMGFMT_RG4B:

  694. 	case IMGFMT_BG4B:

  695. 		{

  696. 			const uint8_t * const d64= dither_8x8_73 [y&7];

  697. 			const uint8_t * const d128=dither_8x8_220[y&7];

  698. 			YSCALE_YUV_2_RGBX_C(uint8_t)

  699. 				((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];

  700. 				((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];

  701. 			}

  702. 		}

  703. 		break;

  704. 	case IMGFMT_RGB1:

  705. 	case IMGFMT_BGR1:

  706. 		{

  707. 			const uint8_t * const d128=dither_8x8_220[y&7];

  708. 			uint8_t *g= c->table_gU[128] + c->table_gV[128];

  709. 			int acc=0;

  710. 			for(i=0; i<dstW-1; i+=2){

  711. 				int j;

  712. 				int Y1=1<<18;

  713. 				int Y2=1<<18;

  714. 

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

  716. 				{

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

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

  719. 				}

  720. 				Y1>>=19;

  721. 				Y2>>=19;

  722. 				if((Y1|Y2)&256)

  723. 				{

  724. 					if(Y1>255)   Y1=255;

  725. 					else if(Y1<0)Y1=0;

  726. 					if(Y2>255)   Y2=255;

  727. 					else if(Y2<0)Y2=0;

  728. 				}

  729. 				acc+= acc + g[Y1+d128[(i+0)&7]];

  730. 				acc+= acc + g[Y2+d128[(i+1)&7]];

  731. 				if((i&7)==6){

  732. 					((uint8_t*)dest)[0]= acc;

  733. 					dest++;

  734. 				}

  735. 			}

  736. 		}

  737. 		break;

  738. 	case IMGFMT_YUY2:

  739. 		YSCALE_YUV_2_PACKEDX_C(void)

  740. 			((uint8_t*)dest)[2*i2+0]= Y1;

  741. 			((uint8_t*)dest)[2*i2+1]= U;

  742. 			((uint8_t*)dest)[2*i2+2]= Y2;

  743. 			((uint8_t*)dest)[2*i2+3]= V;

  744. 		}

  745.                 break;

  746. 	case IMGFMT_UYVY:

  747. 		YSCALE_YUV_2_PACKEDX_C(void)

  748. 			((uint8_t*)dest)[2*i2+0]= U;

  749. 			((uint8_t*)dest)[2*i2+1]= Y1;

  750. 			((uint8_t*)dest)[2*i2+2]= V;

  751. 			((uint8_t*)dest)[2*i2+3]= Y2;

  752. 		}

  753.                 break;

  754. 	}

  755. }

  756. 

  757. 

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

  759. //Plain C versions

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

  761. #define COMPILE_C

  762. #endif

  763. 

  764. #ifdef ARCH_POWERPC

  765. #if defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)

  766. #define COMPILE_ALTIVEC

  767. #endif //HAVE_ALTIVEC

  768. #endif //ARCH_POWERPC

  769. 

  770. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  771. 

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

  773. #define COMPILE_MMX

  774. #endif

  775. 

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

  777. #define COMPILE_MMX2

  778. #endif

  779. 

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

  781. #define COMPILE_3DNOW

  782. #endif

  783. #endif //ARCH_X86 || ARCH_X86_64

  784. 

  785. #undef HAVE_MMX

  786. #undef HAVE_MMX2

  787. #undef HAVE_3DNOW

  788. 

  789. #ifdef COMPILE_C

  790. #undef HAVE_MMX

  791. #undef HAVE_MMX2

  792. #undef HAVE_3DNOW

  793. #undef HAVE_ALTIVEC

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

  795. #include "swscale_template.c"

  796. #endif

  797. 

  798. #ifdef ARCH_POWERPC

  799. #ifdef COMPILE_ALTIVEC

  800. #undef RENAME

  801. #define HAVE_ALTIVEC

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

  803. #include "swscale_template.c"

  804. #endif

  805. #endif //ARCH_POWERPC

  806. 

  807. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  808. 

  809. //X86 versions

  810. /*

  811. #undef RENAME

  812. #undef HAVE_MMX

  813. #undef HAVE_MMX2

  814. #undef HAVE_3DNOW

  815. #define ARCH_X86

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

  817. #include "swscale_template.c"

  818. */

  819. //MMX versions

  820. #ifdef COMPILE_MMX

  821. #undef RENAME

  822. #define HAVE_MMX

  823. #undef HAVE_MMX2

  824. #undef HAVE_3DNOW

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

  826. #include "swscale_template.c"

  827. #endif

  828. 

  829. //MMX2 versions

  830. #ifdef COMPILE_MMX2

  831. #undef RENAME

  832. #define HAVE_MMX

  833. #define HAVE_MMX2

  834. #undef HAVE_3DNOW

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

  836. #include "swscale_template.c"

  837. #endif

  838. 

  839. //3DNOW versions

  840. #ifdef COMPILE_3DNOW

  841. #undef RENAME

  842. #define HAVE_MMX

  843. #undef HAVE_MMX2

  844. #define HAVE_3DNOW

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

  846. #include "swscale_template.c"

  847. #endif

  848. 

  849. #endif //ARCH_X86 || ARCH_X86_64

  850. 

  851. // minor note: the HAVE_xyz is messed up after that line so don't use it

  852. 

  853. static double getSplineCoeff(double a, double b, double c, double d, double dist)

  854. {

  855. //	printf("%f %f %f %f %f\n", a,b,c,d,dist);

  856. 	if(dist<=1.0) 	return ((d*dist + c)*dist + b)*dist +a;

  857. 	else		return getSplineCoeff(	0.0, 

  858. 						 b+ 2.0*c + 3.0*d,

  859. 						        c + 3.0*d,

  860. 						-b- 3.0*c - 6.0*d,

  861. 						dist-1.0);

  862. }

  863. 

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

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

  866. 			      SwsVector *srcFilter, SwsVector *dstFilter, double param[2])

  867. {

  868. 	int i;

  869. 	int filterSize;

  870. 	int filter2Size;

  871. 	int minFilterSize;

  872. 	double *filter=NULL;

  873. 	double *filter2=NULL;

  874. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  875. 	if(flags & SWS_CPU_CAPS_MMX)

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

  877. #endif

  878. 

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

  880. 	*filterPos = av_malloc((dstW+1)*sizeof(int16_t));

  881. 

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

  883. 	{

  884. 		int i;

  885. 		filterSize= 1;

  886. 		filter= av_malloc(dstW*sizeof(double)*filterSize);

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

  888. 

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

  890. 		{

  891. 			filter[i*filterSize]=1;

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

  893. 		}

  894. 

  895. 	}

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

  897. 	{

  898. 		int i;

  899. 		int xDstInSrc;

  900. 		filterSize= 1;

  901. 		filter= av_malloc(dstW*sizeof(double)*filterSize);

  902. 		

  903. 		xDstInSrc= xInc/2 - 0x8000;

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

  905. 		{

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

  907. 

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

  909. 			filter[i]= 1.0;

  910. 			xDstInSrc+= xInc;

  911. 		}

  912. 	}

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

  914. 	{

  915. 		int i;

  916. 		int xDstInSrc;

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

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

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

  920. 		filter= av_malloc(dstW*sizeof(double)*filterSize);

  921. 

  922. 		xDstInSrc= xInc/2 - 0x8000;

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

  924. 		{

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

  926. 			int j;

  927. 

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

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

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

  931. 				{

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

  933. 					double coeff= 1.0 - d;

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

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

  936. 					xx++;

  937. 				}

  938. 			xDstInSrc+= xInc;

  939. 		}

  940. 	}

  941. 	else

  942. 	{

  943. 		double xDstInSrc;

  944. 		double sizeFactor, filterSizeInSrc;

  945. 		const double xInc1= (double)xInc / (double)(1<<16);

  946. 

  947. 		if     (flags&SWS_BICUBIC)	sizeFactor= 4.0;

  948. 		else if(flags&SWS_X)		sizeFactor= 8.0;

  949. 		else if(flags&SWS_AREA)		sizeFactor= 1.0; //downscale only, for upscale it is bilinear

  950. 		else if(flags&SWS_GAUSS)	sizeFactor= 8.0;   // infinite ;)

  951. 		else if(flags&SWS_LANCZOS)	sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;

  952. 		else if(flags&SWS_SINC)		sizeFactor= 20.0; // infinite ;)

  953. 		else if(flags&SWS_SPLINE)	sizeFactor= 20.0;  // infinite ;)

  954. 		else if(flags&SWS_BILINEAR)	sizeFactor= 2.0;

  955. 		else {

  956. 			sizeFactor= 0.0; //GCC warning killer

  957. 			ASSERT(0)

  958. 		}

  959. 		

  960. 		if(xInc1 <= 1.0)	filterSizeInSrc= sizeFactor; // upscale

  961. 		else			filterSizeInSrc= sizeFactor*srcW / (double)dstW;

  962. 

  963. 		filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible

  964. 		if(filterSize > srcW-2) filterSize=srcW-2;

  965. 

  966. 		filter= av_malloc(dstW*sizeof(double)*filterSize);

  967. 

  968. 		xDstInSrc= xInc1 / 2.0 - 0.5;

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

  970. 		{

  971. 			int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);

  972. 			int j;

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

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

  975. 			{

  976. 				double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;

  977. 				double coeff;

  978. 				if(flags & SWS_BICUBIC)

  979. 				{

  980. 					double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;

  981. 					double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;

  982. 

  983. 					if(d<1.0) 

  984. 						coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;

  985. 					else if(d<2.0)

  986. 						coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;

  987. 					else

  988. 						coeff=0.0;

  989. 				}

  990. /*				else if(flags & SWS_X)

  991. 				{

  992. 					double p= param ? param*0.01 : 0.3;

  993. 					coeff = d ? sin(d*PI)/(d*PI) : 1.0;

  994. 					coeff*= pow(2.0, - p*d*d);

  995. 				}*/

  996. 				else if(flags & SWS_X)

  997. 				{

  998. 					double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;

  999. 					

  1000. 					if(d<1.0)

  1001. 						coeff = cos(d*PI);

  1002. 					else

  1003. 						coeff=-1.0;

  1004. 					if(coeff<0.0) 	coeff= -pow(-coeff, A);

  1005. 					else		coeff=  pow( coeff, A);

  1006. 					coeff= coeff*0.5 + 0.5;

  1007. 				}

  1008. 				else if(flags & SWS_AREA)

  1009. 				{

  1010. 					double srcPixelSize= 1.0/xInc1;

  1011. 					if(d + srcPixelSize/2 < 0.5) coeff= 1.0;

  1012. 					else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;

  1013. 					else coeff=0.0;

  1014. 				}

  1015. 				else if(flags & SWS_GAUSS)

  1016. 				{

  1017. 					double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;

  1018. 					coeff = pow(2.0, - p*d*d);

  1019. 				}

  1020. 				else if(flags & SWS_SINC)

  1021. 				{

  1022. 					coeff = d ? sin(d*PI)/(d*PI) : 1.0;

  1023. 				}

  1024. 				else if(flags & SWS_LANCZOS)

  1025. 				{

  1026. 					double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 

  1027. 					coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;

  1028. 					if(d>p) coeff=0;

  1029. 				}

  1030. 				else if(flags & SWS_BILINEAR)

  1031. 				{

  1032. 					coeff= 1.0 - d;

  1033. 					if(coeff<0) coeff=0;

  1034. 				}

  1035. 				else if(flags & SWS_SPLINE)

  1036. 				{

  1037. 					double p=-2.196152422706632;

  1038. 					coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);

  1039. 				}

  1040. 				else {

  1041. 					coeff= 0.0; //GCC warning killer

  1042. 					ASSERT(0)

  1043. 				}

  1044. 

  1045. 				filter[i*filterSize + j]= coeff;

  1046. 				xx++;

  1047. 			}

  1048. 			xDstInSrc+= xInc1;

  1049. 		}

  1050. 	}

  1051. 

  1052. 	/* apply src & dst Filter to filter -> filter2

  1053. 	   av_free(filter);

  1054. 	*/

  1055. 	ASSERT(filterSize>0)

  1056. 	filter2Size= filterSize;

  1057. 	if(srcFilter) filter2Size+= srcFilter->length - 1;

  1058. 	if(dstFilter) filter2Size+= dstFilter->length - 1;

  1059. 	ASSERT(filter2Size>0)

  1060. 	filter2= av_malloc(filter2Size*dstW*sizeof(double));

  1061. 

  1062. 	for(i=0; i<dstW; i++)

  1063. 	{

  1064. 		int j;

  1065. 		SwsVector scaleFilter;

  1066. 		SwsVector *outVec;

  1067. 

  1068. 		scaleFilter.coeff= filter + i*filterSize;

  1069. 		scaleFilter.length= filterSize;

  1070. 

  1071. 		if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);

  1072. 		else	      outVec= &scaleFilter;

  1073. 

  1074. 		ASSERT(outVec->length == filter2Size)

  1075. 		//FIXME dstFilter

  1076. 

  1077. 		for(j=0; j<outVec->length; j++)

  1078. 		{

  1079. 			filter2[i*filter2Size + j]= outVec->coeff[j];

  1080. 		}

  1081. 

  1082. 		(*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;

  1083. 

  1084. 		if(outVec != &scaleFilter) sws_freeVec(outVec);

  1085. 	}

  1086. 	av_free(filter); filter=NULL;

  1087. 

  1088. 	/* try to reduce the filter-size (step1 find size and shift left) */

  1089. 	// Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)

  1090. 	minFilterSize= 0;

  1091. 	for(i=dstW-1; i>=0; i--)

  1092. 	{

  1093. 		int min= filter2Size;

  1094. 		int j;

  1095. 		double cutOff=0.0;

  1096. 

  1097. 		/* get rid off near zero elements on the left by shifting left */

  1098. 		for(j=0; j<filter2Size; j++)

  1099. 		{

  1100. 			int k;

  1101. 			cutOff += ABS(filter2[i*filter2Size]);

  1102. 

  1103. 			if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;

  1104. 

  1105. 			/* preserve Monotonicity because the core can't handle the filter otherwise */

  1106. 			if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;

  1107. 

  1108. 			// Move filter coeffs left

  1109. 			for(k=1; k<filter2Size; k++)

  1110. 				filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];

  1111. 			filter2[i*filter2Size + k - 1]= 0.0;

  1112. 			(*filterPos)[i]++;

  1113. 		}

  1114. 

  1115. 		cutOff=0.0;

  1116. 		/* count near zeros on the right */

  1117. 		for(j=filter2Size-1; j>0; j--)

  1118. 		{

  1119. 			cutOff += ABS(filter2[i*filter2Size + j]);

  1120. 

  1121. 			if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;

  1122. 			min--;

  1123. 		}

  1124. 

  1125. 		if(min>minFilterSize) minFilterSize= min;

  1126. 	}

  1127. 

  1128.         if (flags & SWS_CPU_CAPS_ALTIVEC) {

  1129.           // we can handle the special case 4,

  1130.           // so we don't want to go to the full 8

  1131.           if (minFilterSize < 5)

  1132.             filterAlign = 4;

  1133. 

  1134.           // we really don't want to waste our time

  1135.           // doing useless computation, so fall-back on

  1136.           // the scalar C code for very small filter.

  1137.           // vectorizing is worth it only if you have

  1138.           // decent-sized vector.

  1139.           if (minFilterSize < 3)

  1140.             filterAlign = 1;

  1141.         }

  1142. 

  1143.         if (flags & SWS_CPU_CAPS_MMX) {

  1144.                 // special case for unscaled vertical filtering

  1145.                 if(minFilterSize == 1 && filterAlign == 2)

  1146.                         filterAlign= 1;

  1147.         }

  1148. 

  1149. 	ASSERT(minFilterSize > 0)

  1150. 	filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));

  1151. 	ASSERT(filterSize > 0)

  1152. 	filter= av_malloc(filterSize*dstW*sizeof(double));

  1153.         if(filterSize >= MAX_FILTER_SIZE)

  1154.                 return -1;

  1155. 	*outFilterSize= filterSize;

  1156. 

  1157. 	if(flags&SWS_PRINT_INFO)

  1158. 		MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);

  1159. 	/* try to reduce the filter-size (step2 reduce it) */

  1160. 	for(i=0; i<dstW; i++)

  1161. 	{

  1162. 		int j;

  1163. 

  1164. 		for(j=0; j<filterSize; j++)

  1165. 		{

  1166. 			if(j>=filter2Size) filter[i*filterSize + j]= 0.0;

  1167. 			else		   filter[i*filterSize + j]= filter2[i*filter2Size + j];

  1168. 		}

  1169. 	}

  1170. 	av_free(filter2); filter2=NULL;

  1171. 	

  1172. 

  1173. 	//FIXME try to align filterpos if possible

  1174. 

  1175. 	//fix borders

  1176. 	for(i=0; i<dstW; i++)

  1177. 	{

  1178. 		int j;

  1179. 		if((*filterPos)[i] < 0)

  1180. 		{

  1181. 			// Move filter coeffs left to compensate for filterPos

  1182. 			for(j=1; j<filterSize; j++)

  1183. 			{

  1184. 				int left= FFMAX(j + (*filterPos)[i], 0);

  1185. 				filter[i*filterSize + left] += filter[i*filterSize + j];

  1186. 				filter[i*filterSize + j]=0;

  1187. 			}

  1188. 			(*filterPos)[i]= 0;

  1189. 		}

  1190. 

  1191. 		if((*filterPos)[i] + filterSize > srcW)

  1192. 		{

  1193. 			int shift= (*filterPos)[i] + filterSize - srcW;

  1194. 			// Move filter coeffs right to compensate for filterPos

  1195. 			for(j=filterSize-2; j>=0; j--)

  1196. 			{

  1197. 				int right= FFMIN(j + shift, filterSize-1);

  1198. 				filter[i*filterSize +right] += filter[i*filterSize +j];

  1199. 				filter[i*filterSize +j]=0;

  1200. 			}

  1201. 			(*filterPos)[i]= srcW - filterSize;

  1202. 		}

  1203. 	}

  1204. 

  1205. 	// Note the +1 is for the MMXscaler which reads over the end

  1206. 	/* align at 16 for AltiVec (needed by hScale_altivec_real) */

  1207. 	*outFilter= av_malloc(*outFilterSize*(dstW+1)*sizeof(int16_t));

  1208. 	memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));

  1209. 

  1210. 	/* Normalize & Store in outFilter */

  1211. 	for(i=0; i<dstW; i++)

  1212. 	{

  1213. 		int j;

  1214. 		double error=0;

  1215. 		double sum=0;

  1216. 		double scale= one;

  1217. 

  1218. 		for(j=0; j<filterSize; j++)

  1219. 		{

  1220. 			sum+= filter[i*filterSize + j];

  1221. 		}

  1222. 		scale/= sum;

  1223. 		for(j=0; j<*outFilterSize; j++)

  1224. 		{

  1225. 			double v= filter[i*filterSize + j]*scale + error;

  1226. 			int intV= floor(v + 0.5);

  1227. 			(*outFilter)[i*(*outFilterSize) + j]= intV;

  1228. 			error = v - intV;

  1229. 		}

  1230. 	}

  1231. 	

  1232. 	(*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end

  1233. 	for(i=0; i<*outFilterSize; i++)

  1234. 	{

  1235. 		int j= dstW*(*outFilterSize);

  1236. 		(*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];

  1237. 	}

  1238. 

  1239. 	av_free(filter);

  1240.         return 0;

  1241. }

  1242. 

  1243. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  1244. static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)

  1245. {

  1246. 	uint8_t *fragmentA;

  1247. 	long imm8OfPShufW1A;

  1248. 	long imm8OfPShufW2A;

  1249. 	long fragmentLengthA;

  1250. 	uint8_t *fragmentB;

  1251. 	long imm8OfPShufW1B;

  1252. 	long imm8OfPShufW2B;

  1253. 	long fragmentLengthB;

  1254. 	int fragmentPos;

  1255. 

  1256. 	int xpos, i;

  1257. 

  1258. 	// create an optimized horizontal scaling routine

  1259. 

  1260. 	//code fragment

  1261. 

  1262. 	asm volatile(

  1263. 		"jmp 9f				\n\t"

  1264. 	// Begin

  1265. 		"0:				\n\t"

  1266. 		"movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 

  1267. 		"movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 

  1268. 		"movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"

  1269. 		"punpcklbw %%mm7, %%mm1		\n\t"

  1270. 		"punpcklbw %%mm7, %%mm0		\n\t"

  1271. 		"pshufw $0xFF, %%mm1, %%mm1	\n\t"

  1272. 		"1:				\n\t"

  1273. 		"pshufw $0xFF, %%mm0, %%mm0	\n\t"

  1274. 		"2:				\n\t"

  1275. 		"psubw %%mm1, %%mm0		\n\t"

  1276. 		"movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"

  1277. 		"pmullw %%mm3, %%mm0		\n\t"

  1278. 		"psllw $7, %%mm1		\n\t"

  1279. 		"paddw %%mm1, %%mm0		\n\t"

  1280. 

  1281. 		"movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"

  1282. 

  1283. 		"add $8, %%"REG_a"		\n\t"

  1284. 	// End

  1285. 		"9:				\n\t"

  1286. //		"int $3\n\t"

  1287. 		"lea 0b, %0			\n\t"

  1288. 		"lea 1b, %1			\n\t"

  1289. 		"lea 2b, %2			\n\t"

  1290. 		"dec %1				\n\t"

  1291. 		"dec %2				\n\t"

  1292. 		"sub %0, %1			\n\t"

  1293. 		"sub %0, %2			\n\t"

  1294. 		"lea 9b, %3			\n\t"

  1295. 		"sub %0, %3			\n\t"

  1296. 

  1297. 

  1298. 		:"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),

  1299. 		"=r" (fragmentLengthA)

  1300. 	);

  1301. 

  1302. 	asm volatile(

  1303. 		"jmp 9f				\n\t"

  1304. 	// Begin

  1305. 		"0:				\n\t"

  1306. 		"movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 

  1307. 		"movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 

  1308. 		"punpcklbw %%mm7, %%mm0		\n\t"

  1309. 		"pshufw $0xFF, %%mm0, %%mm1	\n\t"

  1310. 		"1:				\n\t"

  1311. 		"pshufw $0xFF, %%mm0, %%mm0	\n\t"

  1312. 		"2:				\n\t"

  1313. 		"psubw %%mm1, %%mm0		\n\t"

  1314. 		"movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"

  1315. 		"pmullw %%mm3, %%mm0		\n\t"

  1316. 		"psllw $7, %%mm1		\n\t"

  1317. 		"paddw %%mm1, %%mm0		\n\t"

  1318. 

  1319. 		"movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"

  1320. 

  1321. 		"add $8, %%"REG_a"		\n\t"

  1322. 	// End

  1323. 		"9:				\n\t"

  1324. //		"int $3\n\t"

  1325. 		"lea 0b, %0			\n\t"

  1326. 		"lea 1b, %1			\n\t"

  1327. 		"lea 2b, %2			\n\t"

  1328. 		"dec %1				\n\t"

  1329. 		"dec %2				\n\t"

  1330. 		"sub %0, %1			\n\t"

  1331. 		"sub %0, %2			\n\t"

  1332. 		"lea 9b, %3			\n\t"

  1333. 		"sub %0, %3			\n\t"

  1334. 

  1335. 

  1336. 		:"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),

  1337. 		"=r" (fragmentLengthB)

  1338. 	);

  1339. 

  1340. 	xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers

  1341. 	fragmentPos=0;

  1342. 	

  1343. 	for(i=0; i<dstW/numSplits; i++)

  1344. 	{

  1345. 		int xx=xpos>>16;

  1346. 

  1347. 		if((i&3) == 0)

  1348. 		{

  1349. 			int a=0;

  1350. 			int b=((xpos+xInc)>>16) - xx;

  1351. 			int c=((xpos+xInc*2)>>16) - xx;

  1352. 			int d=((xpos+xInc*3)>>16) - xx;

  1353. 

  1354. 			filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;

  1355. 			filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;

  1356. 			filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;

  1357. 			filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;

  1358. 			filterPos[i/2]= xx;

  1359. 

  1360. 			if(d+1<4)

  1361. 			{

  1362. 				int maxShift= 3-(d+1);

  1363. 				int shift=0;

  1364. 

  1365. 				memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);

  1366. 

  1367. 				funnyCode[fragmentPos + imm8OfPShufW1B]=

  1368. 					(a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);

  1369. 				funnyCode[fragmentPos + imm8OfPShufW2B]=

  1370. 					a | (b<<2) | (c<<4) | (d<<6);

  1371. 

  1372. 				if(i+3>=dstW) shift=maxShift; //avoid overread

  1373. 				else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align

  1374. 

  1375. 				if(shift && i>=shift)

  1376. 				{

  1377. 					funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;

  1378. 					funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;

  1379. 					filterPos[i/2]-=shift;

  1380. 				}

  1381. 

  1382. 				fragmentPos+= fragmentLengthB;

  1383. 			}

  1384. 			else

  1385. 			{

  1386. 				int maxShift= 3-d;

  1387. 				int shift=0;

  1388. 

  1389. 				memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);

  1390. 

  1391. 				funnyCode[fragmentPos + imm8OfPShufW1A]=

  1392. 				funnyCode[fragmentPos + imm8OfPShufW2A]=

  1393. 					a | (b<<2) | (c<<4) | (d<<6);

  1394. 

  1395. 				if(i+4>=dstW) shift=maxShift; //avoid overread

  1396. 				else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align

  1397. 

  1398. 				if(shift && i>=shift)

  1399. 				{

  1400. 					funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;

  1401. 					funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;

  1402. 					filterPos[i/2]-=shift;

  1403. 				}

  1404. 

  1405. 				fragmentPos+= fragmentLengthA;

  1406. 			}

  1407. 

  1408. 			funnyCode[fragmentPos]= RET;

  1409. 		}

  1410. 		xpos+=xInc;

  1411. 	}

  1412. 	filterPos[i/2]= xpos>>16; // needed to jump to the next part

  1413. }

  1414. #endif // ARCH_X86 || ARCH_X86_64

  1415. 

  1416. static void globalInit(void){

  1417.     // generating tables:

  1418.     int i;

  1419.     for(i=0; i<768; i++){

  1420. 	int c= FFMIN(FFMAX(i-256, 0), 255);

  1421. 	clip_table[i]=c;

  1422.     }

  1423. }

  1424. 

  1425. static SwsFunc getSwsFunc(int flags){

  1426.     

  1427. #ifdef RUNTIME_CPUDETECT

  1428. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  1429. 	// ordered per speed fasterst first

  1430. 	if(flags & SWS_CPU_CAPS_MMX2)

  1431. 		return swScale_MMX2;

  1432. 	else if(flags & SWS_CPU_CAPS_3DNOW)

  1433. 		return swScale_3DNow;

  1434. 	else if(flags & SWS_CPU_CAPS_MMX)

  1435. 		return swScale_MMX;

  1436. 	else

  1437. 		return swScale_C;

  1438. 

  1439. #else

  1440. #ifdef ARCH_POWERPC

  1441. 	if(flags & SWS_CPU_CAPS_ALTIVEC)

  1442. 	  return swScale_altivec;

  1443. 	else

  1444. 	  return swScale_C;

  1445. #endif

  1446. 	return swScale_C;

  1447. #endif /* defined(ARCH_X86) || defined(ARCH_X86_64) */

  1448. #else //RUNTIME_CPUDETECT

  1449. #ifdef HAVE_MMX2

  1450. 	return swScale_MMX2;

  1451. #elif defined (HAVE_3DNOW)

  1452. 	return swScale_3DNow;

  1453. #elif defined (HAVE_MMX)

  1454. 	return swScale_MMX;

  1455. #elif defined (HAVE_ALTIVEC)

  1456. 	return swScale_altivec;

  1457. #else

  1458. 	return swScale_C;

  1459. #endif

  1460. #endif //!RUNTIME_CPUDETECT

  1461. }

  1462. 

  1463. static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1464.              int srcSliceH, uint8_t* dstParam[], int dstStride[]){

  1465. 	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;

  1466. 	/* Copy Y plane */

  1467. 	if(dstStride[0]==srcStride[0] && srcStride[0] > 0)

  1468. 		memcpy(dst, src[0], srcSliceH*dstStride[0]);

  1469. 	else

  1470. 	{

  1471. 		int i;

  1472. 		uint8_t *srcPtr= src[0];

  1473. 		uint8_t *dstPtr= dst;

  1474. 		for(i=0; i<srcSliceH; i++)

  1475. 		{

  1476. 			memcpy(dstPtr, srcPtr, c->srcW);

  1477. 			srcPtr+= srcStride[0];

  1478. 			dstPtr+= dstStride[0];

  1479. 		}

  1480. 	}

  1481. 	dst = dstParam[1] + dstStride[1]*srcSliceY/2;

  1482. 	if (c->dstFormat == IMGFMT_NV12)

  1483. 		interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );

  1484. 	else

  1485. 		interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );

  1486. 

  1487. 	return srcSliceH;

  1488. }

  1489. 

  1490. static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1491.              int srcSliceH, uint8_t* dstParam[], int dstStride[]){

  1492. 	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;

  1493. 

  1494. 	yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );

  1495. 

  1496. 	return srcSliceH;

  1497. }

  1498. 

  1499. static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1500.              int srcSliceH, uint8_t* dstParam[], int dstStride[]){

  1501. 	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;

  1502. 

  1503. 	yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );

  1504. 

  1505. 	return srcSliceH;

  1506. }

  1507. 

  1508. /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */

  1509. static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1510. 			   int srcSliceH, uint8_t* dst[], int dstStride[]){

  1511. 	const int srcFormat= c->srcFormat;

  1512. 	const int dstFormat= c->dstFormat;

  1513. 	const int srcBpp= ((srcFormat&0xFF) + 7)>>3;

  1514. 	const int dstBpp= ((dstFormat&0xFF) + 7)>>3;

  1515. 	const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 

  1516. 	const int dstId= (dstFormat&0xFF)>>2;

  1517. 	void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;

  1518. 

  1519. 	/* BGR -> BGR */

  1520. 	if(   (isBGR(srcFormat) && isBGR(dstFormat))

  1521. 	   || (isRGB(srcFormat) && isRGB(dstFormat))){

  1522. 		switch(srcId | (dstId<<4)){

  1523. 		case 0x34: conv= rgb16to15; break;

  1524. 		case 0x36: conv= rgb24to15; break;

  1525. 		case 0x38: conv= rgb32to15; break;

  1526. 		case 0x43: conv= rgb15to16; break;

  1527. 		case 0x46: conv= rgb24to16; break;

  1528. 		case 0x48: conv= rgb32to16; break;

  1529. 		case 0x63: conv= rgb15to24; break;

  1530. 		case 0x64: conv= rgb16to24; break;

  1531. 		case 0x68: conv= rgb32to24; break;

  1532. 		case 0x83: conv= rgb15to32; break;

  1533. 		case 0x84: conv= rgb16to32; break;

  1534. 		case 0x86: conv= rgb24to32; break;

  1535. 		default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 

  1536. 				 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;

  1537. 		}

  1538. 	}else if(   (isBGR(srcFormat) && isRGB(dstFormat))

  1539. 		 || (isRGB(srcFormat) && isBGR(dstFormat))){

  1540. 		switch(srcId | (dstId<<4)){

  1541. 		case 0x33: conv= rgb15tobgr15; break;

  1542. 		case 0x34: conv= rgb16tobgr15; break;

  1543. 		case 0x36: conv= rgb24tobgr15; break;

  1544. 		case 0x38: conv= rgb32tobgr15; break;

  1545. 		case 0x43: conv= rgb15tobgr16; break;

  1546. 		case 0x44: conv= rgb16tobgr16; break;

  1547. 		case 0x46: conv= rgb24tobgr16; break;

  1548. 		case 0x48: conv= rgb32tobgr16; break;

  1549. 		case 0x63: conv= rgb15tobgr24; break;

  1550. 		case 0x64: conv= rgb16tobgr24; break;

  1551. 		case 0x66: conv= rgb24tobgr24; break;

  1552. 		case 0x68: conv= rgb32tobgr24; break;

  1553. 		case 0x83: conv= rgb15tobgr32; break;

  1554. 		case 0x84: conv= rgb16tobgr32; break;

  1555. 		case 0x86: conv= rgb24tobgr32; break;

  1556. 		case 0x88: conv= rgb32tobgr32; break;

  1557. 		default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 

  1558. 				 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;

  1559. 		}

  1560. 	}else{

  1561. 		MSG_ERR("swScaler: internal error %s -> %s converter\n", 

  1562. 			 sws_format_name(srcFormat), sws_format_name(dstFormat));

  1563. 	}

  1564. 

  1565. 	if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)

  1566. 		conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);

  1567. 	else

  1568. 	{

  1569. 		int i;

  1570. 		uint8_t *srcPtr= src[0];

  1571. 		uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;

  1572. 

  1573. 		for(i=0; i<srcSliceH; i++)

  1574. 		{

  1575. 			conv(srcPtr, dstPtr, c->srcW*srcBpp);

  1576. 			srcPtr+= srcStride[0];

  1577. 			dstPtr+= dstStride[0];

  1578. 		}

  1579. 	}     

  1580. 	return srcSliceH;

  1581. }

  1582. 

  1583. static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1584.              int srcSliceH, uint8_t* dst[], int dstStride[]){

  1585. 

  1586. 	rgb24toyv12(

  1587. 		src[0], 

  1588. 		dst[0]+ srcSliceY    *dstStride[0], 

  1589. 		dst[1]+(srcSliceY>>1)*dstStride[1], 

  1590. 		dst[2]+(srcSliceY>>1)*dstStride[2],

  1591. 		c->srcW, srcSliceH, 

  1592. 		dstStride[0], dstStride[1], srcStride[0]);

  1593. 	return srcSliceH;

  1594. }

  1595. 

  1596. static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1597.              int srcSliceH, uint8_t* dst[], int dstStride[]){

  1598. 	int i;

  1599. 

  1600. 	/* copy Y */

  1601. 	if(srcStride[0]==dstStride[0] && srcStride[0] > 0) 

  1602. 		memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);

  1603. 	else{

  1604. 		uint8_t *srcPtr= src[0];

  1605. 		uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;

  1606. 

  1607. 		for(i=0; i<srcSliceH; i++)

  1608. 		{

  1609. 			memcpy(dstPtr, srcPtr, c->srcW);

  1610. 			srcPtr+= srcStride[0];

  1611. 			dstPtr+= dstStride[0];

  1612. 		}

  1613. 	}

  1614. 

  1615. 	if(c->dstFormat==IMGFMT_YV12){

  1616. 		planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);

  1617. 		planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);

  1618. 	}else{

  1619. 		planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);

  1620. 		planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);

  1621. 	}

  1622. 	return srcSliceH;

  1623. }

  1624. 

  1625. /**

  1626.  * bring pointers in YUV order instead of YVU

  1627.  */

  1628. static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){

  1629. 	if(format == IMGFMT_YV12 || format == IMGFMT_YVU9

  1630.            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){

  1631. 		sortedP[0]= p[0];

  1632. 		sortedP[1]= p[2];

  1633. 		sortedP[2]= p[1];

  1634. 		sortedStride[0]= stride[0];

  1635. 		sortedStride[1]= stride[2];

  1636. 		sortedStride[2]= stride[1];

  1637. 	}

  1638. 	else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)

  1639. 	{

  1640. 		sortedP[0]= p[0];

  1641. 		sortedP[1]= 

  1642. 		sortedP[2]= NULL;

  1643. 		sortedStride[0]= stride[0];

  1644. 		sortedStride[1]= 

  1645. 		sortedStride[2]= 0;

  1646. 	}

  1647. 	else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)

  1648. 	{

  1649. 		sortedP[0]= p[0];

  1650. 		sortedP[1]= p[1];

  1651. 		sortedP[2]= p[2];

  1652. 		sortedStride[0]= stride[0];

  1653. 		sortedStride[1]= stride[1];

  1654. 		sortedStride[2]= stride[2];

  1655. 	}

  1656. 	else if(format == IMGFMT_NV12 || format == IMGFMT_NV21)

  1657. 	{

  1658. 		sortedP[0]= p[0];

  1659. 		sortedP[1]= p[1];

  1660. 		sortedP[2]= NULL;

  1661. 		sortedStride[0]= stride[0];

  1662. 		sortedStride[1]= stride[1];

  1663. 		sortedStride[2]= 0;

  1664. 	}else{

  1665. 		MSG_ERR("internal error in orderYUV\n");

  1666. 	}

  1667. }

  1668. 

  1669. /* unscaled copy like stuff (assumes nearly identical formats) */

  1670. static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  1671.              int srcSliceH, uint8_t* dst[], int dstStride[]){

  1672. 

  1673. 	if(isPacked(c->srcFormat))

  1674. 	{

  1675. 		if(dstStride[0]==srcStride[0] && srcStride[0] > 0)

  1676. 			memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);

  1677. 		else

  1678. 		{

  1679. 			int i;

  1680. 			uint8_t *srcPtr= src[0];

  1681. 			uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;

  1682. 			int length=0;

  1683. 

  1684. 			/* universal length finder */

  1685. 			while(length+c->srcW <= ABS(dstStride[0]) 

  1686. 			   && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;

  1687. 			ASSERT(length!=0);

  1688. 

  1689. 			for(i=0; i<srcSliceH; i++)

  1690. 			{

  1691. 				memcpy(dstPtr, srcPtr, length);

  1692. 				srcPtr+= srcStride[0];

  1693. 				dstPtr+= dstStride[0];

  1694. 			}

  1695. 		}

  1696. 	}

  1697. 	else 

  1698. 	{ /* Planar YUV or gray */

  1699. 		int plane;

  1700. 		for(plane=0; plane<3; plane++)

  1701. 		{

  1702. 			int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);

  1703. 			int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);

  1704. 			int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);

  1705. 

  1706. 			if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)

  1707. 			{

  1708. 				if(!isGray(c->dstFormat))

  1709. 					memset(dst[plane], 128, dstStride[plane]*height);

  1710. 			}

  1711. 			else

  1712. 			{

  1713. 				if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)

  1714. 					memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);

  1715. 				else

  1716. 				{

  1717. 					int i;

  1718. 					uint8_t *srcPtr= src[plane];

  1719. 					uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;

  1720. 					for(i=0; i<height; i++)

  1721. 					{

  1722. 						memcpy(dstPtr, srcPtr, length);

  1723. 						srcPtr+= srcStride[plane];

  1724. 						dstPtr+= dstStride[plane];

  1725. 					}

  1726. 				}

  1727. 			}

  1728. 		}

  1729. 	}

  1730. 	return srcSliceH;

  1731. }

  1732. 

  1733. static int remove_dup_fourcc(int fourcc)

  1734. {

  1735. 	switch(fourcc)

  1736. 	{

  1737. 	    case IMGFMT_I420:

  1738. 	    case IMGFMT_IYUV: return IMGFMT_YV12;

  1739. 	    case IMGFMT_Y8  : return IMGFMT_Y800;

  1740. 	    case IMGFMT_IF09: return IMGFMT_YVU9;

  1741. 	    default: return fourcc;

  1742. 	}

  1743. }

  1744. 

  1745. static void getSubSampleFactors(int *h, int *v, int format){

  1746. 	switch(format){

  1747. 	case IMGFMT_UYVY:

  1748. 	case IMGFMT_YUY2:

  1749. 		*h=1;

  1750. 		*v=0;

  1751. 		break;

  1752. 	case IMGFMT_YV12:

  1753. 	case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented

  1754. 	case IMGFMT_NV12:

  1755. 	case IMGFMT_NV21:

  1756. 		*h=1;

  1757. 		*v=1;

  1758. 		break;

  1759. 	case IMGFMT_YVU9:

  1760. 		*h=2;

  1761. 		*v=2;

  1762. 		break;

  1763. 	case IMGFMT_444P:

  1764. 		*h=0;

  1765. 		*v=0;

  1766. 		break;

  1767. 	case IMGFMT_422P:

  1768. 		*h=1;

  1769. 		*v=0;

  1770. 		break;

  1771. 	case IMGFMT_411P:

  1772. 		*h=2;

  1773. 		*v=0;

  1774. 		break;

  1775. 	default:

  1776. 		*h=0;

  1777. 		*v=0;

  1778. 		break;

  1779. 	}

  1780. }

  1781. 

  1782. static uint16_t roundToInt16(int64_t f){

  1783. 	int r= (f + (1<<15))>>16;

  1784. 	     if(r<-0x7FFF) return 0x8000;

  1785. 	else if(r> 0x7FFF) return 0x7FFF;

  1786. 	else               return r;

  1787. }

  1788. 

  1789. /**

  1790.  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]

  1791.  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235

  1792.  * @return -1 if not supported

  1793.  */

  1794. int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){

  1795. 	int64_t crv =  inv_table[0];

  1796. 	int64_t cbu =  inv_table[1];

  1797. 	int64_t cgu = -inv_table[2];

  1798. 	int64_t cgv = -inv_table[3];

  1799. 	int64_t cy  = 1<<16;

  1800. 	int64_t oy  = 0;

  1801. 

  1802. 	if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;

  1803. 	memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);

  1804. 	memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);

  1805. 

  1806. 	c->brightness= brightness;

  1807. 	c->contrast  = contrast;

  1808. 	c->saturation= saturation;

  1809. 	c->srcRange  = srcRange;

  1810. 	c->dstRange  = dstRange;

  1811. 

  1812. 	c->uOffset=   0x0400040004000400LL;

  1813. 	c->vOffset=   0x0400040004000400LL;

  1814. 

  1815. 	if(!srcRange){

  1816. 		cy= (cy*255) / 219;

  1817. 		oy= 16<<16;

  1818. 	}

  1819. 

  1820. 	cy = (cy *contrast             )>>16;

  1821. 	crv= (crv*contrast * saturation)>>32;

  1822. 	cbu= (cbu*contrast * saturation)>>32;

  1823. 	cgu= (cgu*contrast * saturation)>>32;

  1824. 	cgv= (cgv*contrast * saturation)>>32;

  1825. 

  1826. 	oy -= 256*brightness;

  1827. 

  1828. 	c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;

  1829. 	c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;

  1830. 	c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;

  1831. 	c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;

  1832. 	c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;

  1833. 	c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;

  1834. 

  1835. 	yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);

  1836. 	//FIXME factorize

  1837. 

  1838. #ifdef COMPILE_ALTIVEC

  1839. 	if (c->flags & SWS_CPU_CAPS_ALTIVEC)

  1840. 	    yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);

  1841. #endif	

  1842. 	return 0;

  1843. }

  1844. 

  1845. /**

  1846.  * @return -1 if not supported

  1847.  */

  1848. int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){

  1849. 	if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;

  1850. 

  1851. 	*inv_table = c->srcColorspaceTable;

  1852. 	*table     = c->dstColorspaceTable;

  1853. 	*srcRange  = c->srcRange;

  1854. 	*dstRange  = c->dstRange;

  1855. 	*brightness= c->brightness;

  1856. 	*contrast  = c->contrast;

  1857. 	*saturation= c->saturation;

  1858. 	

  1859. 	return 0;	

  1860. }

  1861. 

  1862. SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,

  1863.                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){

  1864. 

  1865. 	SwsContext *c;

  1866. 	int i;

  1867. 	int usesVFilter, usesHFilter;

  1868. 	int unscaled, needsDither;

  1869. 	int srcFormat, dstFormat;

  1870. 	SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};

  1871. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  1872. 	if(flags & SWS_CPU_CAPS_MMX)

  1873. 		asm volatile("emms\n\t"::: "memory");

  1874. #endif

  1875. 

  1876. #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off

  1877. 	flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);

  1878. #ifdef HAVE_MMX2

  1879. 	flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;

  1880. #elif defined (HAVE_3DNOW)

  1881. 	flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;

  1882. #elif defined (HAVE_MMX)

  1883. 	flags |= SWS_CPU_CAPS_MMX;

  1884. #elif defined (HAVE_ALTIVEC)

  1885. 	flags |= SWS_CPU_CAPS_ALTIVEC;

  1886. #endif

  1887. #endif /* RUNTIME_CPUDETECT */

  1888. 	if(clip_table[512] != 255) globalInit();

  1889. 	if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);

  1890. 

  1891. 	/* avoid duplicate Formats, so we don't need to check to much */

  1892.         if (origSrcFormat < PIX_FMT_NB) {

  1893.             origSrcFormat = fmt_name[origSrcFormat];

  1894.         }

  1895.         if (origDstFormat < PIX_FMT_NB) {

  1896.             origDstFormat = fmt_name[origDstFormat];

  1897.         }

  1898. 	srcFormat = remove_dup_fourcc(origSrcFormat);

  1899. 	dstFormat = remove_dup_fourcc(origDstFormat);

  1900. 

  1901. 	unscaled = (srcW == dstW && srcH == dstH);

  1902. 	needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 

  1903. 		     && (dstFormat&0xFF)<24

  1904. 		     && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));

  1905. 

  1906. 	if(!isSupportedIn(srcFormat)) 

  1907. 	{

  1908. 		MSG_ERR("swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));

  1909. 		return NULL;

  1910. 	}

  1911. 	if(!isSupportedOut(dstFormat))

  1912. 	{

  1913. 		MSG_ERR("swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));

  1914. 		return NULL;

  1915. 	}

  1916. 

  1917. 	/* sanity check */

  1918. 	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

  1919. 	{

  1920. 		 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 

  1921. 			srcW, srcH, dstW, dstH);

  1922. 		return NULL;

  1923. 	}

  1924. 

  1925. 	if(!dstFilter) dstFilter= &dummyFilter;

  1926. 	if(!srcFilter) srcFilter= &dummyFilter;

  1927. 

  1928. 	c= av_malloc(sizeof(SwsContext));

  1929. 	memset(c, 0, sizeof(SwsContext));

  1930. 

  1931. 	c->srcW= srcW;

  1932. 	c->srcH= srcH;

  1933. 	c->dstW= dstW;

  1934. 	c->dstH= dstH;

  1935. 	c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;

  1936. 	c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;

  1937. 	c->flags= flags;

  1938. 	c->dstFormat= dstFormat;

  1939. 	c->srcFormat= srcFormat;

  1940. 	c->origDstFormat= origDstFormat;

  1941. 	c->origSrcFormat= origSrcFormat;

  1942.         c->vRounder= 4* 0x0001000100010001ULL;

  1943. 

  1944. 	usesHFilter= usesVFilter= 0;

  1945. 	if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;

  1946. 	if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;

  1947. 	if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;

  1948. 	if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;

  1949. 	if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;

  1950. 	if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;

  1951. 	if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;

  1952. 	if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;

  1953. 

  1954. 	getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);

  1955. 	getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);

  1956. 

  1957. 	// reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation

  1958. 	if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;

  1959. 

  1960. 	// drop some chroma lines if the user wants it

  1961. 	c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;

  1962. 	c->chrSrcVSubSample+= c->vChrDrop;

  1963. 

  1964. 	// drop every 2. pixel for chroma calculation unless user wants full chroma

  1965. 	if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 

  1966. 		c->chrSrcHSubSample=1;

  1967. 

  1968. 	if(param){

  1969. 		c->param[0] = param[0];

  1970. 		c->param[1] = param[1];

  1971. 	}else{

  1972. 		c->param[0] =

  1973. 		c->param[1] = SWS_PARAM_DEFAULT;

  1974. 	}

  1975. 

  1976. 	c->chrIntHSubSample= c->chrDstHSubSample;

  1977. 	c->chrIntVSubSample= c->chrSrcVSubSample;

  1978. 

  1979. 	// note the -((-x)>>y) is so that we allways round toward +inf

  1980. 	c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);

  1981. 	c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);

  1982. 	c->chrDstW= -((-dstW) >> c->chrDstHSubSample);

  1983. 	c->chrDstH= -((-dstH) >> c->chrDstVSubSample);

  1984. 

  1985. 	sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], 0, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, 0, 0, 1<<16, 1<<16); 

  1986. 

  1987. 	/* unscaled special Cases */

  1988. 	if(unscaled && !usesHFilter && !usesVFilter)

  1989. 	{

  1990. 		/* yv12_to_nv12 */

  1991. 		if(srcFormat == IMGFMT_YV12 && (dstFormat == IMGFMT_NV12 || dstFormat == IMGFMT_NV21))

  1992. 		{

  1993. 			c->swScale= PlanarToNV12Wrapper;

  1994. 		}

  1995. 		/* yuv2bgr */

  1996. 		if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))

  1997. 		{

  1998. 			c->swScale= yuv2rgb_get_func_ptr(c);

  1999. 		}

  2000. 		

  2001. 		if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )

  2002. 		{

  2003. 			c->swScale= yvu9toyv12Wrapper;

  2004. 		}

  2005. 

  2006. 		/* bgr24toYV12 */

  2007. 		if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)

  2008. 			c->swScale= bgr24toyv12Wrapper;

  2009. 		

  2010. 		/* rgb/bgr -> rgb/bgr (no dither needed forms) */

  2011. 		if(   (isBGR(srcFormat) || isRGB(srcFormat))

  2012. 		   && (isBGR(dstFormat) || isRGB(dstFormat)) 

  2013. 		   && !needsDither)

  2014. 			c->swScale= rgb2rgbWrapper;

  2015. 

  2016. 		/* LQ converters if -sws 0 or -sws 4*/

  2017. 		if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){

  2018. 			/* rgb/bgr -> rgb/bgr (dither needed forms) */

  2019. 			if(  (isBGR(srcFormat) || isRGB(srcFormat))

  2020. 			  && (isBGR(dstFormat) || isRGB(dstFormat)) 

  2021. 			  && needsDither)

  2022. 				c->swScale= rgb2rgbWrapper;

  2023. 

  2024. 			/* yv12_to_yuy2 */

  2025. 			if(srcFormat == IMGFMT_YV12 && 

  2026. 			    (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))

  2027. 			{

  2028. 				if (dstFormat == IMGFMT_YUY2)

  2029. 				    c->swScale= PlanarToYuy2Wrapper;

  2030. 				else

  2031. 				    c->swScale= PlanarToUyvyWrapper;

  2032. 			}

  2033. 		}

  2034. 

  2035. #ifdef COMPILE_ALTIVEC

  2036. 		if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&

  2037. 		    ((srcFormat == IMGFMT_YV12 && 

  2038. 		      (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {

  2039. 		  // unscaled YV12 -> packed YUV, we want speed

  2040. 		  if (dstFormat == IMGFMT_YUY2)

  2041. 		    c->swScale= yv12toyuy2_unscaled_altivec;

  2042. 		  else

  2043. 		    c->swScale= yv12touyvy_unscaled_altivec;

  2044. 		}

  2045. #endif

  2046. 

  2047. 		/* simple copy */

  2048. 		if(   srcFormat == dstFormat

  2049. 		   || (isPlanarYUV(srcFormat) && isGray(dstFormat))

  2050. 		   || (isPlanarYUV(dstFormat) && isGray(srcFormat))

  2051. 		  )

  2052. 		{

  2053. 			c->swScale= simpleCopy;

  2054. 		}

  2055. 

  2056. 		if(c->swScale){

  2057. 			if(flags&SWS_PRINT_INFO)

  2058. 				MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 

  2059. 					sws_format_name(srcFormat), sws_format_name(dstFormat));

  2060. 			return c;

  2061. 		}

  2062. 	}

  2063. 

  2064. 	if(flags & SWS_CPU_CAPS_MMX2)

  2065. 	{

  2066. 		c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;

  2067. 		if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))

  2068. 		{

  2069. 			if(flags&SWS_PRINT_INFO)

  2070. 				MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");

  2071. 		}

  2072. 		if(usesHFilter) c->canMMX2BeUsed=0;

  2073. 	}

  2074. 	else

  2075. 		c->canMMX2BeUsed=0;

  2076. 

  2077. 	c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;

  2078. 	c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;

  2079. 

  2080. 	// match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst

  2081. 	// but only for the FAST_BILINEAR mode otherwise do correct scaling

  2082. 	// n-2 is the last chrominance sample available

  2083. 	// this is not perfect, but noone shuld notice the difference, the more correct variant

  2084. 	// would be like the vertical one, but that would require some special code for the

  2085. 	// first and last pixel

  2086. 	if(flags&SWS_FAST_BILINEAR)

  2087. 	{

  2088. 		if(c->canMMX2BeUsed)

  2089. 		{

  2090. 			c->lumXInc+= 20;

  2091. 			c->chrXInc+= 20;

  2092. 		}

  2093. 		//we don't use the x86asm scaler if mmx is available

  2094. 		else if(flags & SWS_CPU_CAPS_MMX)

  2095. 		{

  2096. 			c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;

  2097. 			c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;

  2098. 		}

  2099. 	}

  2100. 

  2101. 	/* precalculate horizontal scaler filter coefficients */

  2102. 	{

  2103. 		const int filterAlign=

  2104. 		  (flags & SWS_CPU_CAPS_MMX) ? 4 :

  2105. 		  (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  2106. 		  1;

  2107. 

  2108. 		initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,

  2109. 				 srcW      ,       dstW, filterAlign, 1<<14,

  2110. 				 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,

  2111. 				 srcFilter->lumH, dstFilter->lumH, c->param);

  2112. 		initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,

  2113. 				 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,

  2114. 				 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,

  2115. 				 srcFilter->chrH, dstFilter->chrH, c->param);

  2116. 

  2117. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  2118. // can't downscale !!!

  2119. 		if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))

  2120. 		{

  2121. #define MAX_FUNNY_CODE_SIZE 10000

  2122. #ifdef MAP_ANONYMOUS

  2123. 			c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);

  2124. 			c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);

  2125. #else

  2126. 			c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);

  2127. 			c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);

  2128. #endif

  2129. 

  2130. 			c->lumMmx2Filter   = av_malloc((dstW        /8+8)*sizeof(int16_t));

  2131. 			c->chrMmx2Filter   = av_malloc((c->chrDstW  /4+8)*sizeof(int16_t));

  2132. 			c->lumMmx2FilterPos= av_malloc((dstW      /2/8+8)*sizeof(int32_t));

  2133. 			c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));

  2134. 

  2135. 			initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);

  2136. 			initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);

  2137. 		}

  2138. #endif /* defined(ARCH_X86) || defined(ARCH_X86_64) */

  2139. 	} // Init Horizontal stuff

  2140. 

  2141. 

  2142. 

  2143. 	/* precalculate vertical scaler filter coefficients */

  2144. 	{

  2145. 		const int filterAlign=

  2146. 		  (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :

  2147. 		  (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  2148. 		  1;

  2149. 

  2150. 		initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,

  2151. 				srcH      ,        dstH, filterAlign, (1<<12)-4,

  2152. 				(flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,

  2153. 				srcFilter->lumV, dstFilter->lumV, c->param);

  2154. 		initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,

  2155. 				c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,

  2156. 				(flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,

  2157. 				srcFilter->chrV, dstFilter->chrV, c->param);

  2158. 

  2159. #ifdef HAVE_ALTIVEC

  2160. 		c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);

  2161. 		c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);

  2162. 

  2163. 		for (i=0;i<c->vLumFilterSize*c->dstH;i++) {

  2164.                   int j;

  2165. 		  short *p = (short *)&c->vYCoeffsBank[i];

  2166. 		  for (j=0;j<8;j++)

  2167. 		    p[j] = c->vLumFilter[i];

  2168. 		}

  2169. 

  2170. 		for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {

  2171.                   int j;

  2172. 		  short *p = (short *)&c->vCCoeffsBank[i];

  2173. 		  for (j=0;j<8;j++)

  2174. 		    p[j] = c->vChrFilter[i];

  2175. 		}

  2176. #endif

  2177. 	}

  2178. 

  2179. 	// Calculate Buffer Sizes so that they won't run out while handling these damn slices

  2180. 	c->vLumBufSize= c->vLumFilterSize;

  2181. 	c->vChrBufSize= c->vChrFilterSize;

  2182. 	for(i=0; i<dstH; i++)

  2183. 	{

  2184. 		int chrI= i*c->chrDstH / dstH;

  2185. 		int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  2186. 				 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  2187. 

  2188. 		nextSlice>>= c->chrSrcVSubSample;

  2189. 		nextSlice<<= c->chrSrcVSubSample;

  2190. 		if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  2191. 			c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];

  2192. 		if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  2193. 			c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  2194. 	}

  2195. 

  2196. 	// allocate pixbufs (we use dynamic allocation because otherwise we would need to

  2197. 	c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));

  2198. 	c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));

  2199. 	//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)

  2200. 	/* align at 16 bytes for AltiVec */

  2201. 	for(i=0; i<c->vLumBufSize; i++)

  2202. 		c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_malloc(4000);

  2203. 	for(i=0; i<c->vChrBufSize; i++)

  2204. 		c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000);

  2205. 

  2206. 	//try to avoid drawing green stuff between the right end and the stride end

  2207. 	for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);

  2208. 	for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);

  2209. 

  2210. 	ASSERT(c->chrDstH <= dstH)

  2211. 

  2212. 	if(flags&SWS_PRINT_INFO)

  2213. 	{

  2214. #ifdef DITHER1XBPP

  2215. 		char *dither= " dithered";

  2216. #else

  2217. 		char *dither= "";

  2218. #endif

  2219. 		if(flags&SWS_FAST_BILINEAR)

  2220. 			MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");

  2221. 		else if(flags&SWS_BILINEAR)

  2222. 			MSG_INFO("\nSwScaler: BILINEAR scaler, ");

  2223. 		else if(flags&SWS_BICUBIC)

  2224. 			MSG_INFO("\nSwScaler: BICUBIC scaler, ");

  2225. 		else if(flags&SWS_X)

  2226. 			MSG_INFO("\nSwScaler: Experimental scaler, ");

  2227. 		else if(flags&SWS_POINT)

  2228. 			MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");

  2229. 		else if(flags&SWS_AREA)

  2230. 			MSG_INFO("\nSwScaler: Area Averageing scaler, ");

  2231. 		else if(flags&SWS_BICUBLIN)

  2232. 			MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");

  2233. 		else if(flags&SWS_GAUSS)

  2234. 			MSG_INFO("\nSwScaler: Gaussian scaler, ");

  2235. 		else if(flags&SWS_SINC)

  2236. 			MSG_INFO("\nSwScaler: Sinc scaler, ");

  2237. 		else if(flags&SWS_LANCZOS)

  2238. 			MSG_INFO("\nSwScaler: Lanczos scaler, ");

  2239. 		else if(flags&SWS_SPLINE)

  2240. 			MSG_INFO("\nSwScaler: Bicubic spline scaler, ");

  2241. 		else

  2242. 			MSG_INFO("\nSwScaler: ehh flags invalid?! ");

  2243. 

  2244. 		if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)

  2245. 			MSG_INFO("from %s to%s %s ", 

  2246. 				sws_format_name(srcFormat), dither, sws_format_name(dstFormat));

  2247. 		else

  2248. 			MSG_INFO("from %s to %s ", 

  2249. 				sws_format_name(srcFormat), sws_format_name(dstFormat));

  2250. 

  2251. 		if(flags & SWS_CPU_CAPS_MMX2)

  2252. 			MSG_INFO("using MMX2\n");

  2253. 		else if(flags & SWS_CPU_CAPS_3DNOW)

  2254. 			MSG_INFO("using 3DNOW\n");

  2255. 		else if(flags & SWS_CPU_CAPS_MMX)

  2256. 			MSG_INFO("using MMX\n");

  2257. 		else if(flags & SWS_CPU_CAPS_ALTIVEC)

  2258. 			MSG_INFO("using AltiVec\n");

  2259. 		else 

  2260. 			MSG_INFO("using C\n");

  2261. 	}

  2262. 

  2263. 	if(flags & SWS_PRINT_INFO)

  2264. 	{

  2265. 		if(flags & SWS_CPU_CAPS_MMX)

  2266. 		{

  2267. 			if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  2268. 				MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  2269. 			else

  2270. 			{

  2271. 				if(c->hLumFilterSize==4)

  2272. 					MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");

  2273. 				else if(c->hLumFilterSize==8)

  2274. 					MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");

  2275. 				else

  2276. 					MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");

  2277. 

  2278. 				if(c->hChrFilterSize==4)

  2279. 					MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");

  2280. 				else if(c->hChrFilterSize==8)

  2281. 					MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");

  2282. 				else

  2283. 					MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");

  2284. 			}

  2285. 		}

  2286. 		else

  2287. 		{

  2288. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  2289. 			MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");

  2290. #else

  2291. 			if(flags & SWS_FAST_BILINEAR)

  2292. 				MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");

  2293. 			else

  2294. 				MSG_V("SwScaler: using C scaler for horizontal scaling\n");

  2295. #endif

  2296. 		}

  2297. 		if(isPlanarYUV(dstFormat))

  2298. 		{

  2299. 			if(c->vLumFilterSize==1)

  2300. 				MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  2301. 			else

  2302. 				MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  2303. 		}

  2304. 		else

  2305. 		{

  2306. 			if(c->vLumFilterSize==1 && c->vChrFilterSize==2)

  2307. 				MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  2308. 				       "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  2309. 			else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)

  2310. 				MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  2311. 			else

  2312. 				MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  2313. 		}

  2314. 

  2315. 		if(dstFormat==IMGFMT_BGR24)

  2316. 			MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",

  2317. 				(flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));

  2318. 		else if(dstFormat==IMGFMT_BGR32)

  2319. 			MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  2320. 		else if(dstFormat==IMGFMT_BGR16)

  2321. 			MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  2322. 		else if(dstFormat==IMGFMT_BGR15)

  2323. 			MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  2324. 

  2325. 		MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  2326. 	}

  2327. 	if(flags & SWS_PRINT_INFO)

  2328. 	{

  2329. 		MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  2330. 			c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  2331. 		MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  2332. 			c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  2333. 	}

  2334. 

  2335. 	c->swScale= getSwsFunc(flags);

  2336. 	return c;

  2337. }

  2338. 

  2339. /**

  2340.  * swscale warper, so we don't need to export the SwsContext.

  2341.  * assumes planar YUV to be in YUV order instead of YVU

  2342.  */

  2343. int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,

  2344.                            int srcSliceH, uint8_t* dst[], int dstStride[]){

  2345. 	if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {

  2346. 	    MSG_ERR("swScaler: slices start in the middle!\n");

  2347. 	    return 0;

  2348. 	}

  2349. 	if (c->sliceDir == 0) {

  2350. 	    if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;

  2351. 	}

  2352. 

  2353. 	// copy strides, so they can safely be modified

  2354. 	if (c->sliceDir == 1) {

  2355. 	    // slices go from top to bottom

  2356. 	    int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};

  2357. 	    int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};

  2358. 	    return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);

  2359. 	} else {

  2360. 	    // slices go from bottom to top => we flip the image internally

  2361. 	    uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],

  2362. 			       src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],

  2363. 			       src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]

  2364. 	    };

  2365. 	    uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],

  2366. 			       dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],

  2367. 			       dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};

  2368. 	    int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};

  2369. 	    int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};

  2370. 	    

  2371. 	    return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);

  2372. 	}

  2373. }

  2374. 

  2375. /**

  2376.  * swscale warper, so we don't need to export the SwsContext

  2377.  */

  2378. int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,

  2379.                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){

  2380. 	int srcStride[3];

  2381. 	int dstStride[3];

  2382. 	uint8_t *src[3];

  2383. 	uint8_t *dst[3];

  2384. 	sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);

  2385. 	sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);

  2386. //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);

  2387. 

  2388. 	return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);

  2389. }

  2390. 

  2391. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 

  2392. 				float lumaSharpen, float chromaSharpen,

  2393. 				float chromaHShift, float chromaVShift,

  2394. 				int verbose)

  2395. {

  2396. 	SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  2397. 

  2398. 	if(lumaGBlur!=0.0){

  2399. 		filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  2400. 		filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  2401. 	}else{

  2402. 		filter->lumH= sws_getIdentityVec();

  2403. 		filter->lumV= sws_getIdentityVec();

  2404. 	}

  2405. 

  2406. 	if(chromaGBlur!=0.0){

  2407. 		filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  2408. 		filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  2409. 	}else{

  2410. 		filter->chrH= sws_getIdentityVec();

  2411. 		filter->chrV= sws_getIdentityVec();

  2412. 	}

  2413. 

  2414. 	if(chromaSharpen!=0.0){

  2415. 		SwsVector *id= sws_getIdentityVec();

  2416.                 sws_scaleVec(filter->chrH, -chromaSharpen);

  2417.                 sws_scaleVec(filter->chrV, -chromaSharpen);

  2418. 		sws_addVec(filter->chrH, id);

  2419. 		sws_addVec(filter->chrV, id);

  2420. 		sws_freeVec(id);

  2421. 	}

  2422. 

  2423. 	if(lumaSharpen!=0.0){

  2424. 		SwsVector *id= sws_getIdentityVec();

  2425.                 sws_scaleVec(filter->lumH, -lumaSharpen);

  2426.                 sws_scaleVec(filter->lumV, -lumaSharpen);

  2427. 		sws_addVec(filter->lumH, id);

  2428. 		sws_addVec(filter->lumV, id);

  2429. 		sws_freeVec(id);

  2430. 	}

  2431. 

  2432. 	if(chromaHShift != 0.0)

  2433. 		sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  2434. 

  2435. 	if(chromaVShift != 0.0)

  2436. 		sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  2437. 

  2438. 	sws_normalizeVec(filter->chrH, 1.0);

  2439. 	sws_normalizeVec(filter->chrV, 1.0);

  2440. 	sws_normalizeVec(filter->lumH, 1.0);

  2441. 	sws_normalizeVec(filter->lumV, 1.0);

  2442. 

  2443. 	if(verbose) sws_printVec(filter->chrH);

  2444. 	if(verbose) sws_printVec(filter->lumH);

  2445. 

  2446.         return filter;

  2447. }

  2448. 

  2449. /**

  2450.  * returns a normalized gaussian curve used to filter stuff

  2451.  * quality=3 is high quality, lowwer is lowwer quality

  2452.  */

  2453. SwsVector *sws_getGaussianVec(double variance, double quality){

  2454. 	const int length= (int)(variance*quality + 0.5) | 1;

  2455. 	int i;

  2456. 	double *coeff= av_malloc(length*sizeof(double));

  2457. 	double middle= (length-1)*0.5;

  2458. 	SwsVector *vec= av_malloc(sizeof(SwsVector));

  2459. 

  2460. 	vec->coeff= coeff;

  2461. 	vec->length= length;

  2462. 

  2463. 	for(i=0; i<length; i++)

  2464. 	{

  2465. 		double dist= i-middle;

  2466. 		coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);

  2467. 	}

  2468. 

  2469. 	sws_normalizeVec(vec, 1.0);

  2470. 

  2471. 	return vec;

  2472. }

  2473. 

  2474. SwsVector *sws_getConstVec(double c, int length){

  2475. 	int i;

  2476. 	double *coeff= av_malloc(length*sizeof(double));

  2477. 	SwsVector *vec= av_malloc(sizeof(SwsVector));

  2478. 

  2479. 	vec->coeff= coeff;

  2480. 	vec->length= length;

  2481. 

  2482. 	for(i=0; i<length; i++)

  2483. 		coeff[i]= c;

  2484. 

  2485. 	return vec;

  2486. }

  2487. 

  2488. 

  2489. SwsVector *sws_getIdentityVec(void){

  2490.         return sws_getConstVec(1.0, 1);

  2491. }

  2492. 

  2493. double sws_dcVec(SwsVector *a){

  2494. 	int i;

  2495.         double sum=0;

  2496. 

  2497. 	for(i=0; i<a->length; i++)

  2498. 		sum+= a->coeff[i];

  2499. 

  2500.         return sum;

  2501. }

  2502. 

  2503. void sws_scaleVec(SwsVector *a, double scalar){

  2504. 	int i;

  2505. 

  2506. 	for(i=0; i<a->length; i++)

  2507. 		a->coeff[i]*= scalar;

  2508. }

  2509. 

  2510. void sws_normalizeVec(SwsVector *a, double height){

  2511.         sws_scaleVec(a, height/sws_dcVec(a));

  2512. }

  2513. 

  2514. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){

  2515. 	int length= a->length + b->length - 1;

  2516. 	double *coeff= av_malloc(length*sizeof(double));

  2517. 	int i, j;

  2518. 	SwsVector *vec= av_malloc(sizeof(SwsVector));

  2519. 

  2520. 	vec->coeff= coeff;

  2521. 	vec->length= length;

  2522. 

  2523. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  2524. 

  2525. 	for(i=0; i<a->length; i++)

  2526. 	{

  2527. 		for(j=0; j<b->length; j++)

  2528. 		{

  2529. 			coeff[i+j]+= a->coeff[i]*b->coeff[j];

  2530. 		}

  2531. 	}

  2532. 

  2533. 	return vec;

  2534. }

  2535. 

  2536. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){

  2537. 	int length= FFMAX(a->length, b->length);

  2538. 	double *coeff= av_malloc(length*sizeof(double));

  2539. 	int i;

  2540. 	SwsVector *vec= av_malloc(sizeof(SwsVector));

  2541. 

  2542. 	vec->coeff= coeff;

  2543. 	vec->length= length;

  2544. 

  2545. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  2546. 

  2547. 	for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  2548. 	for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  2549. 

  2550. 	return vec;

  2551. }

  2552. 

  2553. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){

  2554. 	int length= FFMAX(a->length, b->length);

  2555. 	double *coeff= av_malloc(length*sizeof(double));

  2556. 	int i;

  2557. 	SwsVector *vec= av_malloc(sizeof(SwsVector));

  2558. 

  2559. 	vec->coeff= coeff;

  2560. 	vec->length= length;

  2561. 

  2562. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  2563. 

  2564. 	for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  2565. 	for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  2566. 

  2567. 	return vec;

  2568. }

  2569. 

  2570. /* shift left / or right if "shift" is negative */

  2571. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){

  2572. 	int length= a->length + ABS(shift)*2;

  2573. 	double *coeff= av_malloc(length*sizeof(double));

  2574. 	int i;

  2575. 	SwsVector *vec= av_malloc(sizeof(SwsVector));

  2576. 

  2577. 	vec->coeff= coeff;

  2578. 	vec->length= length;

  2579. 

  2580. 	for(i=0; i<length; i++) coeff[i]= 0.0;

  2581. 

  2582. 	for(i=0; i<a->length; i++)

  2583. 	{

  2584. 		coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  2585. 	}

  2586. 

  2587. 	return vec;

  2588. }

  2589. 

  2590. void sws_shiftVec(SwsVector *a, int shift){

  2591. 	SwsVector *shifted= sws_getShiftedVec(a, shift);

  2592. 	av_free(a->coeff);

  2593. 	a->coeff= shifted->coeff;

  2594. 	a->length= shifted->length;

  2595. 	av_free(shifted);

  2596. }

  2597. 

  2598. void sws_addVec(SwsVector *a, SwsVector *b){

  2599. 	SwsVector *sum= sws_sumVec(a, b);

  2600. 	av_free(a->coeff);

  2601. 	a->coeff= sum->coeff;

  2602. 	a->length= sum->length;

  2603. 	av_free(sum);

  2604. }

  2605. 

  2606. void sws_subVec(SwsVector *a, SwsVector *b){

  2607. 	SwsVector *diff= sws_diffVec(a, b);

  2608. 	av_free(a->coeff);

  2609. 	a->coeff= diff->coeff;

  2610. 	a->length= diff->length;

  2611. 	av_free(diff);

  2612. }

  2613. 

  2614. void sws_convVec(SwsVector *a, SwsVector *b){

  2615. 	SwsVector *conv= sws_getConvVec(a, b);

  2616. 	av_free(a->coeff);  

  2617. 	a->coeff= conv->coeff;

  2618. 	a->length= conv->length;

  2619. 	av_free(conv);

  2620. }

  2621. 

  2622. SwsVector *sws_cloneVec(SwsVector *a){

  2623. 	double *coeff= av_malloc(a->length*sizeof(double));

  2624. 	int i;

  2625. 	SwsVector *vec= av_malloc(sizeof(SwsVector));

  2626. 

  2627. 	vec->coeff= coeff;

  2628. 	vec->length= a->length;

  2629. 

  2630. 	for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];

  2631. 

  2632. 	return vec;

  2633. }

  2634. 

  2635. void sws_printVec(SwsVector *a){

  2636. 	int i;

  2637. 	double max=0;

  2638. 	double min=0;

  2639. 	double range;

  2640. 

  2641. 	for(i=0; i<a->length; i++)

  2642. 		if(a->coeff[i]>max) max= a->coeff[i];

  2643. 

  2644. 	for(i=0; i<a->length; i++)

  2645. 		if(a->coeff[i]<min) min= a->coeff[i];

  2646. 

  2647. 	range= max - min;

  2648. 

  2649. 	for(i=0; i<a->length; i++)

  2650. 	{

  2651. 		int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  2652. 		MSG_DBG2("%1.3f ", a->coeff[i]);

  2653. 		for(;x>0; x--) MSG_DBG2(" ");

  2654. 		MSG_DBG2("|\n");

  2655. 	}

  2656. }

  2657. 

  2658. void sws_freeVec(SwsVector *a){

  2659. 	if(!a) return;

  2660. 	av_free(a->coeff);

  2661. 	a->coeff=NULL;

  2662. 	a->length=0;

  2663. 	av_free(a);

  2664. }

  2665. 

  2666. void sws_freeFilter(SwsFilter *filter){

  2667. 	if(!filter) return;

  2668. 

  2669. 	if(filter->lumH) sws_freeVec(filter->lumH);

  2670. 	if(filter->lumV) sws_freeVec(filter->lumV);

  2671. 	if(filter->chrH) sws_freeVec(filter->chrH);

  2672. 	if(filter->chrV) sws_freeVec(filter->chrV);

  2673. 	av_free(filter);

  2674. }

  2675. 

  2676. 

  2677. void sws_freeContext(SwsContext *c){

  2678. 	int i;

  2679. 	if(!c) return;

  2680. 

  2681. 	if(c->lumPixBuf)

  2682. 	{

  2683. 		for(i=0; i<c->vLumBufSize; i++)

  2684. 		{

  2685. 			av_free(c->lumPixBuf[i]);

  2686. 			c->lumPixBuf[i]=NULL;

  2687. 		}

  2688. 		av_free(c->lumPixBuf);

  2689. 		c->lumPixBuf=NULL;

  2690. 	}

  2691. 

  2692. 	if(c->chrPixBuf)

  2693. 	{

  2694. 		for(i=0; i<c->vChrBufSize; i++)

  2695. 		{

  2696. 			av_free(c->chrPixBuf[i]);

  2697. 			c->chrPixBuf[i]=NULL;

  2698. 		}

  2699. 		av_free(c->chrPixBuf);

  2700. 		c->chrPixBuf=NULL;

  2701. 	}

  2702. 

  2703. 	av_free(c->vLumFilter);

  2704. 	c->vLumFilter = NULL;

  2705. 	av_free(c->vChrFilter);

  2706. 	c->vChrFilter = NULL;

  2707. 	av_free(c->hLumFilter);

  2708. 	c->hLumFilter = NULL;

  2709. 	av_free(c->hChrFilter);

  2710. 	c->hChrFilter = NULL;

  2711. #ifdef HAVE_ALTIVEC

  2712. 	av_free(c->vYCoeffsBank);

  2713. 	c->vYCoeffsBank = NULL;

  2714. 	av_free(c->vCCoeffsBank);

  2715. 	c->vCCoeffsBank = NULL;

  2716. #endif

  2717. 

  2718. 	av_free(c->vLumFilterPos);

  2719. 	c->vLumFilterPos = NULL;

  2720. 	av_free(c->vChrFilterPos);

  2721. 	c->vChrFilterPos = NULL;

  2722. 	av_free(c->hLumFilterPos);

  2723. 	c->hLumFilterPos = NULL;

  2724. 	av_free(c->hChrFilterPos);

  2725. 	c->hChrFilterPos = NULL;

  2726. 

  2727. #if defined(ARCH_X86) || defined(ARCH_X86_64)

  2728. #ifdef MAP_ANONYMOUS

  2729. 	if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);

  2730. 	if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);

  2731. #else

  2732. 	av_free(c->funnyYCode);

  2733. 	av_free(c->funnyUVCode);

  2734. #endif

  2735. 	c->funnyYCode=NULL;

  2736. 	c->funnyUVCode=NULL;

  2737. #endif /* defined(ARCH_X86) || defined(ARCH_X86_64) */

  2738. 

  2739. 	av_free(c->lumMmx2Filter);

  2740. 	c->lumMmx2Filter=NULL;

  2741. 	av_free(c->chrMmx2Filter);

  2742. 	c->chrMmx2Filter=NULL;

  2743. 	av_free(c->lumMmx2FilterPos);

  2744. 	c->lumMmx2FilterPos=NULL;

  2745. 	av_free(c->chrMmx2FilterPos);

  2746. 	c->chrMmx2FilterPos=NULL;

  2747. 	av_free(c->yuvTable);

  2748. 	c->yuvTable=NULL;

  2749. 

  2750. 	av_free(c);

  2751. }