swscale.c

/*
    Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>

    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation; either version 2 of the License, or
    (at your option) any later version.

    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
*/

/*
  supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
  supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
  {BGR,RGB}{1,4,8,15,16} support dithering
  
  unscaled special converters (YV12=I420=IYUV, Y800=Y8)
  YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
  x -> x
  YUV9 -> YV12
  YUV9/YV12 -> Y800
  Y800 -> YUV9/YV12
  BGR24 -> BGR32 & RGB24 -> RGB32
  BGR32 -> BGR24 & RGB32 -> RGB24
  BGR15 -> BGR16
*/

/* 
tested special converters (most are tested actually but i didnt write it down ...)
 YV12 -> BGR16
 YV12 -> YV12
 BGR15 -> BGR16
 BGR16 -> BGR16
 YVU9 -> YV12

untested special converters
  YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
  YV12/I420 -> YV12/I420
  YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
  BGR24 -> BGR32 & RGB24 -> RGB32
  BGR32 -> BGR24 & RGB32 -> RGB24
  BGR24 -> YV12
*/

#include <inttypes.h>
#include <string.h>
#include <math.h>
#include <stdio.h>
#include "../config.h"
#include "../mangle.h"
#include <assert.h>
#ifdef HAVE_MALLOC_H
#include <malloc.h>
#else
#include <stdlib.h>
#endif
#ifdef HAVE_SYS_MMAN_H
#include <sys/mman.h>
#endif
#include "swscale.h"
#include "swscale_internal.h"
#include "../cpudetect.h"
#include "../bswap.h"
#include "../libvo/img_format.h"
#include "rgb2rgb.h"
#include "../libvo/fastmemcpy.h"

#undef MOVNTQ
#undef PAVGB

//#undef HAVE_MMX2
//#define HAVE_3DNOW
//#undef HAVE_MMX
//#undef ARCH_X86
//#define WORDS_BIGENDIAN
#define DITHER1XBPP

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

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

#ifdef MP_DEBUG
#define ASSERT(x) assert(x);
#else
#define ASSERT(x) ;
#endif

#ifdef M_PI
#define PI M_PI
#else
#define PI 3.14159265358979323846
#endif

//FIXME replace this with something faster
#define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
			|| (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
#define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
#define isGray(x)      ((x)==IMGFMT_Y800)
#define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
#define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
#define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
			|| (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
			|| (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
			|| (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
			|| (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
#define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
			|| (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
			|| isRGB(x) || isBGR(x)\
			|| (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
#define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))

#define RGB2YUV_SHIFT 16
#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))

extern const int32_t Inverse_Table_6_9[8][4];

/*
NOTES
Special versions: fast Y 1:1 scaling (no interpolation in y direction)

TODO
more intelligent missalignment avoidance for the horizontal scaler
write special vertical cubic upscale version
Optimize C code (yv12 / minmax)
add support for packed pixel yuv input & output
add support for Y8 output
optimize bgr24 & bgr32
add BGR4 output support
write special BGR->BGR scaler
*/

#define ABS(a) ((a) > 0 ? (a) : (-(a)))
#define MIN(a,b) ((a) > (b) ? (b) : (a))
#define MAX(a,b) ((a) < (b) ? (b) : (a))

#if defined(ARCH_X86) || defined(ARCH_X86_64)
static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;

static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;

static uint64_t __attribute__((aligned(8))) dither4[2]={
	0x0103010301030103LL,
	0x0200020002000200LL,};

static uint64_t __attribute__((aligned(8))) dither8[2]={
	0x0602060206020602LL,
	0x0004000400040004LL,};

static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;

static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;

#ifdef FAST_BGR2YV12
static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
#else
static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
#endif
static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
#endif

// clipping helper table for C implementations:
static unsigned char clip_table[768];

static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
		  
extern const uint8_t dither_2x2_4[2][8];
extern const uint8_t dither_2x2_8[2][8];
extern const uint8_t dither_8x8_32[8][8];
extern const uint8_t dither_8x8_73[8][8];
extern const uint8_t dither_8x8_220[8][8];

#if defined(ARCH_X86) || defined(ARCH_X86_64)
void in_asm_used_var_warning_killer()
{
 volatile int i= bF8+bFC+w10+
 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
 if(i) i=0;
}
#endif

static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
				    int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
				    uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
{
	//FIXME Optimize (just quickly writen not opti..)
	int i;
	for(i=0; i<dstW; i++)
	{
		int val=1<<18;
		int j;
		for(j=0; j<lumFilterSize; j++)
			val += lumSrc[j][i] * lumFilter[j];

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

	if(uDest != NULL)
		for(i=0; i<chrDstW; i++)
		{
			int u=1<<18;
			int v=1<<18;
			int j;
			for(j=0; j<chrFilterSize; j++)
			{
				u += chrSrc[j][i] * chrFilter[j];
				v += chrSrc[j][i + 2048] * chrFilter[j];
			}

			uDest[i]= MIN(MAX(u>>19, 0), 255);
			vDest[i]= MIN(MAX(v>>19, 0), 255);
		}
}


#define YSCALE_YUV_2_PACKEDX_C(type) \
		for(i=0; i<(dstW>>1); i++){\
			int j;\
			int Y1=1<<18;\
			int Y2=1<<18;\
			int U=1<<18;\
			int V=1<<18;\
			type *r, *b, *g;\
			const int i2= 2*i;\
			\
			for(j=0; j<lumFilterSize; j++)\
			{\
				Y1 += lumSrc[j][i2] * lumFilter[j];\
				Y2 += lumSrc[j][i2+1] * lumFilter[j];\
			}\
			for(j=0; j<chrFilterSize; j++)\
			{\
				U += chrSrc[j][i] * chrFilter[j];\
				V += chrSrc[j][i+2048] * chrFilter[j];\
			}\
			Y1>>=19;\
			Y2>>=19;\
			U >>=19;\
			V >>=19;\
			if((Y1|Y2|U|V)&256)\
			{\
				if(Y1>255)   Y1=255;\
				else if(Y1<0)Y1=0;\
				if(Y2>255)   Y2=255;\
				else if(Y2<0)Y2=0;\
				if(U>255)    U=255;\
				else if(U<0) U=0;\
				if(V>255)    V=255;\
				else if(V<0) V=0;\
			}
                        
#define YSCALE_YUV_2_RGBX_C(type) \
			YSCALE_YUV_2_PACKEDX_C(type)\
			r = c->table_rV[V];\
			g = c->table_gU[U] + c->table_gV[V];\
			b = c->table_bU[U];\

#define YSCALE_YUV_2_PACKED2_C \
		for(i=0; i<(dstW>>1); i++){\
			const int i2= 2*i;\
			int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
			int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
			int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
			int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\

#define YSCALE_YUV_2_RGB2_C(type) \
			YSCALE_YUV_2_PACKED2_C\
			type *r, *b, *g;\
			r = c->table_rV[V];\
			g = c->table_gU[U] + c->table_gV[V];\
			b = c->table_bU[U];\

#define YSCALE_YUV_2_PACKED1_C \
		for(i=0; i<(dstW>>1); i++){\
			const int i2= 2*i;\
			int Y1= buf0[i2  ]>>7;\
			int Y2= buf0[i2+1]>>7;\
			int U= (uvbuf1[i     ])>>7;\
			int V= (uvbuf1[i+2048])>>7;\

#define YSCALE_YUV_2_RGB1_C(type) \
			YSCALE_YUV_2_PACKED1_C\
			type *r, *b, *g;\
			r = c->table_rV[V];\
			g = c->table_gU[U] + c->table_gV[V];\
			b = c->table_bU[U];\

#define YSCALE_YUV_2_PACKED1B_C \
		for(i=0; i<(dstW>>1); i++){\
			const int i2= 2*i;\
			int Y1= buf0[i2  ]>>7;\
			int Y2= buf0[i2+1]>>7;\
			int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
			int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\

#define YSCALE_YUV_2_RGB1B_C(type) \
			YSCALE_YUV_2_PACKED1B_C\
			type *r, *b, *g;\
			r = c->table_rV[V];\
			g = c->table_gU[U] + c->table_gV[V];\
			b = c->table_bU[U];\

#define YSCALE_YUV_2_ANYRGB_C(func, func2)\
	switch(c->dstFormat)\
	{\
	case IMGFMT_BGR32:\
	case IMGFMT_RGB32:\
		func(uint32_t)\
			((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
			((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
		}		\
		break;\
	case IMGFMT_RGB24:\
		func(uint8_t)\
			((uint8_t*)dest)[0]= r[Y1];\
			((uint8_t*)dest)[1]= g[Y1];\
			((uint8_t*)dest)[2]= b[Y1];\
			((uint8_t*)dest)[3]= r[Y2];\
			((uint8_t*)dest)[4]= g[Y2];\
			((uint8_t*)dest)[5]= b[Y2];\
			dest+=6;\
		}\
		break;\
	case IMGFMT_BGR24:\
		func(uint8_t)\
			((uint8_t*)dest)[0]= b[Y1];\
			((uint8_t*)dest)[1]= g[Y1];\
			((uint8_t*)dest)[2]= r[Y1];\
			((uint8_t*)dest)[3]= b[Y2];\
			((uint8_t*)dest)[4]= g[Y2];\
			((uint8_t*)dest)[5]= r[Y2];\
			dest+=6;\
		}\
		break;\
	case IMGFMT_RGB16:\
	case IMGFMT_BGR16:\
		{\
			const int dr1= dither_2x2_8[y&1    ][0];\
			const int dg1= dither_2x2_4[y&1    ][0];\
			const int db1= dither_2x2_8[(y&1)^1][0];\
			const int dr2= dither_2x2_8[y&1    ][1];\
			const int dg2= dither_2x2_4[y&1    ][1];\
			const int db2= dither_2x2_8[(y&1)^1][1];\
			func(uint16_t)\
				((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
				((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
			}\
		}\
		break;\
	case IMGFMT_RGB15:\
	case IMGFMT_BGR15:\
		{\
			const int dr1= dither_2x2_8[y&1    ][0];\
			const int dg1= dither_2x2_8[y&1    ][1];\
			const int db1= dither_2x2_8[(y&1)^1][0];\
			const int dr2= dither_2x2_8[y&1    ][1];\
			const int dg2= dither_2x2_8[y&1    ][0];\
			const int db2= dither_2x2_8[(y&1)^1][1];\
			func(uint16_t)\
				((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
				((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
			}\
		}\
		break;\
	case IMGFMT_RGB8:\
	case IMGFMT_BGR8:\
		{\
			const uint8_t * const d64= dither_8x8_73[y&7];\
			const uint8_t * const d32= dither_8x8_32[y&7];\
			func(uint8_t)\
				((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
				((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
			}\
		}\
		break;\
	case IMGFMT_RGB4:\
	case IMGFMT_BGR4:\
		{\
			const uint8_t * const d64= dither_8x8_73 [y&7];\
			const uint8_t * const d128=dither_8x8_220[y&7];\
			func(uint8_t)\
				((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
				                 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
			}\
		}\
		break;\
	case IMGFMT_RG4B:\
	case IMGFMT_BG4B:\
		{\
			const uint8_t * const d64= dither_8x8_73 [y&7];\
			const uint8_t * const d128=dither_8x8_220[y&7];\
			func(uint8_t)\
				((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
				((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
			}\
		}\
		break;\
	case IMGFMT_RGB1:\
	case IMGFMT_BGR1:\
		{\
			const uint8_t * const d128=dither_8x8_220[y&7];\
			uint8_t *g= c->table_gU[128] + c->table_gV[128];\
			for(i=0; i<dstW-7; i+=8){\
				int acc;\
				acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
				acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
				acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
				acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
				acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
				acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
				acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
				acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
				((uint8_t*)dest)[0]= acc;\
				dest++;\
			}\
\
/*\
((uint8_t*)dest)-= dstW>>4;\
{\
			int acc=0;\
			int left=0;\
			static int top[1024];\
			static int last_new[1024][1024];\
			static int last_in3[1024][1024];\
			static int drift[1024][1024];\
			int topLeft=0;\
			int shift=0;\
			int count=0;\
			const uint8_t * const d128=dither_8x8_220[y&7];\
			int error_new=0;\
			int error_in3=0;\
			int f=0;\
			\
			for(i=dstW>>1; i<dstW; i++){\
				int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
				int in2 = (76309 * (in - 16) + 32768) >> 16;\
				int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
				int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
					+ (last_new[y][i] - in3)*f/256;\
				int new= old> 128 ? 255 : 0;\
\
				error_new+= ABS(last_new[y][i] - new);\
				error_in3+= ABS(last_in3[y][i] - in3);\
				f= error_new - error_in3*4;\
				if(f<0) f=0;\
				if(f>256) f=256;\
\
				topLeft= top[i];\
				left= top[i]= old - new;\
				last_new[y][i]= new;\
				last_in3[y][i]= in3;\
\
				acc+= acc + (new&1);\
				if((i&7)==6){\
					((uint8_t*)dest)[0]= acc;\
					((uint8_t*)dest)++;\
				}\
			}\
}\
*/\
		}\
		break;\
	case IMGFMT_YUY2:\
		func2\
			((uint8_t*)dest)[2*i2+0]= Y1;\
			((uint8_t*)dest)[2*i2+1]= U;\
			((uint8_t*)dest)[2*i2+2]= Y2;\
			((uint8_t*)dest)[2*i2+3]= V;\
		}		\
		break;\
	case IMGFMT_UYVY:\
		func2\
			((uint8_t*)dest)[2*i2+0]= U;\
			((uint8_t*)dest)[2*i2+1]= Y1;\
			((uint8_t*)dest)[2*i2+2]= V;\
			((uint8_t*)dest)[2*i2+3]= Y2;\
		}		\
		break;\
	}\


static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
				    int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
				    uint8_t *dest, int dstW, int y)
{
	int i;
	switch(c->dstFormat)
	{
	case IMGFMT_RGB32:
	case IMGFMT_BGR32:
		YSCALE_YUV_2_RGBX_C(uint32_t)
			((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
			((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
		}
		break;
	case IMGFMT_RGB24:
		YSCALE_YUV_2_RGBX_C(uint8_t)
			((uint8_t*)dest)[0]= r[Y1];
			((uint8_t*)dest)[1]= g[Y1];
			((uint8_t*)dest)[2]= b[Y1];
			((uint8_t*)dest)[3]= r[Y2];
			((uint8_t*)dest)[4]= g[Y2];
			((uint8_t*)dest)[5]= b[Y2];
			dest+=6;
		}
		break;
	case IMGFMT_BGR24:
		YSCALE_YUV_2_RGBX_C(uint8_t)
			((uint8_t*)dest)[0]= b[Y1];
			((uint8_t*)dest)[1]= g[Y1];
			((uint8_t*)dest)[2]= r[Y1];
			((uint8_t*)dest)[3]= b[Y2];
			((uint8_t*)dest)[4]= g[Y2];
			((uint8_t*)dest)[5]= r[Y2];
			dest+=6;
		}
		break;
	case IMGFMT_RGB16:
	case IMGFMT_BGR16:
		{
			const int dr1= dither_2x2_8[y&1    ][0];
			const int dg1= dither_2x2_4[y&1    ][0];
			const int db1= dither_2x2_8[(y&1)^1][0];
			const int dr2= dither_2x2_8[y&1    ][1];
			const int dg2= dither_2x2_4[y&1    ][1];
			const int db2= dither_2x2_8[(y&1)^1][1];
			YSCALE_YUV_2_RGBX_C(uint16_t)
				((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
				((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
			}
		}
		break;
	case IMGFMT_RGB15:
	case IMGFMT_BGR15:
		{
			const int dr1= dither_2x2_8[y&1    ][0];
			const int dg1= dither_2x2_8[y&1    ][1];
			const int db1= dither_2x2_8[(y&1)^1][0];
			const int dr2= dither_2x2_8[y&1    ][1];
			const int dg2= dither_2x2_8[y&1    ][0];
			const int db2= dither_2x2_8[(y&1)^1][1];
			YSCALE_YUV_2_RGBX_C(uint16_t)
				((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
				((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
			}
		}
		break;
	case IMGFMT_RGB8:
	case IMGFMT_BGR8:
		{
			const uint8_t * const d64= dither_8x8_73[y&7];
			const uint8_t * const d32= dither_8x8_32[y&7];
			YSCALE_YUV_2_RGBX_C(uint8_t)
				((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
				((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
			}
		}
		break;
	case IMGFMT_RGB4:
	case IMGFMT_BGR4:
		{
			const uint8_t * const d64= dither_8x8_73 [y&7];
			const uint8_t * const d128=dither_8x8_220[y&7];
			YSCALE_YUV_2_RGBX_C(uint8_t)
				((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
				                  +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
			}
		}
		break;
	case IMGFMT_RG4B:
	case IMGFMT_BG4B:
		{
			const uint8_t * const d64= dither_8x8_73 [y&7];
			const uint8_t * const d128=dither_8x8_220[y&7];
			YSCALE_YUV_2_RGBX_C(uint8_t)
				((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
				((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
			}
		}
		break;
	case IMGFMT_RGB1:
	case IMGFMT_BGR1:
		{
			const uint8_t * const d128=dither_8x8_220[y&7];
			uint8_t *g= c->table_gU[128] + c->table_gV[128];
			int acc=0;
			for(i=0; i<dstW-1; i+=2){
				int j;
				int Y1=1<<18;
				int Y2=1<<18;

				for(j=0; j<lumFilterSize; j++)
				{
					Y1 += lumSrc[j][i] * lumFilter[j];
					Y2 += lumSrc[j][i+1] * lumFilter[j];
				}
				Y1>>=19;
				Y2>>=19;
				if((Y1|Y2)&256)
				{
					if(Y1>255)   Y1=255;
					else if(Y1<0)Y1=0;
					if(Y2>255)   Y2=255;
					else if(Y2<0)Y2=0;
				}
				acc+= acc + g[Y1+d128[(i+0)&7]];
				acc+= acc + g[Y2+d128[(i+1)&7]];
				if((i&7)==6){
					((uint8_t*)dest)[0]= acc;
					dest++;
				}
			}
		}
		break;
	case IMGFMT_YUY2:
		YSCALE_YUV_2_PACKEDX_C(void)
			((uint8_t*)dest)[2*i2+0]= Y1;
			((uint8_t*)dest)[2*i2+1]= U;
			((uint8_t*)dest)[2*i2+2]= Y2;
			((uint8_t*)dest)[2*i2+3]= V;
		}
                break;
	case IMGFMT_UYVY:
		YSCALE_YUV_2_PACKEDX_C(void)
			((uint8_t*)dest)[2*i2+0]= U;
			((uint8_t*)dest)[2*i2+1]= Y1;
			((uint8_t*)dest)[2*i2+2]= V;
			((uint8_t*)dest)[2*i2+3]= Y2;
		}
                break;
	}
}


//Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
//Plain C versions
#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
#define COMPILE_C
#endif

#ifdef ARCH_POWERPC
#ifdef HAVE_ALTIVEC
#define COMPILE_ALTIVEC
#endif //HAVE_ALTIVEC
#endif //ARCH_POWERPC

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

#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
#define COMPILE_MMX
#endif

#if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
#define COMPILE_MMX2
#endif

#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
#define COMPILE_3DNOW
#endif
#endif //ARCH_X86 || ARCH_X86_64

#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW

#ifdef COMPILE_C
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#undef HAVE_ALTIVEC
#define RENAME(a) a ## _C
#include "swscale_template.c"
#endif

#ifdef ARCH_POWERPC
#ifdef COMPILE_ALTIVEC
#undef RENAME
#define HAVE_ALTIVEC
#define RENAME(a) a ## _altivec
#include "swscale_template.c"
#endif
#endif //ARCH_POWERPC

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

//X86 versions
/*
#undef RENAME
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#define ARCH_X86
#define RENAME(a) a ## _X86
#include "swscale_template.c"
*/
//MMX versions
#ifdef COMPILE_MMX
#undef RENAME
#define HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#define RENAME(a) a ## _MMX
#include "swscale_template.c"
#endif

//MMX2 versions
#ifdef COMPILE_MMX2
#undef RENAME
#define HAVE_MMX
#define HAVE_MMX2
#undef HAVE_3DNOW
#define RENAME(a) a ## _MMX2
#include "swscale_template.c"
#endif

//3DNOW versions
#ifdef COMPILE_3DNOW
#undef RENAME
#define HAVE_MMX
#undef HAVE_MMX2
#define HAVE_3DNOW
#define RENAME(a) a ## _3DNow
#include "swscale_template.c"
#endif

#endif //ARCH_X86 || ARCH_X86_64

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

static double getSplineCoeff(double a, double b, double c, double d, double dist)
{
//	printf("%f %f %f %f %f\n", a,b,c,d,dist);
	if(dist<=1.0) 	return ((d*dist + c)*dist + b)*dist +a;
	else		return getSplineCoeff(	0.0, 
						 b+ 2.0*c + 3.0*d,
						        c + 3.0*d,
						-b- 3.0*c - 6.0*d,
						dist-1.0);
}

static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
			      int srcW, int dstW, int filterAlign, int one, int flags,
			      SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
{
	int i;
	int filterSize;
	int filter2Size;
	int minFilterSize;
	double *filter=NULL;
	double *filter2=NULL;
#if defined(ARCH_X86) || defined(ARCH_X86_64)
	if(flags & SWS_CPU_CAPS_MMX)
		asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
#endif

	// Note the +1 is for the MMXscaler which reads over the end
	*filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));

	if(ABS(xInc - 0x10000) <10) // unscaled
	{
		int i;
		filterSize= 1;
		filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
		for(i=0; i<dstW*filterSize; i++) filter[i]=0;

		for(i=0; i<dstW; i++)
		{
			filter[i*filterSize]=1;
			(*filterPos)[i]=i;
		}

	}
	else if(flags&SWS_POINT) // lame looking point sampling mode
	{
		int i;
		int xDstInSrc;
		filterSize= 1;
		filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
		
		xDstInSrc= xInc/2 - 0x8000;
		for(i=0; i<dstW; i++)
		{
			int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;

			(*filterPos)[i]= xx;
			filter[i]= 1.0;
			xDstInSrc+= xInc;
		}
	}
	else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
	{
		int i;
		int xDstInSrc;
		if     (flags&SWS_BICUBIC) filterSize= 4;
		else if(flags&SWS_X      ) filterSize= 4;
		else			   filterSize= 2; // SWS_BILINEAR / SWS_AREA 
		filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);

		xDstInSrc= xInc/2 - 0x8000;
		for(i=0; i<dstW; i++)
		{
			int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
			int j;

			(*filterPos)[i]= xx;
				//Bilinear upscale / linear interpolate / Area averaging
				for(j=0; j<filterSize; j++)
				{
					double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
					double coeff= 1.0 - d;
					if(coeff<0) coeff=0;
					filter[i*filterSize + j]= coeff;
					xx++;
				}
			xDstInSrc+= xInc;
		}
	}
	else
	{
		double xDstInSrc;
		double sizeFactor, filterSizeInSrc;
		const double xInc1= (double)xInc / (double)(1<<16);

		if     (flags&SWS_BICUBIC)	sizeFactor= 4.0;
		else if(flags&SWS_X)		sizeFactor= 8.0;
		else if(flags&SWS_AREA)		sizeFactor= 1.0; //downscale only, for upscale it is bilinear
		else if(flags&SWS_GAUSS)	sizeFactor= 8.0;   // infinite ;)
		else if(flags&SWS_LANCZOS)	sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
		else if(flags&SWS_SINC)		sizeFactor= 20.0; // infinite ;)
		else if(flags&SWS_SPLINE)	sizeFactor= 20.0;  // infinite ;)
		else if(flags&SWS_BILINEAR)	sizeFactor= 2.0;
		else {
			sizeFactor= 0.0; //GCC warning killer
			ASSERT(0)
		}
		
		if(xInc1 <= 1.0)	filterSizeInSrc= sizeFactor; // upscale
		else			filterSizeInSrc= sizeFactor*srcW / (double)dstW;

		filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
		if(filterSize > srcW-2) filterSize=srcW-2;

		filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);

		xDstInSrc= xInc1 / 2.0 - 0.5;
		for(i=0; i<dstW; i++)
		{
			int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
			int j;
			(*filterPos)[i]= xx;
			for(j=0; j<filterSize; j++)
			{
				double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
				double coeff;
				if(flags & SWS_BICUBIC)
				{
					double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
					double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;

					if(d<1.0) 
						coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
					else if(d<2.0)
						coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
					else
						coeff=0.0;
				}
/*				else if(flags & SWS_X)
				{
					double p= param ? param*0.01 : 0.3;
					coeff = d ? sin(d*PI)/(d*PI) : 1.0;
					coeff*= pow(2.0, - p*d*d);
				}*/
				else if(flags & SWS_X)
				{
					double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
					
					if(d<1.0)
						coeff = cos(d*PI);
					else
						coeff=-1.0;
					if(coeff<0.0) 	coeff= -pow(-coeff, A);
					else		coeff=  pow( coeff, A);
					coeff= coeff*0.5 + 0.5;
				}
				else if(flags & SWS_AREA)
				{
					double srcPixelSize= 1.0/xInc1;
					if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
					else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
					else coeff=0.0;
				}
				else if(flags & SWS_GAUSS)
				{
					double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
					coeff = pow(2.0, - p*d*d);
				}
				else if(flags & SWS_SINC)
				{
					coeff = d ? sin(d*PI)/(d*PI) : 1.0;
				}
				else if(flags & SWS_LANCZOS)
				{
					double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 
					coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
					if(d>p) coeff=0;
				}
				else if(flags & SWS_BILINEAR)
				{
					coeff= 1.0 - d;
					if(coeff<0) coeff=0;
				}
				else if(flags & SWS_SPLINE)
				{
					double p=-2.196152422706632;
					coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
				}
				else {
					coeff= 0.0; //GCC warning killer
					ASSERT(0)
				}

				filter[i*filterSize + j]= coeff;
				xx++;
			}
			xDstInSrc+= xInc1;
		}
	}

	/* apply src & dst Filter to filter -> filter2
	   free(filter);
	*/
	ASSERT(filterSize>0)
	filter2Size= filterSize;
	if(srcFilter) filter2Size+= srcFilter->length - 1;
	if(dstFilter) filter2Size+= dstFilter->length - 1;
	ASSERT(filter2Size>0)
	filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));

	for(i=0; i<dstW; i++)
	{
		int j;
		SwsVector scaleFilter;
		SwsVector *outVec;

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

		if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
		else	      outVec= &scaleFilter;

		ASSERT(outVec->length == filter2Size)
		//FIXME dstFilter

		for(j=0; j<outVec->length; j++)
		{
			filter2[i*filter2Size + j]= outVec->coeff[j];
		}

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

		if(outVec != &scaleFilter) sws_freeVec(outVec);
	}