vertical cubic/linear scaling

prints some info (MMX, dstbpp, ...)
mainloop rewritten


git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@3345 b3059339-0415-0410-9bf9-f77b7e298cf2

3 files changed, 994 insertions, 327 deletions

diff --git a/postproc/swscale.c b/postproc/swscale.c
index 0383c37fc5..138f35891a 100644
--- a/postproc/swscale.c
+++ b/postproc/swscale.c
@@ -8,8 +8,11 @@
 #include <inttypes.h>
 #include <string.h>
 #include <math.h>
-//#include <stdio.h> //FOR DEBUG ONLY
+#include <stdio.h>
 #include "../config.h"
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif
 #include "swscale.h"
 #include "../cpudetect.h"
 #undef MOVNTQ
@@ -24,22 +27,28 @@ int fullUVIpol=0;
 int allwaysIpol=0;
 
 #define RET 0xC3 //near return opcode
+
+//#define ASSERT(x) if(!(x)) { printf("ASSERT " #x " failed\n"); *((int*)0)=0; }
+#define ASSERT(x) ;
+
+
 /*
 NOTES
 
 known BUGS with known cause (no bugreports please!, but patches are welcome :) )
 horizontal MMX2 scaler reads 1-7 samples too much (might cause a sig11)
 
-Supported output formats BGR15 BGR16 BGR24 BGR32
+Supported output formats BGR15 BGR16 BGR24 BGR32, YV12
 BGR15 & BGR16 MMX verions support dithering
 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
 
 TODO
 more intelligent missalignment avoidance for the horizontal scaler
-bicubic scaler
 dither in C
 change the distance of the u & v buffer
-how to differenciate between x86 an C at runtime ?! (using C for now)
+Move static / global vars into a struct so multiple scalers can be used
+write special vertical cubic upscale version
+Optimize C code (yv12 / minmax)
 */
 
 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
@@ -94,21 +103,35 @@ static uint64_t __attribute__((aligned(8))) temp0;
 static uint64_t __attribute__((aligned(8))) asm_yalpha1;
 static uint64_t __attribute__((aligned(8))) asm_uvalpha1;
 
-// temporary storage for 4 yuv lines:
-// 16bit for now (mmx likes it more compact)
-static uint16_t __attribute__((aligned(8))) pix_buf_y[4][2048];
-static uint16_t __attribute__((aligned(8))) pix_buf_uv[2][2048*2];
+static int16_t __attribute__((aligned(8))) *lumPixBuf[2000];
+static int16_t __attribute__((aligned(8))) *chrPixBuf[2000];
 static int16_t __attribute__((aligned(8))) hLumFilter[8000];
 static int16_t __attribute__((aligned(8))) hLumFilterPos[2000];
 static int16_t __attribute__((aligned(8))) hChrFilter[8000];
 static int16_t __attribute__((aligned(8))) hChrFilterPos[2000];
+static int16_t __attribute__((aligned(8))) vLumFilter[8000];
+static int16_t __attribute__((aligned(8))) vLumFilterPos[2000];
+static int16_t __attribute__((aligned(8))) vChrFilter[8000];
+static int16_t __attribute__((aligned(8))) vChrFilterPos[2000];
+
+// Contain simply the values from v(Lum|Chr)Filter just nicely packed for mmx
+//FIXME these are very likely too small / 8000 caused problems with 480x480
+static int16_t __attribute__((aligned(8))) lumMmxFilter[16000];
+static int16_t __attribute__((aligned(8))) chrMmxFilter[16000];
 #else
-static uint16_t pix_buf_y[4][2048];
-static uint16_t pix_buf_uv[2][2048*2];
+static int16_t *lumPixBuf[2000];
+static int16_t *chrPixBuf[2000];
 static int16_t hLumFilter[8000];
 static int16_t hLumFilterPos[2000];
 static int16_t hChrFilter[8000];
 static int16_t hChrFilterPos[2000];
+static int16_t vLumFilter[8000];
+static int16_t vLumFilterPos[2000];
+static int16_t vChrFilter[8000];
+static int16_t vChrFilterPos[2000];
+//FIXME just dummy vars
+static int16_t lumMmxFilter[1];
+static int16_t chrMmxFilter[1];
 #endif
 
 // clipping helper table for C implementations:
@@ -127,9 +150,19 @@ static    int yuvtab_3343[256];
 static    int yuvtab_0c92[256];
 static    int yuvtab_1a1e[256];
 static    int yuvtab_40cf[256];
-
-static int hLumFilterSize;
-static int hChrFilterSize;
+// Needed for cubic scaler to catch overflows
+static    int clip_yuvtab_2568[768];
+static    int clip_yuvtab_3343[768];
+static    int clip_yuvtab_0c92[768];
+static    int clip_yuvtab_1a1e[768];
+static    int clip_yuvtab_40cf[768];
+
+static int hLumFilterSize=0;
+static int hChrFilterSize=0;
+static int vLumFilterSize=0;
+static int vChrFilterSize=0;
+static int vLumBufSize=0;
+static int vChrBufSize=0;
 
 int sws_flags=0;
 
@@ -274,15 +307,14 @@ void SwScale_YV12slice(unsigned char* srcptr[],int stride[], int srcSliceY ,
 void SwScale_Init(){
     // generating tables:
     int i;
-    for(i=0;i<256;i++){
-        clip_table[i]=0;
-        clip_table[i+256]=i;
-        clip_table[i+512]=255;
-	yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
-	yuvtab_3343[i]=0x3343*(i-128);
-	yuvtab_0c92[i]=-0x0c92*(i-128);
-	yuvtab_1a1e[i]=-0x1a1e*(i-128);
-	yuvtab_40cf[i]=0x40cf*(i-128);
+    for(i=0; i<768; i++){
+	int c= MIN(MAX(i-256, 0), 255);
+	clip_table[i]=c;
+	yuvtab_2568[c]= clip_yuvtab_2568[i]=(0x2568*(c-16))+(256<<13);
+	yuvtab_3343[c]= clip_yuvtab_3343[i]=0x3343*(c-128);
+	yuvtab_0c92[c]= clip_yuvtab_0c92[i]=-0x0c92*(c-128);
+	yuvtab_1a1e[c]= clip_yuvtab_1a1e[i]=-0x1a1e*(c-128);
+	yuvtab_40cf[c]= clip_yuvtab_40cf[i]=0x40cf*(c-128);
     }
 
     for(i=0; i<768; i++)
@@ -295,5 +327,6 @@ void SwScale_Init(){
 	clip_table15g[i]= (v<<2)&0x03E0;
 	clip_table15r[i]= (v<<7)&0x7C00;
     }
+
 }
 
diff --git a/postproc/swscale.h b/postproc/swscale.h
index 2c75a9007c..2757ba5f20 100644
--- a/postproc/swscale.h
+++ b/postproc/swscale.h
@@ -11,4 +11,4 @@ void SwScale_YV12slice(unsigned char* srcptr[],int stride[], int srcSliceY,
 			     int srcSliceH, uint8_t* dstptr[], int dststride, int dstbpp,
 			     int srcW, int srcH, int dstW, int dstH);
 // generating tables
-void SwScale_Init();
+void SwScale_Init();
\ No newline at end of file
diff --git a/postproc/swscale_template.c b/postproc/swscale_template.c
index c7bb86e009..be0962495e 100644
--- a/postproc/swscale_template.c
+++ b/postproc/swscale_template.c
@@ -5,11 +5,6 @@
 // current version mostly by Michael Niedermayer (michaelni@gmx.at)
 // the parts written by michael are under GNU GPL
 
-/* TODO
-Move static / global vars into a struct so multiple scalers can be used
-write vertical cubic upscale / linear downscale stuff
-*/
-
 #undef MOVNTQ
 #undef PAVGB
 #undef PREFETCH
@@ -53,6 +48,129 @@ write vertical cubic upscale / linear downscale stuff
 #define MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
 #endif
 
+
+#define YSCALEYUV2YV12X(x) \
+			"xorl %%eax, %%eax		\n\t"\
+			"pxor %%mm3, %%mm3		\n\t"\
+			"pxor %%mm4, %%mm4		\n\t"\
+			"movl %0, %%edx			\n\t"\
+			".balign 16			\n\t" /* FIXME Unroll? */\
+			"1:				\n\t"\
+			"movl (%1, %%edx, 4), %%esi	\n\t"\
+			"movq (%2, %%edx, 8), %%mm0	\n\t" /* filterCoeff */\
+			"movq " #x "(%%esi, %%eax, 2), %%mm2	\n\t" /* srcData */\
+			"movq 8+" #x "(%%esi, %%eax, 2), %%mm5	\n\t" /* srcData */\
+			"pmulhw %%mm0, %%mm2		\n\t"\
+			"pmulhw %%mm0, %%mm5		\n\t"\
+			"paddw %%mm2, %%mm3		\n\t"\
+			"paddw %%mm5, %%mm4		\n\t"\
+			"addl $1, %%edx			\n\t"\
+			" jnz 1b			\n\t"\
+			"psraw $3, %%mm3		\n\t"\
+			"psraw $3, %%mm4		\n\t"\
+			"packuswb %%mm4, %%mm3		\n\t"\
+			MOVNTQ(%%mm3, (%3, %%eax))\
+			"addl $8, %%eax			\n\t"\
+			"cmpl %4, %%eax			\n\t"\
+			"pxor %%mm3, %%mm3		\n\t"\
+			"pxor %%mm4, %%mm4		\n\t"\
+			"movl %0, %%edx			\n\t"\
+			"jb 1b				\n\t"
+
+#define YSCALEYUV2YV121 \
+			"movl %2, %%eax			\n\t"\
+			".balign 16			\n\t" /* FIXME Unroll? */\
+			"1:				\n\t"\
+			"movq (%0, %%eax, 2), %%mm0	\n\t"\
+			"movq 8(%0, %%eax, 2), %%mm1	\n\t"\
+			"psraw $7, %%mm0		\n\t"\
+			"psraw $7, %%mm1		\n\t"\
+			"packuswb %%mm1, %%mm0		\n\t"\
+			MOVNTQ(%%mm0, (%1, %%eax))\
+			"addl $8, %%eax			\n\t"\
+			"jnc 1b				\n\t"
+
+/*
+			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
+			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
+			   "r" (dest), "m" (dstW),
+			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
+			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
+*/
+#define YSCALEYUV2RGBX \
+		"xorl %%eax, %%eax		\n\t"\
+		".balign 16			\n\t"\
+		"1:				\n\t"\
+		"movl %1, %%edx			\n\t" /* -chrFilterSize */\
+		"movl %3, %%ebx			\n\t" /* chrMmxFilter+lumFilterSize */\
+		"movl %7, %%ecx			\n\t" /* chrSrc+lumFilterSize */\
+		"pxor %%mm3, %%mm3		\n\t"\
+		"pxor %%mm4, %%mm4		\n\t"\
+		"2:				\n\t"\
+		"movl (%%ecx, %%edx, 4), %%esi	\n\t"\
+		"movq (%%ebx, %%edx, 8), %%mm0	\n\t" /* filterCoeff */\
+		"movq (%%esi, %%eax), %%mm2	\n\t" /* UsrcData */\
+		"movq 4096(%%esi, %%eax), %%mm5	\n\t" /* VsrcData */\
+		"pmulhw %%mm0, %%mm2		\n\t"\
+		"pmulhw %%mm0, %%mm5		\n\t"\
+		"paddw %%mm2, %%mm3		\n\t"\
+		"paddw %%mm5, %%mm4		\n\t"\
+		"addl $1, %%edx			\n\t"\
+		" jnz 2b			\n\t"\
+\
+		"movl %0, %%edx			\n\t" /* -lumFilterSize */\
+		"movl %2, %%ebx			\n\t" /* lumMmxFilter+lumFilterSize */\
+		"movl %6, %%ecx			\n\t" /* lumSrc+lumFilterSize */\
+		"pxor %%mm1, %%mm1		\n\t"\
+		"pxor %%mm7, %%mm7		\n\t"\
+		"2:				\n\t"\
+		"movl (%%ecx, %%edx, 4), %%esi	\n\t"\
+		"movq (%%ebx, %%edx, 8), %%mm0	\n\t" /* filterCoeff */\
+		"movq (%%esi, %%eax, 2), %%mm2	\n\t" /* Y1srcData */\
+		"movq 8(%%esi, %%eax, 2), %%mm5	\n\t" /* Y2srcData */\
+		"pmulhw %%mm0, %%mm2		\n\t"\
+		"pmulhw %%mm0, %%mm5		\n\t"\
+		"paddw %%mm2, %%mm1		\n\t"\
+		"paddw %%mm5, %%mm7		\n\t"\
+		"addl $1, %%edx			\n\t"\
+		" jnz 2b			\n\t"\
+\
+		"psubw w400, %%mm3		\n\t" /* (U-128)8*/\
+		"psubw w400, %%mm4		\n\t" /* (V-128)8*/\
+		"movq %%mm3, %%mm2		\n\t" /* (U-128)8*/\
+		"movq %%mm4, %%mm5		\n\t" /* (V-128)8*/\
+		"pmulhw ugCoeff, %%mm3		\n\t"\
+		"pmulhw vgCoeff, %%mm4		\n\t"\
+	/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
+		"pmulhw ubCoeff, %%mm2		\n\t"\
+		"pmulhw vrCoeff, %%mm5		\n\t"\
+		"psubw w80, %%mm1		\n\t" /* 8(Y-16)*/\
+		"psubw w80, %%mm7		\n\t" /* 8(Y-16)*/\
+		"pmulhw yCoeff, %%mm1		\n\t"\
+		"pmulhw yCoeff, %%mm7		\n\t"\
+	/* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
+		"paddw %%mm3, %%mm4		\n\t"\
+		"movq %%mm2, %%mm0		\n\t"\
+		"movq %%mm5, %%mm6		\n\t"\
+		"movq %%mm4, %%mm3		\n\t"\
+		"punpcklwd %%mm2, %%mm2		\n\t"\
+		"punpcklwd %%mm5, %%mm5		\n\t"\
+		"punpcklwd %%mm4, %%mm4		\n\t"\
+		"paddw %%mm1, %%mm2		\n\t"\
+		"paddw %%mm1, %%mm5		\n\t"\
+		"paddw %%mm1, %%mm4		\n\t"\
+		"punpckhwd %%mm0, %%mm0		\n\t"\
+		"punpckhwd %%mm6, %%mm6		\n\t"\
+		"punpckhwd %%mm3, %%mm3		\n\t"\
+		"paddw %%mm7, %%mm0		\n\t"\
+		"paddw %%mm7, %%mm6		\n\t"\
+		"paddw %%mm7, %%mm3		\n\t"\
+		/* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
+		"packuswb %%mm0, %%mm2		\n\t"\
+		"packuswb %%mm6, %%mm5		\n\t"\
+		"packuswb %%mm3, %%mm4		\n\t"\
+		"pxor %%mm7, %%mm7		\n\t"
+
 #define FULL_YSCALEYUV2RGB \
 		"pxor %%mm7, %%mm7		\n\t"\
 		"movd %6, %%mm6			\n\t" /*yalpha1*/\
@@ -238,8 +356,8 @@ write vertical cubic upscale / linear downscale stuff
 		"movq 4096(%3, %%eax), %%mm4	\n\t" /* uvbuf1[eax+2048]*/\
 		"paddw %%mm2, %%mm3		\n\t" /* uvbuf0[eax] + uvbuf1[eax]*/\
 		"paddw %%mm5, %%mm4		\n\t" /* uvbuf0[eax+2048] + uvbuf1[eax+2048]*/\
-		"psrlw $5, %%mm3		\n\t"\
-		"psrlw $5, %%mm4		\n\t"\
+		"psrlw $5, %%mm3		\n\t" /*FIXME might overflow*/\
+		"psrlw $5, %%mm4		\n\t" /*FIXME might overflow*/\
 		"psubw w400, %%mm3		\n\t" /* (U-128)8*/\
 		"psubw w400, %%mm4		\n\t" /* (V-128)8*/\
 		"movq %%mm3, %%mm2		\n\t" /* (U-128)8*/\
@@ -524,36 +642,366 @@ write vertical cubic upscale / linear downscale stuff
 #define WRITEBGR24 WRITEBGR24MMX
 #endif
 
-static inline void RENAME(yuv2yuv)(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
-			   uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int yalpha, int uvalpha)
+static inline void RENAME(yuv2yuvX)(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,
+				    int16_t * lumMmxFilter, int16_t * chrMmxFilter)
 {
-	int yalpha1=yalpha^4095;
-	int uvalpha1=uvalpha^4095;
+#ifdef HAVE_MMX
+	if(uDest != NULL)
+	{
+		asm volatile(
+				YSCALEYUV2YV12X(0)
+				:: "m" (-chrFilterSize), "r" (chrSrc+chrFilterSize),
+				"r" (chrMmxFilter+chrFilterSize*4), "r" (uDest), "m" (dstW>>1)
+				: "%eax", "%edx", "%esi"
+			);
+
+		asm volatile(
+				YSCALEYUV2YV12X(4096)
+				:: "m" (-chrFilterSize), "r" (chrSrc+chrFilterSize),
+				"r" (chrMmxFilter+chrFilterSize*4), "r" (vDest), "m" (dstW>>1)
+				: "%eax", "%edx", "%esi"
+			);
+	}
+
+	asm volatile(
+			YSCALEYUV2YV12X(0)
+			:: "m" (-lumFilterSize), "r" (lumSrc+lumFilterSize),
+			   "r" (lumMmxFilter+lumFilterSize*4), "r" (dest), "m" (dstW)
+			: "%eax", "%edx", "%esi"
+		);
+#else
+	//FIXME Optimize (just quickly writen not opti..)
 	int i;
+	for(i=0; i<dstW; i++)
+	{
+		int val=0;
+		int j;
+		for(j=0; j<lumFilterSize; j++)
+			val += lumSrc[j][i] * lumFilter[j];
 
-#ifdef ARCH_X86
-	asm volatile ("\n\t"::: "memory");
+		dest[i]= MIN(MAX(val>>19, 0), 255);
+	}
+
+	if(uDest != NULL)
+		for(i=0; i<(dstW>>1); i++)
+		{
+			int u=0;
+			int v=0;
+			int j;
+			for(j=0; j<lumFilterSize; 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);
+		}
 #endif
+}
 
-	for(i=0;i<dstW;i++)
+static inline void RENAME(yuv2yuv1)(int16_t *lumSrc, int16_t *chrSrc,
+				    uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW)
+{
+#ifdef HAVE_MMX
+	if(uDest != NULL)
 	{
-		((uint8_t*)dest)[i] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
+		asm volatile(
+				YSCALEYUV2YV121
+				:: "r" (chrSrc + (dstW>>1)), "r" (uDest + (dstW>>1)),
+				"g" (-(dstW>>1))
+				: "%eax"
+			);
+
+		asm volatile(
+				YSCALEYUV2YV121
+				:: "r" (chrSrc + 2048 + (dstW>>1)), "r" (vDest + (dstW>>1)),
+				"g" (-(dstW>>1))
+				: "%eax"
+			);
 	}
 
-	if(uvalpha != -1)
+	asm volatile(
+		YSCALEYUV2YV121
+		:: "r" (lumSrc + dstW), "r" (dest + dstW),
+		"g" (-dstW)
+		: "%eax"
+	);
+#else
+	//FIXME Optimize (just quickly writen not opti..)
+	//FIXME replace MINMAX with LUTs
+	int i;
+	for(i=0; i<dstW; i++)
 	{
+		int val= lumSrc[i]>>7;
+
+		dest[i]= MIN(MAX(val>>19, 0), 255);
+	}
+
+	if(uDest != NULL)
 		for(i=0; i<(dstW>>1); i++)
 		{
-			((uint8_t*)uDest)[i] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
-			((uint8_t*)vDest)[i] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
+			int u=chrSrc[i]>>7;
+			int v=chrSrc[i + 2048]>>7;
+
+			uDest[i]= MIN(MAX(u>>19, 0), 255);
+			vDest[i]= MIN(MAX(v>>19, 0), 255);
 		}
-	}
+#endif
 }
 
+
 /**
  * vertical scale YV12 to RGB
  */
-static inline void RENAME(yuv2rgbX)(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
+static inline void RENAME(yuv2rgbX)(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+				    int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+			    uint8_t *dest, int dstW, int dstbpp, int16_t * lumMmxFilter, int16_t * chrMmxFilter)
+{
+	if(fullUVIpol)
+	{
+//FIXME
+	}//FULL_UV_IPOL
+	else
+	{
+#ifdef HAVE_MMX
+		if(dstbpp == 32) //FIXME untested
+		{
+			asm volatile(
+				YSCALEYUV2RGBX
+				WRITEBGR32
+
+			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
+			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
+			   "r" (dest), "m" (dstW),
+			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
+			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
+			);
+		}
+		else if(dstbpp==24) //FIXME untested
+		{
+			asm volatile(
+				YSCALEYUV2RGBX
+				"leal (%%eax, %%eax, 2), %%ebx	\n\t" //FIXME optimize
+				"addl %4, %%ebx			\n\t"
+				WRITEBGR24
+
+			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
+			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
+			   "r" (dest), "m" (dstW),
+			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
+			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
+			);
+		}
+		else if(dstbpp==15)
+		{
+			asm volatile(
+				YSCALEYUV2RGBX
+		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
+#ifdef DITHER1XBPP
+				"paddusb b5Dither, %%mm2	\n\t"
+				"paddusb g5Dither, %%mm4	\n\t"
+				"paddusb r5Dither, %%mm5	\n\t"
+#endif
+
+				WRITEBGR15
+
+			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
+			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
+			   "r" (dest), "m" (dstW),
+			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
+			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
+			);
+		}
+		else if(dstbpp==16)
+		{
+			asm volatile(
+				YSCALEYUV2RGBX
+		/* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
+#ifdef DITHER1XBPP
+				"paddusb b5Dither, %%mm2	\n\t"
+				"paddusb g6Dither, %%mm4	\n\t"
+				"paddusb r5Dither, %%mm5	\n\t"
+#endif
+
+				WRITEBGR16
+
+			:: "m" (-lumFilterSize), "m" (-chrFilterSize),
+			   "m" (lumMmxFilter+lumFilterSize*4), "m" (chrMmxFilter+chrFilterSize*4),
+			   "r" (dest), "m" (dstW),
+			   "m" (lumSrc+lumFilterSize), "m" (chrSrc+chrFilterSize)
+			: "%eax", "%ebx", "%ecx", "%edx", "%esi"
+			);
+		}
+#else
+		if(dstbpp==32)
+		{
+			int i;
+			for(i=0; i<(dstW>>1); i++){
+				int j;
+				int Y1=0;
+				int Y2=0;
+				int U=0;
+				int V=0;
+				int Cb, Cr, Cg;
+				for(j=0; j<lumFilterSize; j++)
+				{
+					Y1 += lumSrc[j][2*i] * lumFilter[j];
+					Y2 += lumSrc[j][2*i+1] * lumFilter[j];
+				}
+				for(j=0; j<chrFilterSize; j++)
+				{
+					U += chrSrc[j][i] * chrFilter[j];
+					V += chrSrc[j][i+2048] * chrFilter[j];
+				}
+				Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];
+				Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];
+				U >>= 19;
+				V >>= 19;
+
+				Cb= clip_yuvtab_40cf[U+ 256];
+				Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];
+				Cr= clip_yuvtab_3343[V+ 256];
+
+				dest[8*i+0]=clip_table[((Y1 + Cb) >>13)];
+				dest[8*i+1]=clip_table[((Y1 + Cg) >>13)];
+				dest[8*i+2]=clip_table[((Y1 + Cr) >>13)];
+
+				dest[8*i+4]=clip_table[((Y2 + Cb) >>13)];
+				dest[8*i+5]=clip_table[((Y2 + Cg) >>13)];
+				dest[8*i+6]=clip_table[((Y2 + Cr) >>13)];
+			}
+		}
+		else if(dstbpp==24)
+		{
+			int i;
+			for(i=0; i<(dstW>>1); i++){
+				int j;
+				int Y1=0;
+				int Y2=0;
+				int U=0;
+				int V=0;
+				int Cb, Cr, Cg;
+				for(j=0; j<lumFilterSize; j++)
+				{
+					Y1 += lumSrc[j][2*i] * lumFilter[j];
+					Y2 += lumSrc[j][2*i+1] * lumFilter[j];
+				}
+				for(j=0; j<chrFilterSize; j++)
+				{
+					U += chrSrc[j][i] * chrFilter[j];
+					V += chrSrc[j][i+2048] * chrFilter[j];
+				}
+				Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];
+				Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];
+				U >>= 19;
+				V >>= 19;
+
+				Cb= clip_yuvtab_40cf[U+ 256];
+				Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];
+				Cr= clip_yuvtab_3343[V+ 256];
+
+				dest[0]=clip_table[((Y1 + Cb) >>13)];
+				dest[1]=clip_table[((Y1 + Cg) >>13)];
+				dest[2]=clip_table[((Y1 + Cr) >>13)];
+
+				dest[3]=clip_table[((Y2 + Cb) >>13)];
+				dest[4]=clip_table[((Y2 + Cg) >>13)];
+				dest[5]=clip_table[((Y2 + Cr) >>13)];
+				dest+=6;
+			}
+		}
+		else if(dstbpp==16)
+		{
+			int i;
+			for(i=0; i<(dstW>>1); i++){
+				int j;
+				int Y1=0;
+				int Y2=0;
+				int U=0;
+				int V=0;
+				int Cb, Cr, Cg;
+				for(j=0; j<lumFilterSize; j++)
+				{
+					Y1 += lumSrc[j][2*i] * lumFilter[j];
+					Y2 += lumSrc[j][2*i+1] * lumFilter[j];
+				}
+				for(j=0; j<chrFilterSize; j++)
+				{
+					U += chrSrc[j][i] * chrFilter[j];
+					V += chrSrc[j][i+2048] * chrFilter[j];
+				}
+				Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];
+				Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];
+				U >>= 19;
+				V >>= 19;
+
+				Cb= clip_yuvtab_40cf[U+ 256];
+				Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];
+				Cr= clip_yuvtab_3343[V+ 256];
+
+				((uint16_t*)dest)[2*i] =
+					clip_table16b[(Y1 + Cb) >>13] |
+					clip_table16g[(Y1 + Cg) >>13] |
+					clip_table16r[(Y1 + Cr) >>13];
+
+				((uint16_t*)dest)[2*i+1] =
+					clip_table16b[(Y2 + Cb) >>13] |
+					clip_table16g[(Y2 + Cg) >>13] |
+					clip_table16r[(Y2 + Cr) >>13];
+			}
+		}
+		else if(dstbpp==15)
+		{
+			int i;
+			for(i=0; i<(dstW>>1); i++){
+				int j;
+				int Y1=0;
+				int Y2=0;
+				int U=0;
+				int V=0;
+				int Cb, Cr, Cg;
+				for(j=0; j<lumFilterSize; j++)
+				{
+					Y1 += lumSrc[j][2*i] * lumFilter[j];
+					Y2 += lumSrc[j][2*i+1] * lumFilter[j];
+				}
+				for(j=0; j<chrFilterSize; j++)
+				{
+					U += chrSrc[j][i] * chrFilter[j];
+					V += chrSrc[j][i+2048] * chrFilter[j];
+				}
+				Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];
+				Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];
+				U >>= 19;
+				V >>= 19;
+
+				Cb= clip_yuvtab_40cf[U+ 256];
+				Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];
+				Cr= clip_yuvtab_3343[V+ 256];
+
+				((uint16_t*)dest)[2*i] =
+					clip_table15b[(Y1 + Cb) >>13] |
+					clip_table15g[(Y1 + Cg) >>13] |
+					clip_table15r[(Y1 + Cr) >>13];
+
+				((uint16_t*)dest)[2*i+1] =
+					clip_table15b[(Y2 + Cb) >>13] |
+					clip_table15g[(Y2 + Cg) >>13] |
+					clip_table15r[(Y2 + Cr) >>13];
+			}
+		}
+#endif
+	} //!FULL_UV_IPOL
+}
+
+
+/**
+ * vertical bilinear scale YV12 to RGB
+ */
+static inline void RENAME(yuv2rgb2)(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
 			    uint8_t *dest, int dstW, int yalpha, int uvalpha, int dstbpp)
 {
 	int yalpha1=yalpha^4095;
@@ -708,8 +1156,6 @@ FULL_YSCALEYUV2RGB
 			);
 		}
 #else
-		asm volatile ("\n\t"::: "memory");
-
 		if(dstbpp==32 || dstbpp==24)
 		{
 			int i;
@@ -819,8 +1265,6 @@ FULL_YSCALEYUV2RGB
 			);
 		}
 #else
-		asm volatile ("\n\t"::: "memory");
-
 		if(dstbpp==32)
 		{
 			int i;
@@ -844,7 +1288,7 @@ FULL_YSCALEYUV2RGB
 				dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
 			}
 		}
-		if(dstbpp==24)
+		else if(dstbpp==24)
 		{
 			int i;
 			for(i=0; i<dstW-1; i+=2){
@@ -925,20 +1369,18 @@ FULL_YSCALEYUV2RGB
 /**
  * YV12 to RGB without scaling or interpolating
  */
-static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
-			    uint8_t *dest, int dstW, int yalpha, int uvalpha, int dstbpp)
+static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *uvbuf0, uint16_t *uvbuf1,
+			    uint8_t *dest, int dstW, int uvalpha, int dstbpp)
 {
 	int uvalpha1=uvalpha^4095;
-#ifdef HAVE_MMX
-	int yalpha1=yalpha^4095;
-#endif
+	const int yalpha=0;
+	const int yalpha1=0;
 
 	if(fullUVIpol || allwaysIpol)
 	{
-		RENAME(yuv2rgbX)(buf0, buf1, uvbuf0, uvbuf1, dest, dstW, yalpha, uvalpha, dstbpp);
+		RENAME(yuv2rgb2)(buf0, buf0, uvbuf0, uvbuf1, dest, dstW, 0, uvalpha, dstbpp);
 		return;
 	}
-	if( yalpha > 2048 ) buf0 = buf1;
 
 #ifdef HAVE_MMX
 	if( uvalpha < 2048 ) // note this is not correct (shifts chrominance by 0.5 pixels) but its a bit faster
@@ -948,7 +1390,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 			asm volatile(
 				YSCALEYUV2RGB1
 				WRITEBGR32
-			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
+			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
 			"m" (yalpha1), "m" (uvalpha1)
 			: "%eax"
 			);
@@ -959,7 +1401,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 				"movl %4, %%ebx			\n\t"
 				YSCALEYUV2RGB1
 				WRITEBGR24
-			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstW),
+			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstW),
 			"m" (yalpha1), "m" (uvalpha1)
 			: "%eax", "%ebx"
 			);
@@ -975,7 +1417,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 				"paddusb r5Dither, %%mm5	\n\t"
 #endif
 				WRITEBGR15
-			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
+			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
 			"m" (yalpha1), "m" (uvalpha1)
 			: "%eax"
 			);
@@ -992,7 +1434,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 #endif
 
 				WRITEBGR16
-			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
+			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
 			"m" (yalpha1), "m" (uvalpha1)
 			: "%eax"
 			);
@@ -1005,7 +1447,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 			asm volatile(
 				YSCALEYUV2RGB1b
 				WRITEBGR32
-			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
+			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
 			"m" (yalpha1), "m" (uvalpha1)
 			: "%eax"
 			);
@@ -1016,7 +1458,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 				"movl %4, %%ebx			\n\t"
 				YSCALEYUV2RGB1b
 				WRITEBGR24
-			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstW),
+			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstW),
 			"m" (yalpha1), "m" (uvalpha1)
 			: "%eax", "%ebx"
 			);
@@ -1032,7 +1474,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 				"paddusb r5Dither, %%mm5	\n\t"
 #endif
 				WRITEBGR15
-			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
+			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
 			"m" (yalpha1), "m" (uvalpha1)
 			: "%eax"
 			);
@@ -1049,7 +1491,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 #endif
 
 				WRITEBGR16
-			:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
+			:: "r" (buf0), "r" (buf0), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstW),
 			"m" (yalpha1), "m" (uvalpha1)
 			: "%eax"
 			);
@@ -1057,7 +1499,6 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 	}
 #else
 //FIXME write 2 versions (for even & odd lines)
-	asm volatile ("\n\t"::: "memory");
 
 	if(dstbpp==32)
 	{
@@ -1082,7 +1523,7 @@ static inline void RENAME(yuv2rgb1)(uint16_t *buf0, uint16_t *buf1, uint16_t *uv
 			dest[4*i+6]=clip_table[((Y2 + Cr) >>13)];
 		}
 	}
-	if(dstbpp==24)
+	else if(dstbpp==24)
 	{
 		int i;
 		for(i=0; i<dstW-1; i+=2){
@@ -1245,7 +1686,7 @@ static inline void RENAME(hScale)(int16_t *dst, int dstW, uint8_t *src, int srcW
 			"movd %%mm0, (%4, %%ebp)	\n\t"
 			"addl $4, %%ebp			\n\t"
 			" jnc 1b			\n\t"
-			
+
 			"popl %%ebp			\n\t"
 			: "+a" (counter)
 			: "c" (filter), "d" (filterPos), "S" (src), "D" (dst)
@@ -1294,7 +1735,7 @@ static inline void RENAME(hScale)(int16_t *dst, int dstW, uint8_t *src, int srcW
 			"movd %%mm4, (%%eax, %0)	\n\t"
 			"addl $4, %0			\n\t"
 			" jnc 1b			\n\t"
-			
+
 			: "+r" (counter)
 			: "r" (filter), "m" (filterPos), "m" (dst), "m"(src+filterSize),
 			  "m" (src), "r" (filterSize*2)
@@ -1308,7 +1749,7 @@ static inline void RENAME(hScale)(int16_t *dst, int dstW, uint8_t *src, int srcW
 		int j;
 		int srcPos= filterPos[i];
 		int val=0;
-//		printf("filterPos: %d\n", hFilterPos[i]);
+//		printf("filterPos: %d\n", filterPos[i]);
 		for(j=0; j<filterSize; j++)
 		{
 //			printf("filter: %d, src: %d\n", filter[i], src[srcPos + j]);
@@ -1516,10 +1957,11 @@ FUNNYUVCODE
 		  "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF), "m" (src2)
 		: "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
 	);
-		for(i=dstWidth-1; (i*xInc)>>16 >=srcW/2-1; i--)
+		for(i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--)
 		{
-			dst[i] = src1[srcW/2-1]*128;
-			dst[i+2048] = src2[srcW/2-1]*128;
+//			printf("%d %d %d\n", dstWidth, i, srcW);
+			dst[i] = src1[srcW-1]*128;
+			dst[i+2048] = src2[srcW-1]*128;
 		}
 	}
 	else
@@ -1584,33 +2026,45 @@ FUNNYUVCODE
    }
 }
 
-static void inline RENAME(initFilter)(int16_t *filter, int16_t *filterPos, int *filterSize, int xInc,
-				      int srcW, int dstW)
+static inline void RENAME(initFilter)(int16_t *filter, int16_t *filterPos, int *filterSize, int xInc,
+				      int srcW, int dstW, int filterAlign, int one)
 {
 	int i;
 #ifdef HAVE_MMX
-	asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS
+	asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions) 
 #endif
-	
-	if(xInc <= (1<<16)) // upscale / cubic interpolate
+
+	if(ABS(xInc - 0x10000) <10) // unscaled
+	{
+		int i;
+		*filterSize= (1 +(filterAlign-1)) & (~(filterAlign-1)); // 1 or 4 normaly
+		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(xInc <= (1<<16) || sws_flags==SWS_FAST_BILINEAR) // upscale
 	{
 		int i;
 		int xDstInSrc;
 		if(sws_flags==SWS_BICUBIC) *filterSize= 4;
 		else			   *filterSize= 2;
 //		printf("%d %d %d\n", filterSize, srcW, dstW);
-#ifdef HAVE_MMX
-		*filterSize= (*filterSize +3) & (~3); // -> *filterSize %4 == 0
-#endif
+		*filterSize= (*filterSize +(filterAlign-1)) & (~(filterAlign-1));
+
 		xDstInSrc= xInc - 0x8000;
 		for(i=0; i<dstW; i++)
 		{
 			int xx= (xDstInSrc>>16) - (*filterSize>>1) + 1;
 			int j;
-			
+
 			filterPos[i]= xx;
 			if(sws_flags == SWS_BICUBIC)
-			{				
+			{
 				double d= ABS(((xx+1)<<16) - xDstInSrc)/(double)(1<<16);
 //				int coeff;
 				int y1,y2,y3,y4;
@@ -1620,7 +2074,7 @@ static void inline RENAME(initFilter)(int16_t *filter, int16_t *filterPos, int *
 		y2 = (int)floor(0.5 + (+ 1.0             -     (A+3.0)*d*d + (A+2.0)*d*d*d) * 16384.0);
 		y3 = (int)floor(0.5 + (        -     A*d + (2.0*A+3.0)*d*d - (A+2.0)*d*d*d) * 16384.0);
 		y4 = (int)floor(0.5 + (                  +           A*d*d -       A*d*d*d) * 16384.0);
-						
+
 //				printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
 				filter[i*(*filterSize) + 0]= y1;
 				filter[i*(*filterSize) + 1]= y2;
@@ -1628,20 +2082,20 @@ static void inline RENAME(initFilter)(int16_t *filter, int16_t *filterPos, int *
 				filter[i*(*filterSize) + 3]= y4;
 //				printf("%1.3f %d, %d, %d, %d\n",d , y1, y2, y3, y4);
 			}
-			else 
+			else
 			{
 				for(j=0; j<*filterSize; j++)
 				{
 					double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
 					int coeff;
 					coeff= (int)(0.5 + (1.0 - d)*(1<<14));
-					if(coeff<0) coeff=0;					
+					if(coeff<0) coeff=0;
 	//				printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
 					filter[i*(*filterSize) + j]= coeff;
 					xx++;
 				}
-			}		
-			xDstInSrc+= xInc;			
+			}
+			xDstInSrc+= xInc;
 		}
 	}
 	else // downscale
@@ -1650,47 +2104,46 @@ static void inline RENAME(initFilter)(int16_t *filter, int16_t *filterPos, int *
 		if(sws_flags==SWS_BICUBIC) *filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);
 		else			   *filterSize= (int)ceil(1 + 2.0*srcW / (double)dstW);
 //		printf("%d %d %d\n", *filterSize, srcW, dstW);
-#ifdef HAVE_MMX
-		*filterSize= (*filterSize +3) & (~3); // -> *filterSize %4 == 0
-#endif
+		*filterSize= (*filterSize +(filterAlign-1)) & (~(filterAlign-1));
+
 		xDstInSrc= xInc - 0x8000;
 		for(i=0; i<dstW; i++)
 		{
 			int xx= (int)((double)xDstInSrc/(double)(1<<16) - *filterSize*0.5 + 0.5);
 			int j;
-			
+
 			filterPos[i]= xx;
 			for(j=0; j<*filterSize; j++)
 			{
 				double d= ABS((xx<<16) - xDstInSrc)/(double)xInc;
 				int coeff;
 				if(sws_flags == SWS_BICUBIC)
-				{				
+				{
 					double A= -0.75;
 //					d*=2;
 					// Equation is from VirtualDub
 					if(d<1.0)
-						coeff = (int)floor(0.5 + (1.0 - (A+3.0)*d*d 
+						coeff = (int)floor(0.5 + (1.0 - (A+3.0)*d*d
 						        + (A+2.0)*d*d*d) * (1<<14));
 					else if(d<2.0)
-						coeff = (int)floor(0.5 + (-4.0*A + 8.0*A*d 
+						coeff = (int)floor(0.5 + (-4.0*A + 8.0*A*d
 						        - 5.0*A*d*d + A*d*d*d) * (1<<14));
 					else
 						coeff=0;
 				}
-				else 
+				else
 				{
 					coeff= (int)(0.5 + (1.0 - d)*(1<<14));
-					if(coeff<0) coeff=0;					
+					if(coeff<0) coeff=0;
 				}
-//				printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
+//				if(filterAlign==1) printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
 				filter[i*(*filterSize) + j]= coeff;
 				xx++;
 			}
 			xDstInSrc+= xInc;
 		}
 	}
-	
+
 	//fix borders
 	for(i=0; i<dstW; i++)
 	{
@@ -1717,16 +2170,19 @@ static void inline RENAME(initFilter)(int16_t *filter, int16_t *filterPos, int *
 				filter[i*(*filterSize) +right] += filter[i*(*filterSize) +j];
 				filter[i*(*filterSize) +j]=0;
 			}
-			filterPos[i]= srcW - *filterSize; 
+			filterPos[i]= srcW - *filterSize;
 		}
 	}
-	
+
+	//FIXME try to align filterpos if possible / try to shift filterpos to put zeros at the end
+	// and skip these than later
+
 	//Normalize
 	for(i=0; i<dstW; i++)
 	{
 		int j;
 		double sum=0;
-		double scale=1<<14;
+		double scale= one;
 		for(j=0; j<*filterSize; j++)
 		{
 			sum+= filter[i*(*filterSize) + j];
@@ -1739,295 +2195,473 @@