postprocess_template.c

		"movq %%mm2, %%mm1		\n\t"
		"punpcklwd %%mm4, %%mm2		\n\t"
		"punpckhwd %%mm4, %%mm1		\n\t"

		"movd %%mm0, (%0)		\n\t"
		"psrlq $32, %%mm0		\n\t"
		"movd %%mm0, (%%eax)		\n\t"
		"movd %%mm3, (%%eax, %1)	\n\t"
		"psrlq $32, %%mm3		\n\t"
		"movd %%mm3, (%%eax, %1, 2)	\n\t"
		"movd %%mm2, (%0, %1, 4)	\n\t"
		"psrlq $32, %%mm2		\n\t"
		"movd %%mm2, (%%edx)		\n\t"
		"movd %%mm1, (%%edx, %1)	\n\t"
		"psrlq $32, %%mm1		\n\t"
		"movd %%mm1, (%%edx, %1, 2)	\n\t"


		"movq 64(%2), %%mm0		\n\t" // 12345678
		"movq 80(%2), %%mm1		\n\t" // abcdefgh
		"movq %%mm0, %%mm2		\n\t" // 12345678
		"punpcklbw %%mm1, %%mm0		\n\t" // 1a2b3c4d
		"punpckhbw %%mm1, %%mm2		\n\t" // 5e6f7g8h

		"movq 96(%2), %%mm1		\n\t"
		"movq 112(%2), %%mm3		\n\t"
		"movq %%mm1, %%mm4		\n\t"
		"punpcklbw %%mm3, %%mm1		\n\t"
		"punpckhbw %%mm3, %%mm4		\n\t"

		"movq %%mm0, %%mm3		\n\t"
		"punpcklwd %%mm1, %%mm0		\n\t"
		"punpckhwd %%mm1, %%mm3		\n\t"
		"movq %%mm2, %%mm1		\n\t"
		"punpcklwd %%mm4, %%mm2		\n\t"
		"punpckhwd %%mm4, %%mm1		\n\t"

		"movd %%mm0, 4(%0)		\n\t"
		"psrlq $32, %%mm0		\n\t"
		"movd %%mm0, 4(%%eax)		\n\t"
		"movd %%mm3, 4(%%eax, %1)	\n\t"
		"psrlq $32, %%mm3		\n\t"
		"movd %%mm3, 4(%%eax, %1, 2)	\n\t"
		"movd %%mm2, 4(%0, %1, 4)	\n\t"
		"psrlq $32, %%mm2		\n\t"
		"movd %%mm2, 4(%%edx)		\n\t"
		"movd %%mm1, 4(%%edx, %1)	\n\t"
		"psrlq $32, %%mm1		\n\t"
		"movd %%mm1, 4(%%edx, %1, 2)	\n\t"

	:: "r" (dst), "r" (dstStride), "r" (src)
	: "%eax", "%edx"
	);
}
#endif
//static int test=0;

static void inline RENAME(tempNoiseReducer)(uint8_t *src, int stride,
				    uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
{
	// to save a register (FIXME do this outside of the loops)
	tempBluredPast[127]= maxNoise[0];
	tempBluredPast[128]= maxNoise[1];
	tempBluredPast[129]= maxNoise[2];
        
#define FAST_L2_DIFF
//#define L1_DIFF //u should change the thresholds too if u try that one
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
	asm volatile(
		"leal (%2, %2, 2), %%eax			\n\t" // 3*stride
		"leal (%2, %2, 4), %%edx			\n\t" // 5*stride
		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride
//	0	1	2	3	4	5	6	7	8	9
//	%x	%x+%2	%x+2%2	%x+eax	%x+4%2	%x+edx	%x+2eax	%x+ecx	%x+8%2
//FIXME reorder?
#ifdef L1_DIFF //needs mmx2
		"movq (%0), %%mm0				\n\t" // L0
		"psadbw (%1), %%mm0				\n\t" // |L0-R0|
		"movq (%0, %2), %%mm1				\n\t" // L1
		"psadbw (%1, %2), %%mm1				\n\t" // |L1-R1|
		"movq (%0, %2, 2), %%mm2			\n\t" // L2
		"psadbw (%1, %2, 2), %%mm2			\n\t" // |L2-R2|
		"movq (%0, %%eax), %%mm3			\n\t" // L3
		"psadbw (%1, %%eax), %%mm3			\n\t" // |L3-R3|

		"movq (%0, %2, 4), %%mm4			\n\t" // L4
		"paddw %%mm1, %%mm0				\n\t"
		"psadbw (%1, %2, 4), %%mm4			\n\t" // |L4-R4|
		"movq (%0, %%edx), %%mm5			\n\t" // L5
		"paddw %%mm2, %%mm0				\n\t"
		"psadbw (%1, %%edx), %%mm5			\n\t" // |L5-R5|
		"movq (%0, %%eax, 2), %%mm6			\n\t" // L6
		"paddw %%mm3, %%mm0				\n\t"
		"psadbw (%1, %%eax, 2), %%mm6			\n\t" // |L6-R6|
		"movq (%0, %%ecx), %%mm7			\n\t" // L7
		"paddw %%mm4, %%mm0				\n\t"
		"psadbw (%1, %%ecx), %%mm7			\n\t" // |L7-R7|
		"paddw %%mm5, %%mm6				\n\t"
		"paddw %%mm7, %%mm6				\n\t"
		"paddw %%mm6, %%mm0				\n\t"
#elif defined (FAST_L2_DIFF)
		"pcmpeqb %%mm7, %%mm7				\n\t"
		"movq "MANGLE(b80)", %%mm6			\n\t"
		"pxor %%mm0, %%mm0				\n\t"
#define L2_DIFF_CORE(a, b)\
		"movq " #a ", %%mm5				\n\t"\
		"movq " #b ", %%mm2				\n\t"\
		"pxor %%mm7, %%mm2				\n\t"\
		PAVGB(%%mm2, %%mm5)\
		"paddb %%mm6, %%mm5				\n\t"\
		"movq %%mm5, %%mm2				\n\t"\
		"psllw $8, %%mm5				\n\t"\
		"pmaddwd %%mm5, %%mm5				\n\t"\
		"pmaddwd %%mm2, %%mm2				\n\t"\
		"paddd %%mm2, %%mm5				\n\t"\
		"psrld $14, %%mm5				\n\t"\
		"paddd %%mm5, %%mm0				\n\t"

L2_DIFF_CORE((%0), (%1))
L2_DIFF_CORE((%0, %2), (%1, %2))
L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2))
L2_DIFF_CORE((%0, %%eax), (%1, %%eax))
L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4))
L2_DIFF_CORE((%0, %%edx), (%1, %%edx))
L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2))
L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx))

#else
		"pxor %%mm7, %%mm7				\n\t"
		"pxor %%mm0, %%mm0				\n\t"
#define L2_DIFF_CORE(a, b)\
		"movq " #a ", %%mm5				\n\t"\
		"movq " #b ", %%mm2				\n\t"\
		"movq %%mm5, %%mm1				\n\t"\
		"movq %%mm2, %%mm3				\n\t"\
		"punpcklbw %%mm7, %%mm5				\n\t"\
		"punpckhbw %%mm7, %%mm1				\n\t"\
		"punpcklbw %%mm7, %%mm2				\n\t"\
		"punpckhbw %%mm7, %%mm3				\n\t"\
		"psubw %%mm2, %%mm5				\n\t"\
		"psubw %%mm3, %%mm1				\n\t"\
		"pmaddwd %%mm5, %%mm5				\n\t"\
		"pmaddwd %%mm1, %%mm1				\n\t"\
		"paddd %%mm1, %%mm5				\n\t"\
		"paddd %%mm5, %%mm0				\n\t"

L2_DIFF_CORE((%0), (%1))
L2_DIFF_CORE((%0, %2), (%1, %2))
L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2))
L2_DIFF_CORE((%0, %%eax), (%1, %%eax))
L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4))
L2_DIFF_CORE((%0, %%edx), (%1, %%edx))
L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2))
L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx))

#endif

		"movq %%mm0, %%mm4				\n\t"
		"psrlq $32, %%mm0				\n\t"
		"paddd %%mm0, %%mm4				\n\t"
		"movd %%mm4, %%ecx				\n\t"
		"shll $2, %%ecx					\n\t"
		"movl %3, %%edx					\n\t"
		"addl -4(%%edx), %%ecx				\n\t"
		"addl 4(%%edx), %%ecx				\n\t"
		"addl -1024(%%edx), %%ecx			\n\t"
		"addl $4, %%ecx					\n\t"
		"addl 1024(%%edx), %%ecx			\n\t"
		"shrl $3, %%ecx					\n\t"
		"movl %%ecx, (%%edx)				\n\t"

//		"movl %3, %%ecx					\n\t"
//		"movl %%ecx, test				\n\t"
//		"jmp 4f \n\t"
		"cmpl 512(%%edx), %%ecx				\n\t"
		" jb 2f						\n\t"
		"cmpl 516(%%edx), %%ecx				\n\t"
		" jb 1f						\n\t"

		"leal (%%eax, %2, 2), %%edx			\n\t" // 5*stride
		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride
		"movq (%0), %%mm0				\n\t" // L0
		"movq (%0, %2), %%mm1				\n\t" // L1
		"movq (%0, %2, 2), %%mm2			\n\t" // L2
		"movq (%0, %%eax), %%mm3			\n\t" // L3
		"movq (%0, %2, 4), %%mm4			\n\t" // L4
		"movq (%0, %%edx), %%mm5			\n\t" // L5
		"movq (%0, %%eax, 2), %%mm6			\n\t" // L6
		"movq (%0, %%ecx), %%mm7			\n\t" // L7
		"movq %%mm0, (%1)				\n\t" // L0
		"movq %%mm1, (%1, %2)				\n\t" // L1
		"movq %%mm2, (%1, %2, 2)			\n\t" // L2
		"movq %%mm3, (%1, %%eax)			\n\t" // L3
		"movq %%mm4, (%1, %2, 4)			\n\t" // L4
		"movq %%mm5, (%1, %%edx)			\n\t" // L5
		"movq %%mm6, (%1, %%eax, 2)			\n\t" // L6
		"movq %%mm7, (%1, %%ecx)			\n\t" // L7
		"jmp 4f						\n\t"

		"1:						\n\t"
		"leal (%%eax, %2, 2), %%edx			\n\t" // 5*stride
		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride
		"movq (%0), %%mm0				\n\t" // L0
		PAVGB((%1), %%mm0)				      // L0
		"movq (%0, %2), %%mm1				\n\t" // L1
		PAVGB((%1, %2), %%mm1)				      // L1
		"movq (%0, %2, 2), %%mm2			\n\t" // L2
		PAVGB((%1, %2, 2), %%mm2)			      // L2
		"movq (%0, %%eax), %%mm3			\n\t" // L3
		PAVGB((%1, %%eax), %%mm3)			      // L3
		"movq (%0, %2, 4), %%mm4			\n\t" // L4
		PAVGB((%1, %2, 4), %%mm4)			      // L4
		"movq (%0, %%edx), %%mm5			\n\t" // L5
		PAVGB((%1, %%edx), %%mm5)			      // L5
		"movq (%0, %%eax, 2), %%mm6			\n\t" // L6
		PAVGB((%1, %%eax, 2), %%mm6)			      // L6
		"movq (%0, %%ecx), %%mm7			\n\t" // L7
		PAVGB((%1, %%ecx), %%mm7)			      // L7
		"movq %%mm0, (%1)				\n\t" // R0
		"movq %%mm1, (%1, %2)				\n\t" // R1
		"movq %%mm2, (%1, %2, 2)			\n\t" // R2
		"movq %%mm3, (%1, %%eax)			\n\t" // R3
		"movq %%mm4, (%1, %2, 4)			\n\t" // R4
		"movq %%mm5, (%1, %%edx)			\n\t" // R5
		"movq %%mm6, (%1, %%eax, 2)			\n\t" // R6
		"movq %%mm7, (%1, %%ecx)			\n\t" // R7
		"movq %%mm0, (%0)				\n\t" // L0
		"movq %%mm1, (%0, %2)				\n\t" // L1
		"movq %%mm2, (%0, %2, 2)			\n\t" // L2
		"movq %%mm3, (%0, %%eax)			\n\t" // L3
		"movq %%mm4, (%0, %2, 4)			\n\t" // L4
		"movq %%mm5, (%0, %%edx)			\n\t" // L5
		"movq %%mm6, (%0, %%eax, 2)			\n\t" // L6
		"movq %%mm7, (%0, %%ecx)			\n\t" // L7
		"jmp 4f						\n\t"

		"2:						\n\t"
		"cmpl 508(%%edx), %%ecx				\n\t"
		" jb 3f						\n\t"

		"leal (%%eax, %2, 2), %%edx			\n\t" // 5*stride
		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride
		"movq (%0), %%mm0				\n\t" // L0
		"movq (%0, %2), %%mm1				\n\t" // L1
		"movq (%0, %2, 2), %%mm2			\n\t" // L2
		"movq (%0, %%eax), %%mm3			\n\t" // L3
		"movq (%1), %%mm4				\n\t" // R0
		"movq (%1, %2), %%mm5				\n\t" // R1
		"movq (%1, %2, 2), %%mm6			\n\t" // R2
		"movq (%1, %%eax), %%mm7			\n\t" // R3
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		"movq %%mm0, (%1)				\n\t" // R0
		"movq %%mm1, (%1, %2)				\n\t" // R1
		"movq %%mm2, (%1, %2, 2)			\n\t" // R2
		"movq %%mm3, (%1, %%eax)			\n\t" // R3
		"movq %%mm0, (%0)				\n\t" // L0
		"movq %%mm1, (%0, %2)				\n\t" // L1
		"movq %%mm2, (%0, %2, 2)			\n\t" // L2
		"movq %%mm3, (%0, %%eax)			\n\t" // L3

		"movq (%0, %2, 4), %%mm0			\n\t" // L4
		"movq (%0, %%edx), %%mm1			\n\t" // L5
		"movq (%0, %%eax, 2), %%mm2			\n\t" // L6
		"movq (%0, %%ecx), %%mm3			\n\t" // L7
		"movq (%1, %2, 4), %%mm4			\n\t" // R4
		"movq (%1, %%edx), %%mm5			\n\t" // R5
		"movq (%1, %%eax, 2), %%mm6			\n\t" // R6
		"movq (%1, %%ecx), %%mm7			\n\t" // R7
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		"movq %%mm0, (%1, %2, 4)			\n\t" // R4
		"movq %%mm1, (%1, %%edx)			\n\t" // R5
		"movq %%mm2, (%1, %%eax, 2)			\n\t" // R6
		"movq %%mm3, (%1, %%ecx)			\n\t" // R7
		"movq %%mm0, (%0, %2, 4)			\n\t" // L4
		"movq %%mm1, (%0, %%edx)			\n\t" // L5
		"movq %%mm2, (%0, %%eax, 2)			\n\t" // L6
		"movq %%mm3, (%0, %%ecx)			\n\t" // L7
		"jmp 4f						\n\t"

		"3:						\n\t"
		"leal (%%eax, %2, 2), %%edx			\n\t" // 5*stride
		"leal (%%edx, %2, 2), %%ecx			\n\t" // 7*stride
		"movq (%0), %%mm0				\n\t" // L0
		"movq (%0, %2), %%mm1				\n\t" // L1
		"movq (%0, %2, 2), %%mm2			\n\t" // L2
		"movq (%0, %%eax), %%mm3			\n\t" // L3
		"movq (%1), %%mm4				\n\t" // R0
		"movq (%1, %2), %%mm5				\n\t" // R1
		"movq (%1, %2, 2), %%mm6			\n\t" // R2
		"movq (%1, %%eax), %%mm7			\n\t" // R3
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		"movq %%mm0, (%1)				\n\t" // R0
		"movq %%mm1, (%1, %2)				\n\t" // R1
		"movq %%mm2, (%1, %2, 2)			\n\t" // R2
		"movq %%mm3, (%1, %%eax)			\n\t" // R3
		"movq %%mm0, (%0)				\n\t" // L0
		"movq %%mm1, (%0, %2)				\n\t" // L1
		"movq %%mm2, (%0, %2, 2)			\n\t" // L2
		"movq %%mm3, (%0, %%eax)			\n\t" // L3

		"movq (%0, %2, 4), %%mm0			\n\t" // L4
		"movq (%0, %%edx), %%mm1			\n\t" // L5
		"movq (%0, %%eax, 2), %%mm2			\n\t" // L6
		"movq (%0, %%ecx), %%mm3			\n\t" // L7
		"movq (%1, %2, 4), %%mm4			\n\t" // R4
		"movq (%1, %%edx), %%mm5			\n\t" // R5
		"movq (%1, %%eax, 2), %%mm6			\n\t" // R6
		"movq (%1, %%ecx), %%mm7			\n\t" // R7
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		PAVGB(%%mm4, %%mm0)
		PAVGB(%%mm5, %%mm1)
		PAVGB(%%mm6, %%mm2)
		PAVGB(%%mm7, %%mm3)
		"movq %%mm0, (%1, %2, 4)			\n\t" // R4
		"movq %%mm1, (%1, %%edx)			\n\t" // R5
		"movq %%mm2, (%1, %%eax, 2)			\n\t" // R6
		"movq %%mm3, (%1, %%ecx)			\n\t" // R7
		"movq %%mm0, (%0, %2, 4)			\n\t" // L4
		"movq %%mm1, (%0, %%edx)			\n\t" // L5
		"movq %%mm2, (%0, %%eax, 2)			\n\t" // L6
		"movq %%mm3, (%0, %%ecx)			\n\t" // L7

		"4:						\n\t"

		:: "r" (src), "r" (tempBlured), "r"(stride), "m" (tempBluredPast)
		: "%eax", "%edx", "%ecx", "memory"
		);
//printf("%d\n", test);
#else
	int y;
	int d=0;
	int sysd=0;
	int i;

	for(y=0; y<8; y++)
	{
		int x;
		for(x=0; x<8; x++)
		{
			int ref= tempBlured[ x + y*stride ];
			int cur= src[ x + y*stride ];
			int d1=ref - cur;
//			if(x==0 || x==7) d1+= d1>>1;
//			if(y==0 || y==7) d1+= d1>>1;
//			d+= ABS(d1);
			d+= d1*d1;
			sysd+= d1;
		}
	}
	i=d;
	d= 	(
		4*d
		+(*(tempBluredPast-256))
		+(*(tempBluredPast-1))+ (*(tempBluredPast+1))
		+(*(tempBluredPast+256))
		+4)>>3;
	*tempBluredPast=i;
//	((*tempBluredPast)*3 + d + 2)>>2;

//printf("%d %d %d\n", maxNoise[0], maxNoise[1], maxNoise[2]);
/*
Switch between
 1  0  0  0  0  0  0  (0)
64 32 16  8  4  2  1  (1)
64 48 36 27 20 15 11 (33) (approx)
64 56 49 43 37 33 29 (200) (approx)
*/
	if(d > maxNoise[1])
	{
		if(d < maxNoise[2])
		{
			for(y=0; y<8; y++)
			{
				int x;
				for(x=0; x<8; x++)
				{
					int ref= tempBlured[ x + y*stride ];
					int cur= src[ x + y*stride ];
					tempBlured[ x + y*stride ]=
					src[ x + y*stride ]=
						(ref + cur + 1)>>1;
				}
			}
		}
		else
		{
			for(y=0; y<8; y++)
			{
				int x;
				for(x=0; x<8; x++)
				{
					tempBlured[ x + y*stride ]= src[ x + y*stride ];
				}
			}
		}
	}
	else
	{
		if(d < maxNoise[0])
		{
			for(y=0; y<8; y++)
			{
				int x;
				for(x=0; x<8; x++)
				{
					int ref= tempBlured[ x + y*stride ];
					int cur= src[ x + y*stride ];
					tempBlured[ x + y*stride ]=
					src[ x + y*stride ]=
						(ref*7 + cur + 4)>>3;
				}
			}
		}
		else
		{
			for(y=0; y<8; y++)
			{
				int x;
				for(x=0; x<8; x++)
				{
					int ref= tempBlured[ x + y*stride ];
					int cur= src[ x + y*stride ];
					tempBlured[ x + y*stride ]=
					src[ x + y*stride ]=
						(ref*3 + cur + 2)>>2;
				}
			}
		}
	}
#endif
}

static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
	QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);

/**
 * Copies a block from src to dst and fixes the blacklevel
 * levelFix == 0 -> dont touch the brighness & contrast
 */
#undef SCALED_CPY

static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
	int levelFix, int64_t *packedOffsetAndScale)
{
#ifndef HAVE_MMX
	int i;
#endif
	if(levelFix)
	{
#ifdef HAVE_MMX
					asm volatile(
						"movq (%%eax), %%mm2	\n\t" // packedYOffset
						"movq 8(%%eax), %%mm3	\n\t" // packedYScale
						"leal (%2,%4), %%eax	\n\t"
						"leal (%3,%5), %%edx	\n\t"
						"pxor %%mm4, %%mm4	\n\t"
#ifdef HAVE_MMX2
#define SCALED_CPY(src1, src2, dst1, dst2)					\
						"movq " #src1 ", %%mm0	\n\t"\
						"movq " #src1 ", %%mm5	\n\t"\
						"movq " #src2 ", %%mm1	\n\t"\
						"movq " #src2 ", %%mm6	\n\t"\
						"punpcklbw %%mm0, %%mm0 \n\t"\
						"punpckhbw %%mm5, %%mm5 \n\t"\
						"punpcklbw %%mm1, %%mm1 \n\t"\
						"punpckhbw %%mm6, %%mm6 \n\t"\
						"pmulhuw %%mm3, %%mm0	\n\t"\
						"pmulhuw %%mm3, %%mm5	\n\t"\
						"pmulhuw %%mm3, %%mm1	\n\t"\
						"pmulhuw %%mm3, %%mm6	\n\t"\
						"psubw %%mm2, %%mm0	\n\t"\
						"psubw %%mm2, %%mm5	\n\t"\
						"psubw %%mm2, %%mm1	\n\t"\
						"psubw %%mm2, %%mm6	\n\t"\
						"packuswb %%mm5, %%mm0	\n\t"\
						"packuswb %%mm6, %%mm1	\n\t"\
						"movq %%mm0, " #dst1 "	\n\t"\
						"movq %%mm1, " #dst2 "	\n\t"\

#else //HAVE_MMX2
#define SCALED_CPY(src1, src2, dst1, dst2)					\
						"movq " #src1 ", %%mm0	\n\t"\
						"movq " #src1 ", %%mm5	\n\t"\
						"punpcklbw %%mm4, %%mm0 \n\t"\
						"punpckhbw %%mm4, %%mm5 \n\t"\
						"psubw %%mm2, %%mm0	\n\t"\
						"psubw %%mm2, %%mm5	\n\t"\
						"movq " #src2 ", %%mm1	\n\t"\
						"psllw $6, %%mm0	\n\t"\
						"psllw $6, %%mm5	\n\t"\
						"pmulhw %%mm3, %%mm0	\n\t"\
						"movq " #src2 ", %%mm6	\n\t"\
						"pmulhw %%mm3, %%mm5	\n\t"\
						"punpcklbw %%mm4, %%mm1 \n\t"\
						"punpckhbw %%mm4, %%mm6 \n\t"\
						"psubw %%mm2, %%mm1	\n\t"\
						"psubw %%mm2, %%mm6	\n\t"\
						"psllw $6, %%mm1	\n\t"\
						"psllw $6, %%mm6	\n\t"\
						"pmulhw %%mm3, %%mm1	\n\t"\
						"pmulhw %%mm3, %%mm6	\n\t"\
						"packuswb %%mm5, %%mm0	\n\t"\
						"packuswb %%mm6, %%mm1	\n\t"\
						"movq %%mm0, " #dst1 "	\n\t"\
						"movq %%mm1, " #dst2 "	\n\t"\

#endif //!HAVE_MMX2

SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))
SCALED_CPY((%2, %4, 2), (%%eax, %4, 2), (%3, %5, 2), (%%edx, %5, 2))
SCALED_CPY((%2, %4, 4), (%%eax, %4, 4), (%3, %5, 4), (%%edx, %5, 4))
						"leal (%%eax,%4,4), %%eax	\n\t"
						"leal (%%edx,%5,4), %%edx	\n\t"
SCALED_CPY((%%eax, %4), (%%eax, %4, 2), (%%edx, %5), (%%edx, %5, 2))


						: "=&a" (packedOffsetAndScale)
						: "0" (packedOffsetAndScale),
						"r"(src),
						"r"(dst),
						"r" (srcStride),
						"r" (dstStride)
						: "%edx"
					);
#else
				for(i=0; i<8; i++)
					memcpy(	&(dst[dstStride*i]),
						&(src[srcStride*i]), BLOCK_SIZE);
#endif
	}
	else
	{
#ifdef HAVE_MMX
					asm volatile(
						"leal (%0,%2), %%eax	\n\t"
						"leal (%1,%3), %%edx	\n\t"

#define SIMPLE_CPY(src1, src2, dst1, dst2)				\
						"movq " #src1 ", %%mm0	\n\t"\
						"movq " #src2 ", %%mm1	\n\t"\
						"movq %%mm0, " #dst1 "	\n\t"\
						"movq %%mm1, " #dst2 "	\n\t"\

SIMPLE_CPY((%0)       , (%0, %2)      , (%1)       , (%1, %3))
SIMPLE_CPY((%0, %2, 2), (%%eax, %2, 2), (%1, %3, 2), (%%edx, %3, 2))
SIMPLE_CPY((%0, %2, 4), (%%eax, %2, 4), (%1, %3, 4), (%%edx, %3, 4))
						"leal (%%eax,%2,4), %%eax	\n\t"
						"leal (%%edx,%3,4), %%edx	\n\t"
SIMPLE_CPY((%%eax, %2), (%%eax, %2, 2), (%%edx, %3), (%%edx, %3, 2))

						: : "r" (src),
						"r" (dst),
						"r" (srcStride),
						"r" (dstStride)
						: "%eax", "%edx"
					);
#else
				for(i=0; i<8; i++)
					memcpy(	&(dst[dstStride*i]),
						&(src[srcStride*i]), BLOCK_SIZE);
#endif
	}
}

/**
 * Duplicates the given 8 src pixels ? times upward
 */
static inline void RENAME(duplicate)(uint8_t src[], int stride)
{
#ifdef HAVE_MMX
	asm volatile(
		"movq (%0), %%mm0		\n\t"
		"addl %1, %0			\n\t"
		"movq %%mm0, (%0)		\n\t"
		"movq %%mm0, (%0, %1)		\n\t"
		"movq %%mm0, (%0, %1, 2)	\n\t"
		: "+r" (src)
		: "r" (-stride)
	);
#else
	int i;
	uint8_t *p=src;
	for(i=0; i<3; i++)
	{
		p-= stride;
		memcpy(p, src, 8);
	}
#endif
}

/**
 * Filters array of bytes (Y or U or V values)
 */
static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
	QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
{
	PPContext __attribute__((aligned(8))) c= *c2; //copy to stack for faster access
	int x,y;
#ifdef COMPILE_TIME_MODE
	const int mode= COMPILE_TIME_MODE;
#else
	const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
#endif
	int black=0, white=255; // blackest black and whitest white in the picture
	int QPCorrecture= 256*256;

	int copyAhead;

	//FIXME remove
	uint64_t * const yHistogram= c.yHistogram;
	uint8_t * const tempSrc= c.tempSrc;
	uint8_t * const tempDst= c.tempDst;

	c.dcOffset= c.ppMode.maxDcDiff;
	c.dcThreshold= c.ppMode.maxDcDiff*2 + 1;

#ifdef HAVE_MMX
	c.mmxDcOffset= 0x7F - c.dcOffset;
	c.mmxDcThreshold= 0x7F - c.dcThreshold;

	c.mmxDcOffset*= 0x0101010101010101LL;
	c.mmxDcThreshold*= 0x0101010101010101LL;
#endif

	if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
	else if(   (mode & LINEAR_BLEND_DEINT_FILTER)
		|| (mode & FFMPEG_DEINT_FILTER)) copyAhead=14;
	else if(   (mode & V_DEBLOCK)
		|| (mode & LINEAR_IPOL_DEINT_FILTER)
		|| (mode & MEDIAN_DEINT_FILTER)) copyAhead=13;
	else if(mode & V_X1_FILTER) copyAhead=11;
//	else if(mode & V_RK1_FILTER) copyAhead=10;
	else if(mode & DERING) copyAhead=9;
	else copyAhead=8;

	copyAhead-= 8;

	if(!isColor)
	{
		uint64_t sum= 0;
		int i;
		uint64_t maxClipped;
		uint64_t clipped;
		double scale;

		c.frameNum++;
		// first frame is fscked so we ignore it
		if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;

		for(i=0; i<256; i++)
		{
			sum+= yHistogram[i];
//			printf("%d ", yHistogram[i]);
		}
//		printf("\n\n");

		/* we allways get a completly black picture first */
		maxClipped= (uint64_t)(sum * maxClippedThreshold);

		clipped= sum;
		for(black=255; black>0; black--)
		{
			if(clipped < maxClipped) break;
			clipped-= yHistogram[black];
		}

		clipped= sum;
		for(white=0; white<256; white++)
		{
			if(clipped < maxClipped) break;
			clipped-= yHistogram[white];
		}

		scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);

#ifdef HAVE_MMX2
		c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
		c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
#else
		c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
		c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
#endif

		c.packedYOffset|= c.packedYOffset<<32;
		c.packedYOffset|= c.packedYOffset<<16;

		c.packedYScale|= c.packedYScale<<32;
		c.packedYScale|= c.packedYScale<<16;
		
		if(mode & LEVEL_FIX)	QPCorrecture= (int)(scale*256*256 + 0.5);
		else			QPCorrecture= 256*256;
	}
	else
	{
		c.packedYScale= 0x0100010001000100LL;
		c.packedYOffset= 0;
		QPCorrecture= 256*256;
	}

	/* copy & deinterlace first row of blocks */
	y=-BLOCK_SIZE;
	{
		uint8_t *srcBlock= &(src[y*srcStride]);
		uint8_t *dstBlock= tempDst + dstStride;

		// From this point on it is guranteed that we can read and write 16 lines downward
		// finish 1 block before the next otherwise well might have a problem
		// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
		for(x=0; x<width; x+=BLOCK_SIZE)
		{

#ifdef HAVE_MMX2
/*
			prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
			prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
			prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
			prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
*/

			asm(
				"movl %4, %%eax			\n\t"
				"shrl $2, %%eax			\n\t"
				"andl $6, %%eax			\n\t"
				"addl %5, %%eax			\n\t"
				"movl %%eax, %%edx		\n\t"
				"imul %1, %%eax			\n\t"
				"imul %3, %%edx			\n\t"
				"prefetchnta 32(%%eax, %0)	\n\t"
				"prefetcht0 32(%%edx, %2)	\n\t"
				"addl %1, %%eax			\n\t"
				"addl %3, %%edx			\n\t"
				"prefetchnta 32(%%eax, %0)	\n\t"
				"prefetcht0 32(%%edx, %2)	\n\t"
			:: "r" (srcBlock), "r" (srcStride), "r" (dstBlock), "r" (dstStride),
			"m" (x), "m" (copyAhead)
			: "%eax", "%edx"
			);

#elif defined(HAVE_3DNOW)
//FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
/*			prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
			prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
			prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
			prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
*/
#endif

			RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
				srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

			RENAME(duplicate)(dstBlock + dstStride*8, dstStride);

			if(mode & LINEAR_IPOL_DEINT_FILTER)
				RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
			else if(mode & LINEAR_BLEND_DEINT_FILTER)
				RENAME(deInterlaceBlendLinear)(dstBlock, dstStride);
			else if(mode & MEDIAN_DEINT_FILTER)
				RENAME(deInterlaceMedian)(dstBlock, dstStride);
			else if(mode & CUBIC_IPOL_DEINT_FILTER)
				RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
			else if(mode & FFMPEG_DEINT_FILTER)
				RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
/*			else if(mode & CUBIC_BLEND_DEINT_FILTER)
				RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
*/
			dstBlock+=8;
			srcBlock+=8;
		}
		memcpy(dst, tempDst + 9*dstStride, copyAhead*dstStride );
	}

//printf("\n");
	for(y=0; y<height; y+=BLOCK_SIZE)
	{
		//1% speedup if these are here instead of the inner loop
		uint8_t *srcBlock= &(src[y*srcStride]);
		uint8_t *dstBlock= &(dst[y*dstStride]);
#ifdef HAVE_MMX
		uint8_t *tempBlock1= c.tempBlocks;
		uint8_t *tempBlock2= c.tempBlocks + 8;
#endif
#ifdef ARCH_X86
		int *QPptr= isColor ? &QPs[(y>>3)*QPStride] :&QPs[(y>>4)*QPStride];
		int QPDelta= isColor ? (-1) : 1<<31;
		int QPFrac= 1<<30;
#endif
		int QP=0;
		/* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
		   if not than use a temporary buffer */
		if(y+15 >= height)
		{
			int i;
			/* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
			   blockcopy to dst later */
			memcpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
				srcStride*MAX(height-y-copyAhead, 0) );

			/* duplicate last line of src to fill the void upto line (copyAhead+7) */
			for(i=MAX(height-y, 8); i<copyAhead+8; i++)
				memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), srcStride);

			/* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
			memcpy(tempDst, dstBlock - dstStride, dstStride*MIN(height-y+1, copyAhead+1) );

			/* duplicate last line of dst to fill the void upto line (copyAhead) */
			for(i=height-y+1; i<=copyAhead; i++)
				memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), dstStride);

			dstBlock= tempDst + dstStride;
			srcBlock= tempSrc;
		}
//printf("\n");

		// From this point on it is guranteed that we can read and write 16 lines downward
		// finish 1 block before the next otherwise well might have a problem
		// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
		for(x=0; x<width; x+=BLOCK_SIZE)
		{
			const int stride= dstStride;
#ifdef HAVE_MMX
			uint8_t *tmpXchg;
#endif
#ifdef ARCH_X86
			QP= *QPptr;
			asm volatile(
				"addl %2, %1		\n\t"
				"sbbl %%eax, %%eax	\n\t"
				"shll $2, %%eax		\n\t"
				"subl %%eax, %0		\n\t"
				: "+r" (QPptr), "+m" (QPFrac)
				: "r" (QPDelta)
				: "%eax"
			);
#else
			QP= isColor ?
                                QPs[(y>>3)*QPStride + (x>>3)]:
                                QPs[(y>>4)*QPStride + (x>>4)];
#endif
			if(!isColor)
			{
				QP= (QP* QPCorrecture + 256*128)>>16;
				yHistogram[ srcBlock[srcStride*12 + 4] ]++;
			}
//printf("%d ", QP);
			c.QP= QP;
#ifdef HAVE_MMX
			asm volatile(
				"movd %1, %%mm7					\n\t"
				"packuswb %%mm7, %%mm7				\n\t" // 0, 0, 0, QP, 0, 0, 0, QP
				"packuswb %%mm7, %%mm7				\n\t" // 0,QP, 0, QP, 0,QP, 0, QP
				"packuswb %%mm7, %%mm7				\n\t" // QP,..., QP
				"movq %%mm7, %0			\n\t"
				: "=m" (c.pQPb) 
				: "r" (QP)
			);
#endif


#ifdef HAVE_MMX2
/*
			prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
			prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
			prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
			prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
*/

			asm(
				"movl %4, %%eax			\n\t"
				"shrl $2, %%eax			\n\t"
				"andl $6, %%eax			\n\t"
				"addl %5, %%eax			\n\t"
				"movl %%eax, %%edx		\n\t"
				"imul %1, %%eax			\n\t"
				"imul %3, %%edx			\n\t"
				"prefetchnta 32(%%eax, %0)	\n\t"
				"prefetcht0 32(%%edx, %2)	\n\t"
				"addl %1, %%eax			\n\t"
				"addl %3, %%edx			\n\t"
				"prefetchnta 32(%%eax, %0)	\n\t"
				"prefetcht0 32(%%edx, %2)	\n\t"
			:: "r" (srcBlock), "r" (srcStride), "r" (dstBlock), "r" (dstStride),
			"m" (x), "m" (copyAhead)
			: "%eax", "%edx"
			);

#elif defined(HAVE_3DNOW)
//FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
/*			prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
			prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
			prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
			prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
*/
#endif

			RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
				srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

			if(mode & LINEAR_IPOL_DEINT_FILTER)
				RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
			else if(mode & LINEAR_BLEND_DEINT_FILTER)
				RENAME(deInterlaceBlendLinear)(dstBlock, dstStride);
			else if(mode & MEDIAN_DEINT_FILTER)
				RENAME(deInterlaceMedian)(dstBlock, dstStride);
			else if(mode & CUBIC_IPOL_DEINT_FILTER)
				RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
			else if(mode & FFMPEG_DEINT_FILTER)
				RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
/*			else if(mode & CUBIC_BLEND_DEINT_FILTER)
				RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
*/

			/* only deblock if we have 2 blocks */
			if(y + 8 < height)
			{
				if(mode & V_X1_FILTER)
					RENAME(vertX1Filter)(dstBlock, stride, &c);
				else if(mode & V_DEBLOCK)
				{
					if( RENAME(isVertDC)(dstBlock, stride, &c))
					{
						if(RENAME(isVertMinMaxOk)(dstBlock, stride, &c))
							RENAME(doVertLowPass)(dstBlock, stride, &c);
					}
					else
						RENAME(doVertDefFilter)(dstBlock, stride, &c);
				}
			}

#ifdef HAVE_MMX
			RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
#endif
			/* check if we have a previous block to deblock it with dstBlock */
			if(x - 8 >= 0)
			{
#ifdef HAVE_MMX
				if(mode & H_X1_FILTER)
					RENAME(vertX1Filter)(tempBlock1, 16, &c);
				else if(mode & H_DEBLOCK)
				{
					if( RENAME(isVertDC)(tempBlock1, 16, &c))
					{
						if(RENAME(isVertMinMaxOk)(tempBlock1, 16, &c))
							RENAME(doVertLowPass)(tempBlock1, 16, &c);
					}
					else
						RENAME(doVertDefFilter)(tempBlock1, 16, &c);
				}

				RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);

#else
				if(mode & H_X1_FILTER)
					horizX1Filter(dstBlock-4, stride, QP);
				else if(mode & H_DEBLOCK)
				{
					if( isHorizDC(dstBlock-4, stride, &c))
					{
						if(isHorizMinMaxOk(dstBlock-4, stride, QP))
							doHorizLowPass(dstBlock-4, stride, QP);
					}
					else
						doHorizDefFilter(dstBlock-4, stride, QP);
				}
#endif
				if(mode & DERING)
				{
				//FIXME filter first line
					if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);