postprocess.c

#endif // MMX
#else
	//FIXME
	int x;
	src+= 4*stride;
	for(x=0; x<8; x++)
	{
		src[0       ] = (src[0       ] + 2*src[stride  ] + src[stride*2])>>2;
		src[stride  ] = (src[stride  ] + 2*src[stride*2] + src[stride*3])>>2;
		src[stride*2] = (src[stride*2] + 2*src[stride*3] + src[stride*4])>>2;
		src[stride*3] = (src[stride*3] + 2*src[stride*4] + src[stride*5])>>2;
		src[stride*4] = (src[stride*4] + 2*src[stride*5] + src[stride*6])>>2;
		src[stride*5] = (src[stride*5] + 2*src[stride*6] + src[stride*7])>>2;
		src[stride*6] = (src[stride*6] + 2*src[stride*7] + src[stride*8])>>2;
		src[stride*7] = (src[stride*7] + 2*src[stride*8] + src[stride*9])>>2;
		src++;
	}
#endif
}

#ifdef HAVE_MMX
/**
 * transposes and shift the given 8x8 Block into dst1 and dst2
 */
static inline void transpose1(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
{
	asm(
		"leal (%0, %1), %%eax				\n\t"
		"leal (%%eax, %1, 4), %%ebx			\n\t"
//	0	1	2	3	4	5	6	7	8	9
//	%0	eax	eax+%1	eax+2%1	%0+4%1	ebx	ebx+%1	ebx+2%1	%0+8%1	ebx+4%1
		"movq (%0), %%mm0		\n\t" // 12345678
		"movq (%%eax), %%mm1		\n\t" // abcdefgh
		"movq %%mm0, %%mm2		\n\t" // 12345678
		"punpcklbw %%mm1, %%mm0		\n\t" // 1a2b3c4d
		"punpckhbw %%mm1, %%mm2		\n\t" // 5e6f7g8h

		"movq (%%eax, %1), %%mm1	\n\t"
		"movq (%%eax, %1, 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, 128(%2)		\n\t"
		"psrlq $32, %%mm0		\n\t"
		"movd %%mm0, 144(%2)		\n\t"
		"movd %%mm3, 160(%2)		\n\t"
		"psrlq $32, %%mm3		\n\t"
		"movd %%mm3, 176(%2)		\n\t"
		"movd %%mm3, 48(%3)		\n\t"
		"movd %%mm2, 192(%2)		\n\t"
		"movd %%mm2, 64(%3)		\n\t"
		"psrlq $32, %%mm2		\n\t"
		"movd %%mm2, 80(%3)		\n\t"
		"movd %%mm1, 96(%3)		\n\t"
		"psrlq $32, %%mm1		\n\t"
		"movd %%mm1, 112(%3)		\n\t"

		"movq (%0, %1, 4), %%mm0	\n\t" // 12345678
		"movq (%%ebx), %%mm1		\n\t" // abcdefgh
		"movq %%mm0, %%mm2		\n\t" // 12345678
		"punpcklbw %%mm1, %%mm0		\n\t" // 1a2b3c4d
		"punpckhbw %%mm1, %%mm2		\n\t" // 5e6f7g8h

		"movq (%%ebx, %1), %%mm1	\n\t"
		"movq (%%ebx, %1, 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, 132(%2)		\n\t"
		"psrlq $32, %%mm0		\n\t"
		"movd %%mm0, 148(%2)		\n\t"
		"movd %%mm3, 164(%2)		\n\t"
		"psrlq $32, %%mm3		\n\t"
		"movd %%mm3, 180(%2)		\n\t"
		"movd %%mm3, 52(%3)		\n\t"
		"movd %%mm2, 196(%2)		\n\t"
		"movd %%mm2, 68(%3)		\n\t"
		"psrlq $32, %%mm2		\n\t"
		"movd %%mm2, 84(%3)		\n\t"
		"movd %%mm1, 100(%3)		\n\t"
		"psrlq $32, %%mm1		\n\t"
		"movd %%mm1, 116(%3)		\n\t"


	:: "r" (src), "r" (srcStride), "r" (dst1), "r" (dst2)
	: "%eax", "%ebx"
	);
}

/**
 * transposes the given 8x8 block
 */
static inline void transpose2(uint8_t *dst, int dstStride, uint8_t *src)
{
	asm(
		"leal (%0, %1), %%eax				\n\t"
		"leal (%%eax, %1, 4), %%ebx			\n\t"
//	0	1	2	3	4	5	6	7	8	9
//	%0	eax	eax+%1	eax+2%1	%0+4%1	ebx	ebx+%1	ebx+2%1	%0+8%1	ebx+4%1
		"movq (%2), %%mm0		\n\t" // 12345678
		"movq 16(%2), %%mm1		\n\t" // abcdefgh
		"movq %%mm0, %%mm2		\n\t" // 12345678
		"punpcklbw %%mm1, %%mm0		\n\t" // 1a2b3c4d
		"punpckhbw %%mm1, %%mm2		\n\t" // 5e6f7g8h

		"movq 32(%2), %%mm1		\n\t"
		"movq 48(%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, (%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, (%%ebx)		\n\t"
		"movd %%mm1, (%%ebx, %1)	\n\t"
		"psrlq $32, %%mm1		\n\t"
		"movd %%mm1, (%%ebx, %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(%%ebx)		\n\t"
		"movd %%mm1, 4(%%ebx, %1)	\n\t"
		"psrlq $32, %%mm1		\n\t"
		"movd %%mm1, 4(%%ebx, %1, 2)	\n\t"

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

static void inline tempNoiseReducer(uint8_t *src, int stride,
				    uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
{
#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), %%ebx			\n\t" // 5*stride
		"leal (%%ebx, %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+ebx	%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, %%ebx), %%mm5			\n\t" // L5
		"paddw %%mm2, %%mm0				\n\t"
		"psadbw (%1, %%ebx), %%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 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, %%ebx), (%1, %%ebx))
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, %%ebx), (%1, %%ebx))
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, %%ebx					\n\t"
		"addl -4(%%ebx), %%ecx				\n\t"
		"addl 4(%%ebx), %%ecx				\n\t"
		"addl -1024(%%ebx), %%ecx			\n\t"
		"addl $4, %%ecx					\n\t"
		"addl 1024(%%ebx), %%ecx			\n\t"
		"shrl $3, %%ecx					\n\t"
		"movl %%ecx, (%%ebx)				\n\t"
		"leal (%%eax, %2, 2), %%ebx			\n\t" // 5*stride

//		"movl %3, %%ecx				\n\t"
//		"movl %%ecx, test				\n\t"
//		"jmp 4f \n\t"
		"cmpl 4+maxTmpNoise, %%ecx			\n\t"
		" jb 2f						\n\t"
		"cmpl 8+maxTmpNoise, %%ecx			\n\t"
		" jb 1f						\n\t"

		"leal (%%ebx, %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, %%ebx), %%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, %%ebx)			\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 (%%ebx, %2, 2), %%ecx			\n\t" // 7*stride
		"movq (%0), %%mm0				\n\t" // L0
		"pavgb (%1), %%mm0				\n\t" // L0
		"movq (%0, %2), %%mm1				\n\t" // L1
		"pavgb (%1, %2), %%mm1				\n\t" // L1
		"movq (%0, %2, 2), %%mm2			\n\t" // L2
		"pavgb (%1, %2, 2), %%mm2			\n\t" // L2
		"movq (%0, %%eax), %%mm3			\n\t" // L3
		"pavgb (%1, %%eax), %%mm3			\n\t" // L3
		"movq (%0, %2, 4), %%mm4			\n\t" // L4
		"pavgb (%1, %2, 4), %%mm4			\n\t" // L4
		"movq (%0, %%ebx), %%mm5			\n\t" // L5
		"pavgb (%1, %%ebx), %%mm5			\n\t" // L5
		"movq (%0, %%eax, 2), %%mm6			\n\t" // L6
		"pavgb (%1, %%eax, 2), %%mm6			\n\t" // L6
		"movq (%0, %%ecx), %%mm7			\n\t" // L7
		"pavgb (%1, %%ecx), %%mm7			\n\t" // 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, %%ebx)			\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, %%ebx)			\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 maxTmpNoise, %%ecx			\n\t"
		" jb 3f						\n\t"

		"leal (%%ebx, %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, %%ebx), %%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, %%ebx), %%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, %%ebx)			\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, %%ebx)			\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 (%%ebx, %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, %%ebx), %%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, %%ebx), %%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, %%ebx)			\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, %%ebx)			\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", "%ebx", "%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
}

#ifdef HAVE_ODIVX_POSTPROCESS
#include "../opendivx/postprocess.h"
int use_old_pp=0;
#endif

static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
	QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode);

/* -pp Command line Help
NOTE/FIXME: put this at an appropriate place (--help, html docs, man mplayer)?

-pp <filterName>[:<option>[:<option>...]][,[-]<filterName>[:<option>...]]...

long form example:
-pp vdeblock:autoq,hdeblock:autoq,linblenddeint		-pp default,-vdeblock
short form example:
-pp vb:a,hb:a,lb					-pp de,-vb
more examples:
-pp tn:64:128:256

Filters			Options
short	long name	short	long option	Description
*	*		a	autoq		cpu power dependant enabler
			c	chrom		chrominance filtring enabled
			y	nochrom		chrominance filtring disabled
hb	hdeblock				horizontal deblocking filter
vb	vdeblock				vertical deblocking filter
vr	rkvdeblock
h1	x1hdeblock				Experimental horizontal deblock filter 1
v1	x1vdeblock				Experimental vertical deblock filter 1
dr	dering					not implemented yet
al	autolevels				automatic brightness / contrast fixer
			f	fullyrange	stretch luminance range to (0..255)
lb	linblenddeint				linear blend deinterlacer
li	linipoldeint				linear interpolating deinterlacer
ci	cubicipoldeint				cubic interpolating deinterlacer
md	mediandeint				median deinterlacer
de	default					hdeblock:a,vdeblock:a,dering:a,autolevels
fa	fast					x1hdeblock:a,x1vdeblock:a,dering:a,autolevels
tn	tmpnoise	(3 Thresholds)		Temporal Noise Reducer
*/

/**
 * returns a PPMode struct which will have a non 0 error variable if an error occured
 * name is the string after "-pp" on the command line
 * quality is a number from 0 to GET_PP_QUALITY_MAX
 */
struct PPMode getPPModeByNameAndQuality(char *name, int quality)
{
	char temp[GET_MODE_BUFFER_SIZE];
	char *p= temp;
	char *filterDelimiters= ",";
	char *optionDelimiters= ":";
	struct PPMode ppMode= {0,0,0,0,0,0,{150,200,400}};
	char *filterToken;

	strncpy(temp, name, GET_MODE_BUFFER_SIZE);

	printf("%s\n", name);

	for(;;){
		char *filterName;
		int q= 1000000; //GET_PP_QUALITY_MAX;
		int chrom=-1;
		char *option;
		char *options[OPTIONS_ARRAY_SIZE];
		int i;
		int filterNameOk=0;
		int numOfUnknownOptions=0;
		int enable=1; //does the user want us to enabled or disabled the filter

		filterToken= strtok(p, filterDelimiters);
		if(filterToken == NULL) break;
		p+= strlen(filterToken) + 1; // p points to next filterToken
		filterName= strtok(filterToken, optionDelimiters);
		printf("%s::%s\n", filterToken, filterName);

		if(*filterName == '-')
		{
			enable=0;
			filterName++;
		}

		for(;;){ //for all options
			option= strtok(NULL, optionDelimiters);
			if(option == NULL) break;

			printf("%s\n", option);
			if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
			else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
			else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
			else
			{
				options[numOfUnknownOptions] = option;
				numOfUnknownOptions++;
			}
			if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
		}
		options[numOfUnknownOptions] = NULL;

		/* replace stuff from the replace Table */
		for(i=0; replaceTable[2*i]!=NULL; i++)
		{
			if(!strcmp(replaceTable[2*i], filterName))
			{
				int newlen= strlen(replaceTable[2*i + 1]);
				int plen;
				int spaceLeft;

				if(p==NULL) p= temp, *p=0; 	//last filter
				else p--, *p=',';		//not last filter

				plen= strlen(p);
				spaceLeft= (int)p - (int)temp + plen;
				if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE)
				{
					ppMode.error++;
					break;
				}
				memmove(p + newlen, p, plen+1);
				memcpy(p, replaceTable[2*i + 1], newlen);
				filterNameOk=1;
			}
		}

		for(i=0; filters[i].shortName!=NULL; i++)
		{
//			printf("Compareing %s, %s, %s\n", filters[i].shortName,filters[i].longName, filterName);
			if(   !strcmp(filters[i].longName, filterName)
			   || !strcmp(filters[i].shortName, filterName))
			{
				ppMode.lumMode &= ~filters[i].mask;
				ppMode.chromMode &= ~filters[i].mask;

				filterNameOk=1;
				if(!enable) break; // user wants to disable it

				if(q >= filters[i].minLumQuality)
					ppMode.lumMode|= filters[i].mask;
				if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
					if(q >= filters[i].minChromQuality)
						ppMode.chromMode|= filters[i].mask;

				if(filters[i].mask == LEVEL_FIX)
				{
					int o;
					ppMode.minAllowedY= 16;
					ppMode.maxAllowedY= 234;
					for(o=0; options[o]!=NULL; o++)
						if(  !strcmp(options[o],"fullyrange")
						   ||!strcmp(options[o],"f"))
						{
							ppMode.minAllowedY= 0;
							ppMode.maxAllowedY= 255;
							numOfUnknownOptions--;
						}
				}
				else if(filters[i].mask == TEMP_NOISE_FILTER)
				{
					int o;
					int numOfNoises=0;
					ppMode.maxTmpNoise[0]= 150;
					ppMode.maxTmpNoise[1]= 200;
					ppMode.maxTmpNoise[2]= 400;

					for(o=0; options[o]!=NULL; o++)
					{
						char *tail;
						ppMode.maxTmpNoise[numOfNoises]=
							strtol(options[o], &tail, 0);
						if(tail!=options[o])
						{
							numOfNoises++;
							numOfUnknownOptions--;
							if(numOfNoises >= 3) break;
						}
					}
				}
			}
		}
		if(!filterNameOk) ppMode.error++;
		ppMode.error += numOfUnknownOptions;
	}

#ifdef HAVE_ODIVX_POSTPROCESS
	if(ppMode.lumMode & H_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_Y_H;
	if(ppMode.lumMode & V_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_Y_V;
	if(ppMode.chromMode & H_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_C_H;
	if(ppMode.chromMode & V_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_C_V;
	if(ppMode.lumMode & DERING) ppMode.oldMode |= PP_DERING_Y;
	if(ppMode.chromMode & DERING) ppMode.oldMode |= PP_DERING_C;
#endif

	return ppMode;
}

/**
 * Obsolete, dont use it, use postprocess2() instead
 */
void  postprocess(unsigned char * src[], int src_stride,
                 unsigned char * dst[], int dst_stride,
                 int horizontal_size,   int vertical_size,
                 QP_STORE_T *QP_store,  int QP_stride,
					  int mode)
{
	struct PPMode ppMode;
	static QP_STORE_T zeroArray[2048/8];
/*
	static int qual=0;

	ppMode= getPPModeByNameAndQuality("fast,default,-hdeblock,-vdeblock,tmpnoise:150:200:300", qual);
	printf("OK\n");
	qual++;
	qual%=7;
	printf("\n%X %X %X %X :%d: %d %d %d\n", ppMode.lumMode, ppMode.chromMode, ppMode.oldMode, ppMode.error,
		qual, ppMode.maxTmpNoise[0], ppMode.maxTmpNoise[1], ppMode.maxTmpNoise[2]);
	postprocess2(src, src_stride, dst, dst_stride,
                 horizontal_size, vertical_size, QP_store, QP_stride, &ppMode);

	return;
*/
	if(QP_store==NULL)
	{
		QP_store= zeroArray;
		QP_stride= 0;
	}

	ppMode.lumMode= mode;
	mode= ((mode&0xFF)>>4) | (mode&0xFFFFFF00);
	ppMode.chromMode= mode;
	ppMode.maxTmpNoise[0]= 700;
	ppMode.maxTmpNoise[1]= 1500;
	ppMode.maxTmpNoise[2]= 3000;

#ifdef HAVE_ODIVX_POSTPROCESS
// Note: I could make this shit outside of this file, but it would mean one
// more function call...
	if(use_old_pp){
	    odivx_postprocess(src,src_stride,dst,dst_stride,horizontal_size,vertical_size,QP_store,QP_stride,mode);
	    return;
	}
#endif

	postProcess(src[0], src_stride, dst[0], dst_stride,
		horizontal_size, vertical_size, QP_store, QP_stride, 0, &ppMode);

	horizontal_size >>= 1;
	vertical_size   >>= 1;
	src_stride      >>= 1;
	dst_stride      >>= 1;

	if(1)
	{
		postProcess(src[1], src_stride, dst[1], dst_stride,
			horizontal_size, vertical_size, QP_store, QP_stride, 1, &ppMode);
		postProcess(src[2], src_stride, dst[2], dst_stride,
			horizontal_size, vertical_size, QP_store, QP_stride, 2, &ppMode);
	}
	else
	{
		memset(dst[1], 128, dst_stride*vertical_size);
		memset(dst[2], 128, dst_stride*vertical_size);
//		memcpy(dst[1], src[1], src_stride*horizontal_size);
//		memcpy(dst[2], src[2], src_stride*horizontal_size);
	}
}

void  postprocess2(unsigned char * src[], int src_stride,
                 unsigned char * dst[], int dst_stride,
                 int horizontal_size,   int vertical_size,
                 QP_STORE_T *QP_store,  int QP_stride,
		 struct PPMode *mode)
{

	static QP_STORE_T zeroArray[2048/8];
	if(QP_store==NULL)
	{
		QP_store= zeroArray;
		QP_stride= 0;
	}

#ifdef HAVE_ODIVX_POSTPROCESS
// Note: I could make this shit outside of this file, but it would mean one
// more function call...
	if(use_old_pp){
	    odivx_postprocess(src,src_stride,dst,dst_stride,horizontal_size,vertical_size,QP_store,QP_stride,
	    mode->oldMode);
	    return;
	}
#endif

	postProcess(src[0], src_stride, dst[0], dst_stride,
		horizontal_size, vertical_size, QP_store, QP_stride, 0, mode);

	horizontal_size >>= 1;
	vertical_size   >>= 1;
	src_stride      >>= 1;
	dst_stride      >>= 1;

	postProcess(src[1], src_stride, dst[1], dst_stride,
		horizontal_size, vertical_size, QP_store, QP_stride, 1, mode);
	postProcess(src[2], src_stride, dst[2], dst_stride,
		horizontal_size, vertical_size, QP_store, QP_stride, 2, mode);
}


/**
 * gets the mode flags for a given quality (larger values mean slower but better postprocessing)
 * 0 <= quality <= 6
 */
int getPpModeForQuality(int quality){
	int modes[1+GET_PP_QUALITY_MAX]= {
		0,
#if 1
		// horizontal filters first
		LUM_H_DEBLOCK,
		LUM_H_DEBLOCK | LUM_V_DEBLOCK,
		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK,
		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK,
		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK | LUM_DERING,
		LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK | LUM_DERING | CHROM_DERING
#else
		// vertical filters first
		LUM_V_DEBLOCK,
		LUM_V_DEBLOCK | LUM_H_DEBLOCK,
		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK,
		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK,
		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK | LUM_DERING,
		LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK | LUM_DERING | CHROM_DERING
#endif
	};

#ifdef HAVE_ODIVX_POSTPROCESS
	int odivx_modes[1+GET_PP_QUALITY_MAX]= {
		0,
		PP_DEBLOCK_Y_H,
		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V,
		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H,
		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V,
		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V|PP_DERING_Y,
		PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V|PP_DERING_Y|PP_DERING_C
	};
	if(use_old_pp) return odivx_modes[quality];
#endif
	return modes[quality];
}

/**
 * Copies a block from src to dst and fixes the blacklevel
 * numLines must be a multiple of 4
 * levelFix == 0 -> dont touch the brighness & contrast
 */
static inline void blockCopy(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
	int levelFix)
{
#ifndef HAVE_MMX
	int i;
#endif
	if(levelFix)
	{
#ifdef HAVE_MMX
					asm volatile(
						"leal (%2,%2), %%eax	\n\t"
						"leal (%3,%3), %%ebx	\n\t"
						"movq packedYOffset, %%mm2	\n\t"
						"movq packedYScale, %%mm3	\n\t"
						"pxor %%mm4, %%mm4	\n\t"

#define SCALED_CPY					\
						"movq (%0), %%mm0	\n\t"\
						"movq (%0), %%mm5	\n\t"\
						"punpcklbw %%mm4, %%mm0 \n\t"\
						"punpckhbw %%mm4, %%mm5 \n\t"\
						"psubw %%mm2, %%mm0	\n\t"\
						"psubw %%mm2, %%mm5	\n\t"\
						"movq (%0,%2), %%mm1	\n\t"\
						"psllw $6, %%mm0	\n\t"\
						"psllw $6, %%mm5	\n\t"\
						"pmulhw %%mm3, %%mm0	\n\t"\
						"movq (%0,%2), %%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"\
						"addl %%eax, %0		\n\t"\
						"packuswb %%mm5, %%mm0	\n\t"\
						"packuswb %%mm6, %%mm1	\n\t"\
						"movq %%mm0, (%1)	\n\t"\
						"movq %%mm1, (%1, %3)	\n\t"\

SCALED_CPY
						"addl %%ebx, %1		\n\t"
SCALED_CPY
						"addl %%ebx, %1		\n\t"
SCALED_CPY
						"addl %%ebx, %1		\n\t"
SCALED_CPY

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