postprocess_template.c

这是著名的TCPMP播放器在WINDWOWS,和WINCE下编译通过的源程序.笔者对其中的LIBMAD库做了针对ARM MPU的优化. 并增加了词幕功能.

		"movq %%mm4, %%mm3				\n\t" // d
		"psubusb "MANGLE(b01)", %%mm4			\n\t"
		PAVGB(%%mm7, %%mm4)				      // d/32
		PAVGB(%%mm7, %%mm4)				      // (d + 32)/64
		"paddb %%mm3, %%mm4				\n\t" // 5d/64
		"pand %%mm2, %%mm4				\n\t"

		"movq "MANGLE(b80)", %%mm5			\n\t" // 128
		"psubb %%mm0, %%mm5				\n\t" // q
		"paddsb %%mm6, %%mm5				\n\t" // fix bad rounding
		"pcmpgtb %%mm5, %%mm7				\n\t" // SIGN(q)
		"pxor %%mm7, %%mm5				\n\t"

		PMINUB(%%mm5, %%mm4, %%mm3)			      // MIN(|q|, 5d/64)
		"pxor %%mm1, %%mm7				\n\t" // SIGN(d*q)

		"pand %%mm7, %%mm4				\n\t"
		"movq (%%"REG_a", %1, 2), %%mm0			\n\t"
		"movq (%0, %1, 4), %%mm2			\n\t"
		"pxor %%mm1, %%mm0				\n\t"
		"pxor %%mm1, %%mm2				\n\t"
		"paddb %%mm4, %%mm0				\n\t"
		"psubb %%mm4, %%mm2				\n\t"
		"pxor %%mm1, %%mm0				\n\t"
		"pxor %%mm1, %%mm2				\n\t"
		"movq %%mm0, (%%"REG_a", %1, 2)			\n\t"
		"movq %%mm2, (%0, %1, 4)			\n\t"

		:
		: "r" (src), "r" ((long)stride), "m" (c->pQPb)
		: "%"REG_a, "%"REG_c
	);

/*
	{
	int x;
	src-= stride;
	for(x=0; x<BLOCK_SIZE; x++)
	{
		const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
		if(ABS(middleEnergy)< 8*QP)
		{
			const int q=(src[l4] - src[l5])/2;
			const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
			const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

			int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
			d= MAX(d, 0);

			d= (5*d + 32) >> 6;
			d*= SIGN(-middleEnergy);

			if(q>0)
			{
				d= d<0 ? 0 : d;
				d= d>q ? q : d;
			}
			else
			{
				d= d>0 ? 0 : d;
				d= d<q ? q : d;
			}

        		src[l4]-= d;
	        	src[l5]+= d;
		}
		src++;
	}
src-=8;
	for(x=0; x<8; x++)
	{
		int y;
		for(y=4; y<6; y++)
		{
			int d= src[x+y*stride] - tmp[x+(y-4)*8];
			int ad= ABS(d);
			static int max=0;
			static int sum=0;
			static int num=0;
			static int bias=0;

			if(max<ad) max=ad;
			sum+= ad>3 ? 1 : 0;
			if(ad>3)
			{
				src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
			}
			if(y==4) bias+=d;
			num++;
			if(num%1000000 == 0)
			{
				printf(" %d %d %d %d\n", num, sum, max, bias);
			}
		}
	}
}
*/
#elif defined (HAVE_MMX)
	src+= stride*4;
	asm volatile(
		"pxor %%mm7, %%mm7				\n\t"
		"lea -40(%%"REG_SP"), %%"REG_c"			\n\t" // make space for 4 8-byte vars
		"and "ALIGN_MASK", %%"REG_c"			\n\t" // align
//	0	1	2	3	4	5	6	7
//	%0	%0+%1	%0+2%1	eax+2%1	%0+4%1	eax+4%1	edx+%1	edx+2%1
//	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1

		"movq (%0), %%mm0				\n\t"
		"movq %%mm0, %%mm1				\n\t"
		"punpcklbw %%mm7, %%mm0				\n\t" // low part of line 0
		"punpckhbw %%mm7, %%mm1				\n\t" // high part of line 0

		"movq (%0, %1), %%mm2				\n\t"
		"lea (%0, %1, 2), %%"REG_a"			\n\t"
		"movq %%mm2, %%mm3				\n\t"
		"punpcklbw %%mm7, %%mm2				\n\t" // low part of line 1
		"punpckhbw %%mm7, %%mm3				\n\t" // high part of line 1

		"movq (%%"REG_a"), %%mm4			\n\t"
		"movq %%mm4, %%mm5				\n\t"
		"punpcklbw %%mm7, %%mm4				\n\t" // low part of line 2
		"punpckhbw %%mm7, %%mm5				\n\t" // high part of line 2

		"paddw %%mm0, %%mm0				\n\t" // 2L0
		"paddw %%mm1, %%mm1				\n\t" // 2H0
		"psubw %%mm4, %%mm2				\n\t" // L1 - L2
		"psubw %%mm5, %%mm3				\n\t" // H1 - H2
		"psubw %%mm2, %%mm0				\n\t" // 2L0 - L1 + L2
		"psubw %%mm3, %%mm1				\n\t" // 2H0 - H1 + H2

		"psllw $2, %%mm2				\n\t" // 4L1 - 4L2
		"psllw $2, %%mm3				\n\t" // 4H1 - 4H2
		"psubw %%mm2, %%mm0				\n\t" // 2L0 - 5L1 + 5L2
		"psubw %%mm3, %%mm1				\n\t" // 2H0 - 5H1 + 5H2

		"movq (%%"REG_a", %1), %%mm2			\n\t"
		"movq %%mm2, %%mm3				\n\t"
		"punpcklbw %%mm7, %%mm2				\n\t" // L3
		"punpckhbw %%mm7, %%mm3				\n\t" // H3

		"psubw %%mm2, %%mm0				\n\t" // 2L0 - 5L1 + 5L2 - L3
		"psubw %%mm3, %%mm1				\n\t" // 2H0 - 5H1 + 5H2 - H3
		"psubw %%mm2, %%mm0				\n\t" // 2L0 - 5L1 + 5L2 - 2L3
		"psubw %%mm3, %%mm1				\n\t" // 2H0 - 5H1 + 5H2 - 2H3
		"movq %%mm0, (%%"REG_c")			\n\t" // 2L0 - 5L1 + 5L2 - 2L3
		"movq %%mm1, 8(%%"REG_c")			\n\t" // 2H0 - 5H1 + 5H2 - 2H3

		"movq (%%"REG_a", %1, 2), %%mm0			\n\t"
		"movq %%mm0, %%mm1				\n\t"
		"punpcklbw %%mm7, %%mm0				\n\t" // L4
		"punpckhbw %%mm7, %%mm1				\n\t" // H4

		"psubw %%mm0, %%mm2				\n\t" // L3 - L4
		"psubw %%mm1, %%mm3				\n\t" // H3 - H4
		"movq %%mm2, 16(%%"REG_c")			\n\t" // L3 - L4
		"movq %%mm3, 24(%%"REG_c")			\n\t" // H3 - H4
		"paddw %%mm4, %%mm4				\n\t" // 2L2
		"paddw %%mm5, %%mm5				\n\t" // 2H2
		"psubw %%mm2, %%mm4				\n\t" // 2L2 - L3 + L4
		"psubw %%mm3, %%mm5				\n\t" // 2H2 - H3 + H4

		"lea (%%"REG_a", %1), %0			\n\t"
		"psllw $2, %%mm2				\n\t" // 4L3 - 4L4
		"psllw $2, %%mm3				\n\t" // 4H3 - 4H4
		"psubw %%mm2, %%mm4				\n\t" // 2L2 - 5L3 + 5L4
		"psubw %%mm3, %%mm5				\n\t" // 2H2 - 5H3 + 5H4
//50 opcodes so far
		"movq (%0, %1, 2), %%mm2			\n\t"
		"movq %%mm2, %%mm3				\n\t"
		"punpcklbw %%mm7, %%mm2				\n\t" // L5
		"punpckhbw %%mm7, %%mm3				\n\t" // H5
		"psubw %%mm2, %%mm4				\n\t" // 2L2 - 5L3 + 5L4 - L5
		"psubw %%mm3, %%mm5				\n\t" // 2H2 - 5H3 + 5H4 - H5
		"psubw %%mm2, %%mm4				\n\t" // 2L2 - 5L3 + 5L4 - 2L5
		"psubw %%mm3, %%mm5				\n\t" // 2H2 - 5H3 + 5H4 - 2H5

		"movq (%%"REG_a", %1, 4), %%mm6			\n\t"
		"punpcklbw %%mm7, %%mm6				\n\t" // L6
		"psubw %%mm6, %%mm2				\n\t" // L5 - L6
		"movq (%%"REG_a", %1, 4), %%mm6			\n\t"
		"punpckhbw %%mm7, %%mm6				\n\t" // H6
		"psubw %%mm6, %%mm3				\n\t" // H5 - H6

		"paddw %%mm0, %%mm0				\n\t" // 2L4
		"paddw %%mm1, %%mm1				\n\t" // 2H4
		"psubw %%mm2, %%mm0				\n\t" // 2L4 - L5 + L6
		"psubw %%mm3, %%mm1				\n\t" // 2H4 - H5 + H6

		"psllw $2, %%mm2				\n\t" // 4L5 - 4L6
		"psllw $2, %%mm3				\n\t" // 4H5 - 4H6
		"psubw %%mm2, %%mm0				\n\t" // 2L4 - 5L5 + 5L6
		"psubw %%mm3, %%mm1				\n\t" // 2H4 - 5H5 + 5H6

		"movq (%0, %1, 4), %%mm2			\n\t"
		"movq %%mm2, %%mm3				\n\t"
		"punpcklbw %%mm7, %%mm2				\n\t" // L7
		"punpckhbw %%mm7, %%mm3				\n\t" // H7

		"paddw %%mm2, %%mm2				\n\t" // 2L7
		"paddw %%mm3, %%mm3				\n\t" // 2H7
		"psubw %%mm2, %%mm0				\n\t" // 2L4 - 5L5 + 5L6 - 2L7
		"psubw %%mm3, %%mm1				\n\t" // 2H4 - 5H5 + 5H6 - 2H7

		"movq (%%"REG_c"), %%mm2			\n\t" // 2L0 - 5L1 + 5L2 - 2L3
		"movq 8(%%"REG_c"), %%mm3			\n\t" // 2H0 - 5H1 + 5H2 - 2H3

#ifdef HAVE_MMX2
		"movq %%mm7, %%mm6				\n\t" // 0
		"psubw %%mm0, %%mm6				\n\t"
		"pmaxsw %%mm6, %%mm0				\n\t" // |2L4 - 5L5 + 5L6 - 2L7|
		"movq %%mm7, %%mm6				\n\t" // 0
		"psubw %%mm1, %%mm6				\n\t"
		"pmaxsw %%mm6, %%mm1				\n\t" // |2H4 - 5H5 + 5H6 - 2H7|
		"movq %%mm7, %%mm6				\n\t" // 0
		"psubw %%mm2, %%mm6				\n\t"
		"pmaxsw %%mm6, %%mm2				\n\t" // |2L0 - 5L1 + 5L2 - 2L3|
		"movq %%mm7, %%mm6				\n\t" // 0
		"psubw %%mm3, %%mm6				\n\t"
		"pmaxsw %%mm6, %%mm3				\n\t" // |2H0 - 5H1 + 5H2 - 2H3|
#else
		"movq %%mm7, %%mm6				\n\t" // 0
		"pcmpgtw %%mm0, %%mm6				\n\t"
		"pxor %%mm6, %%mm0				\n\t"
		"psubw %%mm6, %%mm0				\n\t" // |2L4 - 5L5 + 5L6 - 2L7|
		"movq %%mm7, %%mm6				\n\t" // 0
		"pcmpgtw %%mm1, %%mm6				\n\t"
		"pxor %%mm6, %%mm1				\n\t"
		"psubw %%mm6, %%mm1				\n\t" // |2H4 - 5H5 + 5H6 - 2H7|
		"movq %%mm7, %%mm6				\n\t" // 0
		"pcmpgtw %%mm2, %%mm6				\n\t"
		"pxor %%mm6, %%mm2				\n\t"
		"psubw %%mm6, %%mm2				\n\t" // |2L0 - 5L1 + 5L2 - 2L3|
		"movq %%mm7, %%mm6				\n\t" // 0
		"pcmpgtw %%mm3, %%mm6				\n\t"
		"pxor %%mm6, %%mm3				\n\t"
		"psubw %%mm6, %%mm3				\n\t" // |2H0 - 5H1 + 5H2 - 2H3|
#endif

#ifdef HAVE_MMX2
		"pminsw %%mm2, %%mm0				\n\t"
		"pminsw %%mm3, %%mm1				\n\t"
#else
		"movq %%mm0, %%mm6				\n\t"
		"psubusw %%mm2, %%mm6				\n\t"
		"psubw %%mm6, %%mm0				\n\t"
		"movq %%mm1, %%mm6				\n\t"
		"psubusw %%mm3, %%mm6				\n\t"
		"psubw %%mm6, %%mm1				\n\t"
#endif

		"movd %2, %%mm2					\n\t" // QP
		"punpcklbw %%mm7, %%mm2				\n\t"

		"movq %%mm7, %%mm6				\n\t" // 0
		"pcmpgtw %%mm4, %%mm6				\n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
		"pxor %%mm6, %%mm4				\n\t"
		"psubw %%mm6, %%mm4				\n\t" // |2L2 - 5L3 + 5L4 - 2L5|
		"pcmpgtw %%mm5, %%mm7				\n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
		"pxor %%mm7, %%mm5				\n\t"
		"psubw %%mm7, %%mm5				\n\t" // |2H2 - 5H3 + 5H4 - 2H5|
// 100 opcodes
		"psllw $3, %%mm2				\n\t" // 8QP
		"movq %%mm2, %%mm3				\n\t" // 8QP
		"pcmpgtw %%mm4, %%mm2				\n\t"
		"pcmpgtw %%mm5, %%mm3				\n\t"
		"pand %%mm2, %%mm4				\n\t"
		"pand %%mm3, %%mm5				\n\t"


		"psubusw %%mm0, %%mm4				\n\t" // hd
		"psubusw %%mm1, %%mm5				\n\t" // ld


		"movq "MANGLE(w05)", %%mm2			\n\t" // 5
		"pmullw %%mm2, %%mm4				\n\t"
		"pmullw %%mm2, %%mm5				\n\t"
		"movq "MANGLE(w20)", %%mm2			\n\t" // 32
		"paddw %%mm2, %%mm4				\n\t"
		"paddw %%mm2, %%mm5				\n\t"
		"psrlw $6, %%mm4				\n\t"
		"psrlw $6, %%mm5				\n\t"

		"movq 16(%%"REG_c"), %%mm0			\n\t" // L3 - L4
		"movq 24(%%"REG_c"), %%mm1			\n\t" // H3 - H4

		"pxor %%mm2, %%mm2				\n\t"
		"pxor %%mm3, %%mm3				\n\t"

		"pcmpgtw %%mm0, %%mm2				\n\t" // sign (L3-L4)
		"pcmpgtw %%mm1, %%mm3				\n\t" // sign (H3-H4)
		"pxor %%mm2, %%mm0				\n\t"
		"pxor %%mm3, %%mm1				\n\t"
		"psubw %%mm2, %%mm0				\n\t" // |L3-L4|
		"psubw %%mm3, %%mm1				\n\t" // |H3-H4|
		"psrlw $1, %%mm0				\n\t" // |L3 - L4|/2
		"psrlw $1, %%mm1				\n\t" // |H3 - H4|/2

		"pxor %%mm6, %%mm2				\n\t"
		"pxor %%mm7, %%mm3				\n\t"
		"pand %%mm2, %%mm4				\n\t"
		"pand %%mm3, %%mm5				\n\t"

#ifdef HAVE_MMX2
		"pminsw %%mm0, %%mm4				\n\t"
		"pminsw %%mm1, %%mm5				\n\t"
#else
		"movq %%mm4, %%mm2				\n\t"
		"psubusw %%mm0, %%mm2				\n\t"
		"psubw %%mm2, %%mm4				\n\t"
		"movq %%mm5, %%mm2				\n\t"
		"psubusw %%mm1, %%mm2				\n\t"
		"psubw %%mm2, %%mm5				\n\t"
#endif
		"pxor %%mm6, %%mm4				\n\t"
		"pxor %%mm7, %%mm5				\n\t"
		"psubw %%mm6, %%mm4				\n\t"
		"psubw %%mm7, %%mm5				\n\t"
		"packsswb %%mm5, %%mm4				\n\t"
		"movq (%0), %%mm0				\n\t"
		"paddb   %%mm4, %%mm0				\n\t"
		"movq %%mm0, (%0)				\n\t"
		"movq (%0, %1), %%mm0				\n\t"
		"psubb %%mm4, %%mm0				\n\t"
		"movq %%mm0, (%0, %1)				\n\t"

		: "+r" (src)
		: "r" ((long)stride), "m" (c->pQPb)
		: "%"REG_a, "%"REG_c
	);
#else
	const int l1= stride;
	const int l2= stride + l1;
	const int l3= stride + l2;
	const int l4= stride + l3;
	const int l5= stride + l4;
	const int l6= stride + l5;
	const int l7= stride + l6;
	const int l8= stride + l7;
//	const int l9= stride + l8;
	int x;
	src+= stride*3;
	for(x=0; x<BLOCK_SIZE; x++)
	{
		const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
		if(ABS(middleEnergy) < 8*c->QP)
		{
			const int q=(src[l4] - src[l5])/2;
			const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
			const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

			int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
			d= MAX(d, 0);

			d= (5*d + 32) >> 6;
			d*= SIGN(-middleEnergy);

			if(q>0)
			{
				d= d<0 ? 0 : d;
				d= d>q ? q : d;
			}
			else
			{
				d= d>0 ? 0 : d;
				d= d<q ? q : d;
			}

        		src[l4]-= d;
	        	src[l5]+= d;
		}
		src++;
	}
#endif
}
#endif //HAVE_ALTIVEC

#ifndef HAVE_ALTIVEC
static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
{
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
	asm volatile(
		"pxor %%mm6, %%mm6				\n\t"
		"pcmpeqb %%mm7, %%mm7				\n\t"
		"movq %2, %%mm0					\n\t"
		"punpcklbw %%mm6, %%mm0				\n\t"
		"psrlw $1, %%mm0				\n\t"
		"psubw %%mm7, %%mm0				\n\t"
		"packuswb %%mm0, %%mm0				\n\t"
		"movq %%mm0, %3					\n\t"

		"lea (%0, %1), %%"REG_a"			\n\t"
		"lea (%%"REG_a", %1, 4), %%"REG_d"		\n\t"
		
//	0	1	2	3	4	5	6	7	8	9
//	%0	eax	eax+%1	eax+2%1	%0+4%1	edx	edx+%1	edx+2%1	%0+8%1	edx+4%1

#undef FIND_MIN_MAX
#ifdef HAVE_MMX2
#define REAL_FIND_MIN_MAX(addr)\
		"movq " #addr ", %%mm0				\n\t"\
		"pminub %%mm0, %%mm7				\n\t"\