predict.c

avi2mpg1_src 中包含了mpeg1编码的源程序

				_asm
				{
					mov esi, s					;// esi = s
					mov edi, d					;// edi = d
					mov ecx, h					;// ecx = h
					mov edx, lx					;// edx = lx
pred_comp__l8:
					movd mm0, [esi]				;// lower 4 bytes into mm0 = esi[0..3]
					movd mm1, [esi+edx]			;// lower 4 bytes into mm1 = (esi + edx)[0..3]
					movd mm2, [esi+4]			;// lower 4 bytes into mm2 = esi[4..7]
					movd mm3, [esi+edx+4]		;// lower 4 bytes into mm3 = (esi + edx)[4..7]
					movd mm4, [esi+8]			;// lower 4 bytes into mm4 = esi[8..11]
					movd mm5, [esi+edx+8]		;// lower 4 bytes into mm5 = (esi + edx)[8..11]
					movd mm6, [esi+12]			;// lower 4 bytes into mm6 = esi[12..15]
					movd mm7, [esi+edx+12]		;// lower 4 bytes into mm7 = (esi + edx)[12..15]
					punpcklbw mm0, PACKED_0		;// unpack the lower 4 bytes into mm0
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm2, PACKED_0		;// unpack the lower 4 bytes into mm2
					punpcklbw mm3, PACKED_0		;// unpack the lower 4 bytes into mm3
					punpcklbw mm4, PACKED_0		;// unpack the lower 4 bytes into mm4
					punpcklbw mm5, PACKED_0		;// unpack the lower 4 bytes into mm5
					punpcklbw mm6, PACKED_0		;// unpack the lower 4 bytes into mm6
					punpcklbw mm7, PACKED_0		;// unpack the lower 4 bytes into mm7
					paddw mm0, mm1
					paddw mm2, mm3
					paddw mm4, mm5
					paddw mm6, mm7
					paddw mm0, PACKED_1			;// mm0 += mm1 + (1, 1, 1, 1)
					paddw mm2, PACKED_1			;// mm2 += mm3 + (1, 1, 1, 1)
					paddw mm4, PACKED_1			;// mm4 += mm5 + (1, 1, 1, 1)
					paddw mm6, PACKED_1			;// mm6 += mm7 + (1, 1, 1, 1)
					psrlw mm0, 1				;// mm0 >>= 1
					psrlw mm2, 1				;// mm2 >>= 1
					psrlw mm4, 1				;// mm4 >>= 1
					psrlw mm6, 1				;// mm6 >>= 1
					packuswb mm0, mm2			;// pack mm0 and mm2
					packuswb mm4, mm6			;// pack mm4 and mm6
					movq [edi], mm0				;// store mm0 into edi[0..7]
					movq [edi+8], mm4			;// store mm4 into edi[8..15]

					add esi, edx				;// esi += edx
					add edi, edx				;// edi += edx
					dec ecx						;// decrement ecx
					jnz pred_comp__l8			;// loop while not zero
					emms						;// empty MMX state
				}
				return;
			}
		}

		for (j=0; j<h; j++)
		{
			for (i=0; i<w; i++)
				d[i] = (unsigned int)(s[i]+s[i+lx]+1)>>1;
				
			s+= lx;
			d+= lx;
		}
	}
	else if (xh && !yh)
	if (addflag)
	{
		if(cpu_MMX)
		{
			if(w == 8)
			{
				_asm
				{
					mov esi, s					;// esi = s
					mov edi, d					;// edi = d
					mov edx, lx					;// edx = lx
					mov ecx, h					;// ecx = h
pred_comp__l9:
					movd mm0, [esi]				;// lower 4 bytes into mm0 = esi[0..3]
					movd mm1, [esi+1]			;// lower 4 bytes into mm1 = esi[1..4]
					movd mm2, [esi+4]			;// lower 4 bytes into mm2 = esi[4..7]
					movd mm3, [esi+5]			;// lower 4 bytes into mm3 = esi[5..8]
					punpcklbw mm0, PACKED_0		;// unpack the lower 4 bytes into mm0
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm2, PACKED_0		;// unpack the lower 4 bytes into mm2
					punpcklbw mm3, PACKED_0		;// unpack the lower 4 bytes into mm3
					paddw mm0, mm1
					paddw mm2, mm3
					paddw mm0, PACKED_1			;// mm0 += mm1 + (1, 1, 1, 1)
					paddw mm2, PACKED_1			;// mm2 += mm3 + (1, 1, 1, 1)
					movd mm1, [edi]				;// lower 4 bytes into mm1 = edi[0..3]
					movd mm3, [edi+4]			;// lower 4 bytes into mm3 = edi[4..7]
					psrlw mm0, 1				;// mm0 >>= 1
					psrlw mm2, 1				;// mm2 >>= 1
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm3, PACKED_0		;// unpack the lower 4 bytes into mm3
					paddw mm0, PACKED_1
					paddw mm2, PACKED_1
					paddw mm0, mm1				;// mm0 += mm1 + (1, 1, 1, 1)
					paddw mm2, mm3				;// mm2 += mm3 + (1, 1, 1, 1)
					psrlw mm0, 1				;// mm0 >>= 1
					psrlw mm2, 1				;// mm2 >>= 1
					packuswb mm0, mm2			;// pack mm0 and mm2
					movq [edi], mm0				;// store mm0 into edi[0..7]
					add esi, edx				;// esi += edx
					add edi, edx				;// edi += edx
					dec ecx						;// decrement ecx
					jnz pred_comp__l9			;// loop while not zero
					emms						;// empty MMX state
				}
				return;
			}

			if(w == 16)
			{
				_asm
				{
					mov esi, s					;// esi = s
					mov edi, d					;// edi = d
					mov edx, lx					;// edx = lx
					mov ecx, h					;// ecx = h
pred_comp__l10:
					movd mm0, [esi]				;// lower 4 bytes into mm0 = esi[0..3]
					movd mm1, [esi+1]			;// lower 4 bytes into mm1 = esi[1..4]
					movd mm2, [esi+4]			;// lower 4 bytes into mm2 = esi[4..7]
					movd mm3, [esi+5]			;// lower 4 bytes into mm3 = esi[5..8]
					movd mm4, [esi+8]			;// lower 4 bytes into mm4 = esi[8..11]
					movd mm5, [esi+9]			;// lower 4 bytes into mm5 = esi[9..12]
					movd mm6, [esi+12]			;// lower 4 bytes into mm6 = esi[12..15]
					movd mm7, [esi+13]			;// lower 4 bytes into mm7 = esi[13..16]
					punpcklbw mm0, PACKED_0		;// unpack the lower 4 bytes into mm0
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm2, PACKED_0		;// unpack the lower 4 bytes into mm2
					punpcklbw mm3, PACKED_0		;// unpack the lower 4 bytes into mm3
					punpcklbw mm4, PACKED_0		;// unpack the lower 4 bytes into mm4
					punpcklbw mm5, PACKED_0		;// unpack the lower 4 bytes into mm5
					punpcklbw mm6, PACKED_0		;// unpack the lower 4 bytes into mm6
					punpcklbw mm7, PACKED_0		;// unpack the lower 4 bytes into mm7
					paddw mm0, mm1
					paddw mm2, mm3
					paddw mm4, mm5
					paddw mm6, mm7
					paddw mm0, PACKED_1			;// mm0 += mm1 + (1, 1, 1, 1)
					paddw mm2, PACKED_1			;// mm2 += mm3 + (1, 1, 1, 1)
					paddw mm4, PACKED_1			;// mm4 += mm5 + (1, 1, 1, 1)
					paddw mm6, PACKED_1			;// mm6 += mm7 + (1, 1, 1, 1)
					movd mm1, [edi]				;// lower 4 bytes into mm1 = edi[0..3]
					movd mm3, [edi+4]			;// lower 4 bytes into mm3 = edi[4..7]
					movd mm5, [edi+8]			;// lower 4 bytes into mm5 = edi[8..11]
					movd mm7, [edi+12]			;// lower 4 bytes into mm7 = edi[12..15]
					psrlw mm0, 1				;// mm0 >>= 1
					psrlw mm2, 1				;// mm2 >>= 1
					psrlw mm4, 1				;// mm4 >>= 1
					psrlw mm6, 1				;// mm6 >>= 1
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm3, PACKED_0		;// unpack the lower 4 bytes into mm3
					punpcklbw mm5, PACKED_0		;// unpack the lower 4 bytes into mm5
					punpcklbw mm7, PACKED_0		;// unpack the lower 4 bytes into mm7
					paddw mm0, PACKED_1
					paddw mm2, PACKED_1
					paddw mm4, PACKED_1
					paddw mm6, PACKED_1
					paddw mm0, mm1				;// mm0 += mm1 + (1, 1, 1, 1)
					paddw mm2, mm3				;// mm2 += mm3 + (1, 1, 1, 1)
					paddw mm4, mm5				;// mm4 += mm5 + (1, 1, 1, 1)
					paddw mm6, mm7				;// mm6 += mm7 + (1, 1, 1, 1)
					psrlw mm0, 1				;// mm0 >>= 1
					psrlw mm2, 1				;// mm2 >>= 1
					psrlw mm4, 1				;// mm4 >>= 1
					psrlw mm6, 1				;// mm6 >>= 1
					packuswb mm0, mm2			;// pack mm0 and mm2
					packuswb mm4, mm6			;// pack mm4 and mm6
					movq [edi], mm0				;// store mm0 into edi[0..7]
					movq [edi+8], mm4			;// store mm4 into edi[8..15]
					add esi, edx				;// esi += edx
					add edi, edx				;// edi += edx
					dec ecx						;// decrement ecx
					jnz pred_comp__l10			;// loop while not zero
					emms						;// empty MMX state
				}
				return;
			}
		}

		for (j=0; j<h; j++)
		{
			for (i=0; i<w; i++)
				d[i] = (d[i] + ((unsigned int)(s[i]+s[i+1]+1)>>1)+1)>>1;
				
			s+= lx;
			d+= lx;
		}
	}
	else
	{
		if(cpu_MMX)
		{
			if(w == 8)
			{
				_asm
				{
					mov esi, s					;// esi = s
					mov edi, d					;// edi = d
					mov edx, lx					;// edx = lx
					mov ecx, h					;// ecx = h
pred_comp__l11:
					movd mm0, [esi]				;// lower 4 bytes into mm0 = esi[0..3]
					movd mm1, [esi+1]			;// lower 4 bytes into mm1 = esi[1..4]
					movd mm2, [esi+4]			;// lower 4 bytes into mm2 = esi[4..7]
					movd mm3, [esi+5]			;// lower 4 bytes into mm3 = esi[0..3]
					punpcklbw mm0, PACKED_0		;// unpack the lower 4 bytes into mm0
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm2, PACKED_0		;// unpack the lower 4 bytes into mm2
					punpcklbw mm3, PACKED_0		;// unpack the lower 4 bytes into mm3
					paddw mm0, mm1
					paddw mm2, mm3
					paddw mm0, PACKED_1			;// mm0 += mm1 + (1, 1, 1, 1)
					paddw mm2, PACKED_1			;// mm2 += mm3 + (1, 1, 1, 1)
					psrlw mm0, 1				;// mm0 >>= 1
					psrlw mm2, 1				;// mm2 >>= 1
					packuswb mm0, mm2			;// pack mm0 and mm2
					movq [edi], mm0				;// store mm0 into edi[0..7]
					add esi, edx				;// esi += edx
					add edi, edx				;// edi += edx
					dec ecx						;// decrement ecx
					jnz pred_comp__l11			;// loop while not zero
					emms						;// empty MMX state
				}
				return;
			}

			if(w == 16)
			{
				_asm
				{
					mov esi, s					;// esi = s
					mov edi, d					;// edi = d
					mov edx, lx					;// edx = lx
					mov ecx, h					;// ecx = h
pred_comp__l12:
					movd mm0, [esi]				;// lower 4 bytes into mm0 = esi[0..3]
					movd mm1, [esi+1]			;// lower 4 bytes into mm1 = esi[1..4]
					movd mm2, [esi+4]			;// lower 4 bytes into mm2 = esi[4..7]
					movd mm3, [esi+5]			;// lower 4 bytes into mm3 = esi[0..3]
					movd mm4, [esi+8]			;// lower 4 bytes into mm4 = esi[8..11]
					movd mm5, [esi+9]			;// lower 4 bytes into mm5 = esi[9..12]
					movd mm6, [esi+12]			;// lower 4 bytes into mm6 = esi[12..15]
					movd mm7, [esi+13]			;// lower 4 bytes into mm7 = esi[13..16]
					punpcklbw mm0, PACKED_0		;// unpack the lower 4 bytes into mm0
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm2, PACKED_0		;// unpack the lower 4 bytes into mm2
					punpcklbw mm3, PACKED_0		;// unpack the lower 4 bytes into mm3
					punpcklbw mm4, PACKED_0		;// unpack the lower 4 bytes into mm4
					punpcklbw mm5, PACKED_0		;// unpack the lower 4 bytes into mm5
					punpcklbw mm6, PACKED_0		;// unpack the lower 4 bytes into mm6
					punpcklbw mm7, PACKED_0		;// unpack the lower 4 bytes into mm7
					paddw mm0, mm1
					paddw mm2, mm3
					paddw mm4, mm5
					paddw mm6, mm7
					paddw mm0, PACKED_1			;// mm0 += mm1 + (1, 1, 1, 1)
					paddw mm2, PACKED_1			;// mm2 += mm3 + (1, 1, 1, 1)
					paddw mm4, PACKED_1			;// mm4 += mm5 + (1, 1, 1, 1)
					paddw mm6, PACKED_1			;// mm6 += mm7 + (1, 1, 1, 1)
					psrlw mm0, 1				;// mm0 >>= 1
					psrlw mm2, 1				;// mm2 >>= 1
					psrlw mm4, 1				;// mm4 >>= 1
					psrlw mm6, 1				;// mm6 >>= 1
					packuswb mm0, mm2			;// pack mm0 and mm2
					packuswb mm4, mm6			;// pack mm4 and mm6
					movq [edi], mm0				;// store mm0 into edi[0..7]
					movq [edi+8], mm4			;// store mm4 into edi[8..15]
					add esi, edx				;// esi += edx
					add edi, edx				;// edi += edx
					dec ecx						;// decrement ecx
					jnz pred_comp__l12			;// loop while not zero
					emms						;// empty MMX state
				}
				return;
			}
		}

		for (j=0; j<h; j++)
		{
			for (i=0; i<w; i++)
				d[i] = (unsigned int)(s[i]+s[i+1]+1)>>1;

			s+= lx;
			d+= lx;
		}
	}
	else /* if (xh && yh) */
	if (addflag)
	{
		if(cpu_MMX)
		{
			if(w == 8)
			{
				_asm
				{
					mov esi, s					;// esi = s
					mov edi, d					;// edi = d
					mov edx, lx					;// edx = lx
					mov ecx, h					;// ecx = h
pred_comp__l13:
					movd mm0, [esi]				;// lower 4 bytes into mm0 = esi[0..3]
					movd mm1, [esi+1]			;// lower 4 bytes into mm1 = esi[1..4]
					movd mm2, [esi+edx]			;// lower 4 bytes into mm2 = (esi + edx)[0..3]
					movd mm3, [esi+edx+1]		;// lower 4 bytes into mm3 = (esi + edx)[1..4]
					movd mm4, [esi+4]			;// lower 4 bytes into mm4 = esi[4..7]
					movd mm5, [esi+5]			;// lower 4 bytes into mm5 = esi[5..8]
					movd mm6, [esi+edx+4]		;// lower 4 bytes into mm6 = (esi + edx)[4..7]
					movd mm7, [esi+edx+5]		;// lower 4 bytes into mm7 = (esi + edx)[5..8]
					punpcklbw mm0, PACKED_0		;// unpack the lower 4 bytes into mm0
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm2, PACKED_0		;// unpack the lower 4 bytes into mm2
					punpcklbw mm3, PACKED_0		;// unpack the lower 4 bytes into mm3
					punpcklbw mm4, PACKED_0		;// unpack the lower 4 bytes into mm4
					punpcklbw mm5, PACKED_0		;// unpack the lower 4 bytes into mm5
					punpcklbw mm6, PACKED_0		;// unpack the lower 4 bytes into mm6
					punpcklbw mm7, PACKED_0		;// unpack the lower 4 bytes into mm7
					paddw mm0, mm1
					paddw mm2, mm3
					paddw mm4, mm5
					paddw mm6, mm7
					paddw mm0, mm2				;// mm0 += mm1 + mm2 + mm3
					paddw mm4, mm6				;// mm4 += mm5 + mm6 + mm7
					movd mm1, [edi]				;// lower 4 bytes into mm1 = edi[0..3]
					movd mm5, [edi+4]			;// lower 4 bytes into mm5 = edi[4..7]
					paddw mm0, PACKED_2			;// mm0 += (2, 2, 2, 2)
					paddw mm4, PACKED_2			;// mm4 += (2, 2, 2, 2)
					punpcklbw mm1, PACKED_0		;// unpack the lower 4 bytes into mm1
					punpcklbw mm5, PACKED_0		;// unpack the lower 4 bytes into mm5
					psrlw mm0, 2
					psrlw mm4, 2
					paddw mm0, PACKED_1
					paddw mm4, PACKED_1
					paddw mm0, mm1
					paddw mm4, mm5
					psrlw mm0, 1				;// mm0 = (mm0 >> 2) + (1, 1, 1, 1) + mm1
					psrlw mm4, 1				;// mm4 = (mm4 >> 2) + (1, 1, 1, 1) + mm5
					packuswb mm0, mm4			;// pack mm0 and mm4
					movq [edi], mm0				;// store mm0 into edi[0..7]

					add esi, edx				;// esi += edx
					add edi, edx				;// edi += edx
					dec ecx						;// decrement ecx
					jnz pred_comp__l13			;// loop while not zero
					emms						;// empty MMX state