halfpel8_refine_ia64.s

这是一个压缩解压包,用C语言进行编程的,里面有详细的源代码.

	ld8 mpr[6] = [refab[1]], iEdgedWidth
	ld8 ref2a[6] = [refaa[2]], iEdgedWidth
	ld8 ref2b[6] = [refab[2]], iEdgedWidth
	;; 
	ld8 cur[7] = [cura]
	ld8 ref0a[7] = [refaa[0]], iEdgedWidth
	ld8 ref0b[7] = [refab[0]], iEdgedWidth
	ld8 ref1a[7] = [refaa[1]], iEdgedWidth
	ld8 mpr[7] = [refab[1]], iEdgedWidth
	ld8 ref2a[7] = [refaa[2]]
	ld8 ref2b[7] = [refab[2]]
	;; 
	ld8 ref0a[8] = [refaa[0]]
	ld8 ref0b[8] = [refab[0]]
	ld8 ref1a[8] = [refaa[1]]
	ld8 mpr[8] = [refab[1]]
	;;


	// Align ref1
	
     	shr.u ref1a[0] = ref1a[0], nob1
     	shr.u ref1a[1] = ref1a[1], nob1
     	shr.u ref1a[2] = ref1a[2], nob1
     	shr.u ref1a[3] = ref1a[3], nob1
     	shr.u ref1a[4] = ref1a[4], nob1
     	shr.u ref1a[5] = ref1a[5], nob1
     	shr.u ref1a[6] = ref1a[6], nob1
     	shr.u ref1a[7] = ref1a[7], nob1
     	shr.u ref1a[8] = ref1a[8], nob1

	shl mpr[0] = mpr[0], nob64m1
	shl mpr[1] = mpr[1], nob64m1
	shl mpr[2] = mpr[2], nob64m1
	shl mpr[3] = mpr[3], nob64m1
	shl mpr[4] = mpr[4], nob64m1
	shl mpr[5] = mpr[5], nob64m1
	shl mpr[6] = mpr[6], nob64m1
	shl mpr[7] = mpr[7], nob64m1
	shl mpr[8] = mpr[8], nob64m1
	;; 
.explicit
{.mii
	or ref1a[0] = ref1a[0], mpr[0]
     	shr.u ref0a[0] = ref0a[0], nob02
     	shr.u ref0a[1] = ref0a[1], nob02
}
{.mmi
	or ref1a[1] = ref1a[1], mpr[1]
	or ref1a[2] = ref1a[2], mpr[2]
     	shr.u ref0a[2] = ref0a[2], nob02
}
{.mii
	or ref1a[3] = ref1a[3], mpr[3]
     	shr.u ref0a[3] = ref0a[3], nob02
     	shr.u ref0a[4] = ref0a[4], nob02
}
{.mmi
	or ref1a[4] = ref1a[4], mpr[4]
	or ref1a[5] = ref1a[5], mpr[5]
	shr.u ref0a[5] = ref0a[5], nob02
}
{.mii
	or ref1a[6] = ref1a[6], mpr[6]
     	shr.u ref0a[6] = ref0a[6], nob02
     	shr.u ref0a[7] = ref0a[7], nob02
}
{.mii
	or ref1a[7] = ref1a[7], mpr[7]
	or ref1a[8] = ref1a[8], mpr[8]
     	shr.u ref0a[8] = ref0a[8], nob02
}
.default
	// ref1a[] now contains center position values
	// mpr[] not used any more
	
	// Align ref0 left
	
	;; 
	shl mpr[0] = ref0b[0], nob56m02
	shl mpr[1] = ref0b[1], nob56m02
	shl mpr[2] = ref0b[2], nob56m02
	shl mpr[3] = ref0b[3], nob56m02
	shl mpr[4] = ref0b[4], nob56m02
	shl mpr[5] = ref0b[5], nob56m02
	shl mpr[6] = ref0b[6], nob56m02
	shl mpr[7] = ref0b[7], nob56m02
	shl mpr[8] = ref0b[8], nob56m02

	shl ref0b[0] = ref0b[0], nob64m02
	shl ref0b[1] = ref0b[1], nob64m02
	shl ref0b[2] = ref0b[2], nob64m02
	shl ref0b[3] = ref0b[3], nob64m02
	shl ref0b[4] = ref0b[4], nob64m02
	shl ref0b[5] = ref0b[5], nob64m02
	shl ref0b[6] = ref0b[6], nob64m02
	shl ref0b[7] = ref0b[7], nob64m02
	shl ref0b[8] = ref0b[8], nob64m02
	;; 
	or ref0a[0] = ref0a[0], ref0b[0]
	or ref0a[1] = ref0a[1], ref0b[1]
	or ref0a[2] = ref0a[2], ref0b[2]
	or ref0a[3] = ref0a[3], ref0b[3]
	or ref0a[4] = ref0a[4], ref0b[4]
	or ref0a[5] = ref0a[5], ref0b[5]
	or ref0a[6] = ref0a[6], ref0b[6]
	or ref0a[7] = ref0a[7], ref0b[7]
	or ref0a[8] = ref0a[8], ref0b[8]
	;;

	// ref0a[] now contains left position values
	// mpr[] contains intermediate result for right position values (former ref0a << 56 - nob02)
	
	// Align ref0 right

	// Shift one byte more to the right (seen als big-endian)
	shr.u ref0b[0] = ref0a[0], 8
	shr.u ref0b[1] = ref0a[1], 8
	shr.u ref0b[2] = ref0a[2], 8
	shr.u ref0b[3] = ref0a[3], 8
	shr.u ref0b[4] = ref0a[4], 8
	shr.u ref0b[5] = ref0a[5], 8
	shr.u ref0b[6] = ref0a[6], 8
	shr.u ref0b[7] = ref0a[7], 8
	shr.u ref0b[8] = ref0a[8], 8
	;;
.explicit
{.mii
	or  ref0b[0] = ref0b[0], mpr[0]
     	shr.u ref2a[0] = ref2a[0], nob02
     	shr.u ref2a[1] = ref2a[1], nob02
}
{.mmi
	or  ref0b[1] = ref0b[1], mpr[1]
	or  ref0b[2] = ref0b[2], mpr[2]
     	shr.u ref2a[2] = ref2a[2], nob02
}
{.mii
	or  ref0b[3] = ref0b[3], mpr[3]
     	shr.u ref2a[3] = ref2a[3], nob02
     	shr.u ref2a[4] = ref2a[4], nob02
}
{.mmi
	or  ref0b[4] = ref0b[4], mpr[4]
	or  ref0b[5] = ref0b[5], mpr[5]
     	shr.u ref2a[5] = ref2a[5], nob02
}
{.mii
	or  ref0b[6] = ref0b[6], mpr[6]
     	shr.u ref2a[6] = ref2a[6], nob02
     	shr.u ref2a[7] = ref2a[7], nob02
}
.default
	or  ref0b[7] = ref0b[7], mpr[7]
	or  ref0b[8] = ref0b[8], mpr[8]
	
	// ref0b[] now contains right position values
	// mpr[] not needed any more

	
	// Align ref2 left
	
	;; 
	shl mpr[0] = ref2b[0], nob56m02
	shl mpr[1] = ref2b[1], nob56m02
	shl mpr[2] = ref2b[2], nob56m02
	shl mpr[3] = ref2b[3], nob56m02
	shl mpr[4] = ref2b[4], nob56m02
	shl mpr[5] = ref2b[5], nob56m02
	shl mpr[6] = ref2b[6], nob56m02
	shl mpr[7] = ref2b[7], nob56m02

	shl ref2b[0] = ref2b[0], nob64m02
	shl ref2b[1] = ref2b[1], nob64m02
	shl ref2b[2] = ref2b[2], nob64m02
	shl ref2b[3] = ref2b[3], nob64m02
	shl ref2b[4] = ref2b[4], nob64m02
	shl ref2b[5] = ref2b[5], nob64m02
	shl ref2b[6] = ref2b[6], nob64m02
	shl ref2b[7] = ref2b[7], nob64m02
	;; 
	or ref2a[0] = ref2a[0], ref2b[0]
	or ref2a[1] = ref2a[1], ref2b[1]
	or ref2a[2] = ref2a[2], ref2b[2]
	or ref2a[3] = ref2a[3], ref2b[3]
	or ref2a[4] = ref2a[4], ref2b[4]
	or ref2a[5] = ref2a[5], ref2b[5]
	or ref2a[6] = ref2a[6], ref2b[6]
	or ref2a[7] = ref2a[7], ref2b[7]
	;;

	// ref2a[] now contains left position values
	// mpr[] contains intermediate result for right position values (former ref2a << 56 - nob02)
	
	// Align ref2 right

	// Shift one byte more to the right (seen als big-endian)
	shr.u ref2b[0] = ref2a[0], 8
	shr.u ref2b[1] = ref2a[1], 8
	shr.u ref2b[2] = ref2a[2], 8
	shr.u ref2b[3] = ref2a[3], 8
	shr.u ref2b[4] = ref2a[4], 8
	shr.u ref2b[5] = ref2a[5], 8
	shr.u ref2b[6] = ref2a[6], 8
	shr.u ref2b[7] = ref2a[7], 8
	;; 
	or  ref2b[0] = ref2b[0], mpr[0]
	or  ref2b[1] = ref2b[1], mpr[1]
	or  ref2b[2] = ref2b[2], mpr[2]
	or  ref2b[3] = ref2b[3], mpr[3]
	or  ref2b[4] = ref2b[4], mpr[4]
	or  ref2b[5] = ref2b[5], mpr[5]
	or  ref2b[6] = ref2b[6], mpr[6]
	or  ref2b[7] = ref2b[7], mpr[7]
	

	// ref2b[] now contains right position values
	// mpr[] not needed any more


		
	// Let's SAD

	// Left top corner
	

	sub dx = backupX, dx
	psad1 mpr[0] = cur[0], ref0a[0]
	psad1 mpr[1] = cur[1], ref0a[1]

	sub dy = backupY, dy
	psad1 mpr[2] = cur[2], ref0a[2]
	psad1 mpr[3] = cur[3], ref0a[3]
	psad1 mpr[4] = cur[4], ref0a[4]
	psad1 mpr[5] = cur[5], ref0a[5]
	psad1 mpr[6] = cur[6], ref0a[6]
	psad1 mpr[7] = cur[7], ref0a[7]
	;; 
.include "../../src/motion/ia64_asm/calc_delta_1.s"
	
	// Top edge

	psad1 mpr[0] = cur[0], ref1a[0]
	psad1 mpr[1] = cur[1], ref1a[1]
	psad1 mpr[2] = cur[2], ref1a[2]
	psad1 mpr[3] = cur[3], ref1a[3]
	psad1 mpr[4] = cur[4], ref1a[4]

	add dx = 1, dx
	psad1 mpr[5] = cur[5], ref1a[5]
	psad1 mpr[6] = cur[6], ref1a[6]

	psad1 mpr[7] = cur[7], ref1a[7]
	;;

.include "../../src/motion/ia64_asm/calc_delta_2.s"
(lt)	cmp.lt.unc fb, p0 = mpr[8], iMinSAD
.include "../../src/motion/ia64_asm/calc_delta_3.s"
	
	// Right top corner


	psad1 mpr[0] = cur[0], ref0b[0]
	psad1 mpr[1] = cur[1], ref0b[1]
	psad1 mpr[2] = cur[2], ref0b[2]
	psad1 mpr[3] = cur[3], ref0b[3]
	psad1 mpr[4] = cur[4], ref0b[4]
	
	add backupX = 1, backupX
	psad1 mpr[5] = cur[5], ref0b[5]
	psad1 mpr[6] = cur[6], ref0b[6]

	add dx = 1, dx
	psad1 mpr[7] = cur[7], ref0b[7]
	;;
	
.include "../../src/motion/ia64_asm/calc_delta_1.s"
(t)	cmp.lt.unc fb, p0 = iSAD, iMinSAD
	;; 
	
	// Left edge

(fb)	mov iMinSAD = iSAD
	psad1 mpr[0] = cur[0], ref2a[0]

(fb)	mov currX = backupX
	psad1 mpr[1] = cur[1], ref2a[1]
	psad1 mpr[2] = cur[2], ref2a[2]

(fb)	mov currY = backupY
	psad1 mpr[3] = cur[3], ref2a[3]
	psad1 mpr[4] = cur[4], ref2a[4]

	add backupX = 1, backupX
	psad1 mpr[5] = cur[5], ref2a[5]
	psad1 mpr[6] = cur[6], ref2a[6]

	psad1 mpr[7] = cur[7], ref2a[7]

	add dx = -2, dx
	add dy = 1, dy
	;;
	
.include "../../src/motion/ia64_asm/calc_delta_2.s"
(rt)	cmp.lt.unc fb, p0 = mpr[8], iMinSAD
.include "../../src/motion/ia64_asm/calc_delta_3.s"
	
	// Right edge

	
	psad1 mpr[0] = cur[0], ref2b[0]
	psad1 mpr[1] = cur[1], ref2b[1]
	psad1 mpr[2] = cur[2], ref2b[2]
	psad1 mpr[3] = cur[3], ref2b[3]
	psad1 mpr[4] = cur[4], ref2b[4]

	add backupX = -2, backupX
	psad1 mpr[5] = cur[5], ref2b[5]
	psad1 mpr[6] = cur[6], ref2b[6]

	add backupY = 1, backupY
	add dx = 2, dx
	psad1 mpr[7] = cur[7], ref2b[7]
	;;
	
.include "../../src/motion/ia64_asm/calc_delta_1.s"
(l)	cmp.lt.unc fb, p0 = iSAD, iMinSAD
	;;
	
	// Left bottom corner

(fb)	mov iMinSAD = iSAD
	psad1 mpr[0] = cur[0], ref0a[1]

(fb)	mov currX = backupX
	psad1 mpr[1] = cur[1], ref0a[2]
	psad1 mpr[2] = cur[2], ref0a[3]

(fb)	mov currY = backupY
	psad1 mpr[3] = cur[3], ref0a[4]
	psad1 mpr[4] = cur[4], ref0a[5]

	add backupX = 2, backupX
	psad1 mpr[5] = cur[5], ref0a[6]
	psad1 mpr[6] = cur[6], ref0a[7]

	psad1 mpr[7] = cur[7], ref0a[8]

	add dx = -2, dx
	add dy = 1, dy
	;;
	
.include "../../src/motion/ia64_asm/calc_delta_2.s"
(r)	cmp.lt.unc fb, p0 = mpr[8], iMinSAD
.include "../../src/motion/ia64_asm/calc_delta_3.s"
	
	// Bottom edge
	
	psad1 mpr[0] = cur[0], ref1a[1]
	psad1 mpr[1] = cur[1], ref1a[2]
	psad1 mpr[2] = cur[2], ref1a[3]
	psad1 mpr[3] = cur[3], ref1a[4]
	psad1 mpr[4] = cur[4], ref1a[5]

	add backupX = -2, backupX
	psad1 mpr[5] = cur[5], ref1a[6]
	psad1 mpr[6] = cur[6], ref1a[7]

	add backupY = 1, backupY
	add dx = 1, dx
	psad1 mpr[7] = cur[7], ref1a[8]
	;;
	
.include "../../src/motion/ia64_asm/calc_delta_1.s"
(lb)	cmp.lt.unc fb, p0 = iSAD, iMinSAD
	;; 
	// Right bottom corner


(fb)	mov iMinSAD = iSAD
	psad1 mpr[0] = cur[0], ref0b[1]

(fb)	mov currX = backupX
	psad1 mpr[1] = cur[1], ref0b[2]
	psad1 mpr[2] = cur[2], ref0b[3]

(fb)	mov currY = backupY
	psad1 mpr[3] = cur[3], ref0b[4]
	psad1 mpr[4] = cur[4], ref0b[5]
	
	add backupX = 1, backupX
	psad1 mpr[5] = cur[5], ref0b[6]
	psad1 mpr[6] = cur[6], ref0b[7]

	add dx = 1, dx
	psad1 mpr[7] = cur[7], ref0b[8]
	;;
	
.include "../../src/motion/ia64_asm/calc_delta_2.s"
(b)	cmp.lt.unc fb, p0 = mpr[8], iMinSAD
.include "../../src/motion/ia64_asm/calc_delta_3.s"

(rb)	getf.sig ret0 = fmv
	add backupX = 1, backupX
	;; 
(rb)	add iSAD = iSAD, ret0
	;; 
(rb)	cmp.lt.unc fb, p0 = iSAD, iMinSAD
	;; 
(fb)	mov iMinSAD = iSAD
(fb)	mov currX = backupX
(fb)	mov currY = backupY
	;; 
	
	// Write back result
 
	st4 [currMV] = currX
	st4 [currYAddress] = currY
	mov ret0 = iMinSAD

	// Restore important registers

	;;
	mov pr = prsave, -1
	mov ar.pfs = pfs	
	br.ret.sptk.many b0
	
	.endp Halfpel8_Refine_ia64#