ADDSW	R19.UXTW, R14, R17              // d141332b
	ADDS	R12.SXTX, R3, R1                // 61e02cab
	SUB	R19.UXTH<<4, R2, R21            // 553033cb
	SUBW	R1.UXTX<<1, R3, R2              // 6264214b
	SUBS	R3.UXTX, R8, R9                 // 096123eb
	SUBSW	R17.UXTH, R15, R21              // f521316b
	SUBW	ZR<<14, R19, R13                // 6d3a1f4b
	CMP	R2.SXTH, R13                    // bfa122eb

		// starting with the largest sizes and generating as
		// many as needed, using the appropriate offset value.
		//	MOVD  n(R21),R31
		//	MOVD  R31,n(R20)
		//	MOVW  n1(R21),R31
		//	MOVW  R31,n1(R20)
		//	MOVH  n2(R21),R31
		//	MOVH  R31,n2(R20)
		//	MOVB  n3(R21),R31
		//	MOVB  R31,n3(R20)

		// Each loop iteration moves 32 bytes
		ctr := v.AuxInt / bytesPerLoop

		// Remainder after the loop

			clobberFlags:   true,
			faultOnNilArg0: true,
		},

		// duffcopy
		// arg0 = address of dst memory (in R21, changed as side effect)
		// arg1 = address of src memory (in R20, changed as side effect)
		// arg2 = mem
		// auxint = offset into duffcopy code to start executing
		// returns mem
		// R20, R21 changed as side effect
		// R16 and R17 may be clobbered by linker trampoline.
		{
			name:      "DUFFCOPY",

TEXT ·p256MovCond(SB), NOSPLIT, $0-32
	MOVD res+0(FP), P3ptr
	MOVD a+8(FP), P1ptr
	MOVD b+16(FP), P2ptr
	MOVD $16, R16
	MOVD $32, R17
	MOVD $48, R18
	MOVD $56, R21
	MOVD $64, R19
	MOVD $80, R20
	// cond is R1 + 24 (cond offset) + 32
	LXVDSX (R1)(R21), SEL
	VSPLTISB $0, ZER
	// SEL controls whether to store a or b
	VCMPEQUD SEL, ZER, SEL

	LXVD2X (P1ptr+R0), X1H
	LXVD2X (P1ptr+R16), X1L