ROUND2(R3,R4,R5,R2, 5,0xe9b6c7aa,20);
	ROUND2(R2,R3,R4,R5,10,0xd62f105d, 5);
	ROUND2(R5,R2,R3,R4,15, 0x2441453, 9);
	ROUND2(R4,R5,R2,R3, 4,0xd8a1e681,14);
	ROUND2(R3,R4,R5,R2, 9,0xe7d3fbc8,20);
	ROUND2(R2,R3,R4,R5,14,0x21e1cde6, 5);
	ROUND2(R5,R2,R3,R4, 3,0xc33707d6, 9);
	ROUND2(R4,R5,R2,R3, 8,0xf4d50d87,14);
	ROUND2(R3,R4,R5,R2,13,0x455a14ed,20);
	ROUND2(R2,R3,R4,R5, 2,0xa9e3e905, 5);
	ROUND2(R5,R2,R3,R4, 7,0xfcefa3f8, 9);

	ROUND2(AX,BX,CX,DX,10,0xd62f105d, 5);
	ROUND2(DX,AX,BX,CX,15, 0x2441453, 9);
	ROUND2(CX,DX,AX,BX, 4,0xd8a1e681,14);
	ROUND2(BX,CX,DX,AX, 9,0xe7d3fbc8,20);
	ROUND2(AX,BX,CX,DX,14,0x21e1cde6, 5);
	ROUND2(DX,AX,BX,CX, 3,0xc33707d6, 9);
	ROUND2(CX,DX,AX,BX, 8,0xf4d50d87,14);
	ROUND2(BX,CX,DX,AX,13,0x455a14ed,20);
	ROUND2(AX,BX,CX,DX, 2,0xa9e3e905, 5);
	ROUND2(DX,AX,BX,CX, 7,0xfcefa3f8, 9);

	ROUND1x(BX, CX, DX, BP, AX, 19)

	ROUND2(AX, BX, CX, DX, BP, 20)
	ROUND2(BP, AX, BX, CX, DX, 21)
	ROUND2(DX, BP, AX, BX, CX, 22)
	ROUND2(CX, DX, BP, AX, BX, 23)
	ROUND2(BX, CX, DX, BP, AX, 24)
	ROUND2(AX, BX, CX, DX, BP, 25)
	ROUND2(BP, AX, BX, CX, DX, 26)
	ROUND2(DX, BP, AX, BX, CX, 27)
	ROUND2(CX, DX, BP, AX, BX, 28)
	ROUND2(BX, CX, DX, BP, AX, 29)
	ROUND2(AX, BX, CX, DX, BP, 30)
	ROUND2(BP, AX, BX, CX, DX, 31)

	ROUND1x(Rd, Re, Ra, Rb, Rc)
	ROUND1x(Rc, Rd, Re, Ra, Rb)
	ROUND1x(Rb, Rc, Rd, Re, Ra)

	MOVW	$0x6ED9EBA1, Rconst
	MOVW	$4, Rctr
loop2:	ROUND2(Ra, Rb, Rc, Rd, Re)
	ROUND2(Re, Ra, Rb, Rc, Rd)
	ROUND2(Rd, Re, Ra, Rb, Rc)
	ROUND2(Rc, Rd, Re, Ra, Rb)
	ROUND2(Rb, Rc, Rd, Re, Ra)
	SUB.S	$1, Rctr
	BNE	loop2

	MOVW	$0x8F1BBCDC, Rconst
	MOVW	$4, Rctr
loop3:	ROUND3(Ra, Rb, Rc, Rd, Re)

	ROUND2(R4,R5,R6,R7,10,0xd62f105d, 5);
	ROUND2(R7,R4,R5,R6,15, 0x2441453, 9);
	ROUND2(R6,R7,R4,R5, 4,0xd8a1e681,14);
	ROUND2(R5,R6,R7,R4, 9,0xe7d3fbc8,20);
	ROUND2(R4,R5,R6,R7,14,0x21e1cde6, 5);
	ROUND2(R7,R4,R5,R6, 3,0xc33707d6, 9);
	ROUND2(R6,R7,R4,R5, 8,0xf4d50d87,14);
	ROUND2(R5,R6,R7,R4,13,0x455a14ed,20);
	ROUND2(R4,R5,R6,R7, 2,0xa9e3e905, 5);
	ROUND2(R7,R4,R5,R6, 7,0xfcefa3f8, 9);
	ROUND2(R6,R7,R4,R5,12,0x676f02d9,14);

	MOVL	(1*4)(SI),	DI
	MOVL	DX,		BP

	ROUND2(AX,BX,CX,DX, 6,0xf61e2562, 5);
	ROUND2(DX,AX,BX,CX,11,0xc040b340, 9);
	ROUND2(CX,DX,AX,BX, 0,0x265e5a51,14);
	ROUND2(BX,CX,DX,AX, 5,0xe9b6c7aa,20);
	ROUND2(AX,BX,CX,DX,10,0xd62f105d, 5);
	ROUND2(DX,AX,BX,CX,15, 0x2441453, 9);
	ROUND2(CX,DX,AX,BX, 4,0xd8a1e681,14);
	ROUND2(BX,CX,DX,AX, 9,0xe7d3fbc8,20);
	ROUND2(AX,BX,CX,DX,14,0x21e1cde6, 5);

	ROUND1(R3,R4,R5,R22,M15,0x49b40821,22);

	ROUND2(R22,R3,R4,R5,M01,0xf61e2562, 5);
	ROUND2(R5,R22,R3,R4,M06,0xc040b340, 9);
	ROUND2(R4,R5,R22,R3,M11,0x265e5a51,14);
	ROUND2(R3,R4,R5,R22,M00,0xe9b6c7aa,20);
	ROUND2(R22,R3,R4,R5,M05,0xd62f105d, 5);
	ROUND2(R5,R22,R3,R4,M10, 0x2441453, 9);
	ROUND2(R4,R5,R22,R3,M15,0xd8a1e681,14);
	ROUND2(R3,R4,R5,R22,M04,0xe7d3fbc8,20);
	ROUND2(R22,R3,R4,R5,M09,0x21e1cde6, 5);

	ROUND2(Rb, Rc, Rd, Ra,  0, 20, Rc3)

	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
	ROUND2(Ra, Rb, Rc, Rd,  5,	5, Rc0)
	ROUND2(Rd, Ra, Rb, Rc, 10,	9, Rc1)
	ROUND2(Rc, Rd, Ra, Rb, 15, 14, Rc2)
	ROUND2(Rb, Rc, Rd, Ra,  4, 20, Rc3)

	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
	ROUND2(Ra, Rb, Rc, Rd,  9,	5, Rc0)
	ROUND2(Rd, Ra, Rb, Rc, 14,	9, Rc1)
	ROUND2(Rc, Rd, Ra, Rb,  3, 14, Rc2)
	ROUND2(Rb, Rc, Rd, Ra,  8, 20, Rc3)

	ROUND1x(BX, CX, DX, BP, AX, 19)

	ROUND2(AX, BX, CX, DX, BP, 20)
	ROUND2(BP, AX, BX, CX, DX, 21)
	ROUND2(DX, BP, AX, BX, CX, 22)
	ROUND2(CX, DX, BP, AX, BX, 23)
	ROUND2(BX, CX, DX, BP, AX, 24)
	ROUND2(AX, BX, CX, DX, BP, 25)
	ROUND2(BP, AX, BX, CX, DX, 26)
	ROUND2(DX, BP, AX, BX, CX, 27)
	ROUND2(CX, DX, BP, AX, BX, 28)
	ROUND2(BX, CX, DX, BP, AX, 29)
	ROUND2(AX, BX, CX, DX, BP, 30)
	ROUND2(BP, AX, BX, CX, DX, 31)

// CHECK: %[[add2:.*]] = "tf.AddV2"(%[[rescale1]]
// CHECK: %[[maximum2:.*]] = "tf.Maximum"(%[[add2]]
// CHECK: %[[minimum2:.*]] = "tf.Minimum"(%[[maximum2]]
// CHECK: %[[round2:.*]] = "tf.Round"(%[[minimum2]]
// CHECK: %[[quant2:.*]] = "tf.Cast"(%[[round2]]) <{Truncate = false}> : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xi8>

// CHECK: %[[pad2:.*]] = "tf.PadV2"(%[[quant2]]
// CHECK: %[[xlaconv2:.*]] = "tf.XlaConvV2"(%[[pad2]]