MOVBU	((index*8)+1)(X29), X6; \
	MOVBU	((index*8)+2)(X29), X7; \
	MOVBU	((index*8)+3)(X29), X8; \
	SLL	$56, X5; \
	SLL	$48, X6; \
	OR	X5, X6, X5; \
	SLL	$40, X7; \
	OR	X5, X7, X5; \
	SLL	$32, X8; \
	OR	X5, X8, X5; \
	MOVBU	((index*8)+4)(X29), X9; \
	MOVBU	((index*8)+5)(X29), X6; \
	MOVBU	((index*8)+6)(X29), X7; \
	MOVBU	((index*8)+7)(X29), X8; \
	SLL	$24, X9; \
	OR	X5, X9, X5; \
	SLL	$16, X6; \
	OR	X5, X6, X5; \

		x3 := byteorder.LeUint32(q[4*0x3:])
		x4 := byteorder.LeUint32(q[4*0x4:])
		x5 := byteorder.LeUint32(q[4*0x5:])
		x6 := byteorder.LeUint32(q[4*0x6:])
		x7 := byteorder.LeUint32(q[4*0x7:])
		x8 := byteorder.LeUint32(q[4*0x8:])
		x9 := byteorder.LeUint32(q[4*0x9:])
		xa := byteorder.LeUint32(q[4*0xa:])
		xb := byteorder.LeUint32(q[4*0xb:])
		xc := byteorder.LeUint32(q[4*0xc:])
		xd := byteorder.LeUint32(q[4*0xd:])

	// scramble 3 times
	AESENC	X8, X8
	AESENC	X9, X9
	AESENC	X10, X10
	AESENC	X11, X11
	AESENC	X12, X12
	AESENC	X13, X13
	AESENC	X14, X14
	AESENC	X15, X15

	AESENC	X8, X8
	AESENC	X9, X9
	AESENC	X10, X10
	AESENC	X11, X11
	AESENC	X12, X12
	AESENC	X13, X13
	AESENC	X14, X14
	AESENC	X15, X15

	AESENC	X8, X8
	AESENC	X9, X9
	AESENC	X10, X10

	MINU	X30, X5					// b3d2e20b
	ORN	X6, X7, X8				// 33e46340
	ORN	X6, X7					// b3e36340
	SEXTB	X16, X17				// 93184860
	SEXTH	X17, X18				// 13995860
	XNOR	X18, X19, X20				// 33ca2941
	XNOR	X18, X19				// b3c92941
	ZEXTH	X19, X20				// 3bca0908

	// 1.3: Bitwise Rotation (Zbb)
	ROL	X8, X9, X10				// 33958460 or b30f8040b3dff4013395840033e5af00
	ROL	X8, X9					// b3948460 or b30f8040b3dff401b3948400b3e49f00

	MOVOU yout(16*0), X2
	MOVOU yout(16*1), X3
	MOVOU zout(16*0), X4
	MOVOU zout(16*1), X5

	MOVL BX, X6
	MOVL CX, X7

	PXOR X8, X8
	PCMPEQL X9, X9

	PSHUFD $0, X6, X6
	PSHUFD $0, X7, X7

	PCMPEQL X8, X6
	PCMPEQL X8, X7

	MOVOU X6, X15
	PANDN X9, X15

	MOVOU x1in(16*0), X9
	MOVOU x1in(16*1), X10
	MOVOU y1in(16*0), X11
	MOVOU y1in(16*1), X12

		reg: regInfo{
			inputs: []inputInfo{
				{0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14
				{1, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14
				{2, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14
			},
			outputs: []outputInfo{
				{0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14
			},
		},
	},
	{
		name:         "MINSD",

	// seq 1 206 | sed 's/.*/	z = x&/'
	var x1 [10<<20]byte
	var x2 [10<<20]byte
	var x3 [10<<20]byte
	var x4 [10<<20]byte
	var x5 [10<<20]byte
	var x6 [10<<20]byte
	var x7 [10<<20]byte
	var x8 [10<<20]byte
	var x9 [10<<20]byte
	var x10 [10<<20]byte
	var x11 [10<<20]byte
	var x12 [10<<20]byte
	var x13 [10<<20]byte
	var x14 [10<<20]byte
	var x15 [10<<20]byte
	var x16 [10<<20]byte
	var x17 [10<<20]byte

	p256OrdSqr(x, _10101, 1)     // _101010
	p256OrdMul(_101111, _101, x) // _101111
	p256OrdMul(x, _10101, x)     // _111111 = x6
	p256OrdSqr(t, x, 2)          // _11111100
	p256OrdMul(t, t, _11)        // _11111111 = x8
	p256OrdSqr(x, t, 8)          // _ff00
	p256OrdMul(x, x, t)          // _ffff = x16
	p256OrdSqr(t, x, 16)         // _ffff0000
	p256OrdMul(t, t, x)          // _ffffffff = x32

	p256OrdSqr(x, t, 64)
	p256OrdMul(x, x, t)

    return retVal;
  }

  private enum SmallEnum {
    X0,
    X1,
    X2
  }

  private enum MediumEnum {
    X0,
    X1,
    X2,
    X3,
    X4,
    X5,
    X6,
    X7,
    X8,
    X9,
    X10,
    X11,
    X12,
    X13,
    X14,
    X15,
    X16,
    X17,
    X18,
    X19,
    X20,
    X21,
    X22,
    X23,
    X24,
    X25,

    return retVal;
  }

  private enum SmallEnum {
    X0,
    X1,
    X2
  }

  private enum MediumEnum {
    X0,
    X1,
    X2,
    X3,
    X4,
    X5,
    X6,
    X7,
    X8,
    X9,
    X10,
    X11,
    X12,
    X13,
    X14,
    X15,
    X16,
    X17,
    X18,
    X19,
    X20,
    X21,
    X22,
    X23,
    X24,
    X25,