s0 = uint32(sbox1[t0>>24])<<24 | uint32(sbox1[t3>>16&0xff])<<16 | uint32(sbox1[t2>>8&0xff])<<8 | uint32(sbox1[t1&0xff])
	s1 = uint32(sbox1[t1>>24])<<24 | uint32(sbox1[t0>>16&0xff])<<16 | uint32(sbox1[t3>>8&0xff])<<8 | uint32(sbox1[t2&0xff])
	s2 = uint32(sbox1[t2>>24])<<24 | uint32(sbox1[t1>>16&0xff])<<16 | uint32(sbox1[t0>>8&0xff])<<8 | uint32(sbox1[t3&0xff])

	for i := 0; i < 256; i++ {
		if j := sbox0[sbox1[i]]; j != byte(i) {
			t.Errorf("sbox0[sbox1[%#x]] = %#x", i, j)
		}
		if j := sbox1[sbox0[i]]; j != byte(i) {
			t.Errorf("sbox1[sbox0[%#x]] = %#x", i, j)
		}
	}
}

// Test that encryption tables are correct.
// (Can adapt this code to generate them too.)
func TestTe(t *testing.T) {
	for i := 0; i < 256; i++ {
		s := uint32(sbox0[i])
		s2 := mul(s, 2)
		s3 := mul(s, 3)

	STPW.P	(R4, R5), 8(R10)
	STPW.P	(R6, R7), 8(R10)
	MOVW	$0x1b, R14
ks128Loop:
		VMOV	R7, V2.S[0]
		WORD	$0x4E030042       // TBL V3.B16, [V2.B16], V2.B16
		AESE	V0.B16, V2.B16    // Use AES to compute the SBOX
		EORW	R13, R4
		LSLW	$1, R13           // Compute next Rcon
		ANDSW	$0x100, R13, ZR
		CSELW	NE, R14, R13, R13 // Fake modulo
		SUBS	$1, R8
		VMOV	V2.S[0], R0
		EORW	R0, R4
		EORW	R4, R5
		EORW	R5, R6