rijndael-alg-fst.c

这个极其小巧（9k）的加密程序对文件进行加密及解密的速度与资源管理器拷贝文件的速度一样快。因为速度快

			Td1[Te4[(rk[2] >> 16) & 0xff] & 0xff] ^
			Td2[Te4[(rk[2] >>  8) & 0xff] & 0xff] ^
			Td3[Te4[(rk[2]      ) & 0xff] & 0xff];
		rk[3] =
			Td0[Te4[(rk[3] >> 24)       ] & 0xff] ^
			Td1[Te4[(rk[3] >> 16) & 0xff] & 0xff] ^
			Td2[Te4[(rk[3] >>  8) & 0xff] & 0xff] ^
			Td3[Te4[(rk[3]      ) & 0xff] & 0xff];
	}
	return Nr;
}

void rijndaelEncrypt(u32 rk[/*4*(Nr + 1)*/], int Nr, const u8 pt[16], u8 ct[16]) {
	u32 s0, s1, s2, s3, t0, t1, t2, t3;
#ifndef FULL_UNROLL
    int r;
#endif /* ?FULL_UNROLL */

    /*
	 * map byte array block to cipher state
	 * and add initial round key:
	 */
	s0 = GETU32(pt     ) ^ rk[0];
	s1 = GETU32(pt +  4) ^ rk[1];
	s2 = GETU32(pt +  8) ^ rk[2];
	s3 = GETU32(pt + 12) ^ rk[3];
#ifdef FULL_UNROLL
    /* round 1: */
   	t0 = Te0[s0 >> 24] ^ Te1[(s1 >> 16) & 0xff] ^ Te2[(s2 >>  8) & 0xff] ^ Te3[s3 & 0xff] ^ rk[ 4];
   	t1 = Te0[s1 >> 24] ^ Te1[(s2 >> 16) & 0xff] ^ Te2[(s3 >>  8) & 0xff] ^ Te3[s0 & 0xff] ^ rk[ 5];
   	t2 = Te0[s2 >> 24] ^ Te1[(s3 >> 16) & 0xff] ^ Te2[(s0 >>  8) & 0xff] ^ Te3[s1 & 0xff] ^ rk[ 6];
   	t3 = Te0[s3 >> 24] ^ Te1[(s0 >> 16) & 0xff] ^ Te2[(s1 >>  8) & 0xff] ^ Te3[s2 & 0xff] ^ rk[ 7];
   	/* round 2: */
   	s0 = Te0[t0 >> 24] ^ Te1[(t1 >> 16) & 0xff] ^ Te2[(t2 >>  8) & 0xff] ^ Te3[t3 & 0xff] ^ rk[ 8];
   	s1 = Te0[t1 >> 24] ^ Te1[(t2 >> 16) & 0xff] ^ Te2[(t3 >>  8) & 0xff] ^ Te3[t0 & 0xff] ^ rk[ 9];
   	s2 = Te0[t2 >> 24] ^ Te1[(t3 >> 16) & 0xff] ^ Te2[(t0 >>  8) & 0xff] ^ Te3[t1 & 0xff] ^ rk[10];
   	s3 = Te0[t3 >> 24] ^ Te1[(t0 >> 16) & 0xff] ^ Te2[(t1 >>  8) & 0xff] ^ Te3[t2 & 0xff] ^ rk[11];
    /* round 3: */
   	t0 = Te0[s0 >> 24] ^ Te1[(s1 >> 16) & 0xff] ^ Te2[(s2 >>  8) & 0xff] ^ Te3[s3 & 0xff] ^ rk[12];
   	t1 = Te0[s1 >> 24] ^ Te1[(s2 >> 16) & 0xff] ^ Te2[(s3 >>  8) & 0xff] ^ Te3[s0 & 0xff] ^ rk[13];
   	t2 = Te0[s2 >> 24] ^ Te1[(s3 >> 16) & 0xff] ^ Te2[(s0 >>  8) & 0xff] ^ Te3[s1 & 0xff] ^ rk[14];
   	t3 = Te0[s3 >> 24] ^ Te1[(s0 >> 16) & 0xff] ^ Te2[(s1 >>  8) & 0xff] ^ Te3[s2 & 0xff] ^ rk[15];
   	/* round 4: */
   	s0 = Te0[t0 >> 24] ^ Te1[(t1 >> 16) & 0xff] ^ Te2[(t2 >>  8) & 0xff] ^ Te3[t3 & 0xff] ^ rk[16];
   	s1 = Te0[t1 >> 24] ^ Te1[(t2 >> 16) & 0xff] ^ Te2[(t3 >>  8) & 0xff] ^ Te3[t0 & 0xff] ^ rk[17];
   	s2 = Te0[t2 >> 24] ^ Te1[(t3 >> 16) & 0xff] ^ Te2[(t0 >>  8) & 0xff] ^ Te3[t1 & 0xff] ^ rk[18];
   	s3 = Te0[t3 >> 24] ^ Te1[(t0 >> 16) & 0xff] ^ Te2[(t1 >>  8) & 0xff] ^ Te3[t2 & 0xff] ^ rk[19];
    /* round 5: */
   	t0 = Te0[s0 >> 24] ^ Te1[(s1 >> 16) & 0xff] ^ Te2[(s2 >>  8) & 0xff] ^ Te3[s3 & 0xff] ^ rk[20];
   	t1 = Te0[s1 >> 24] ^ Te1[(s2 >> 16) & 0xff] ^ Te2[(s3 >>  8) & 0xff] ^ Te3[s0 & 0xff] ^ rk[21];
   	t2 = Te0[s2 >> 24] ^ Te1[(s3 >> 16) & 0xff] ^ Te2[(s0 >>  8) & 0xff] ^ Te3[s1 & 0xff] ^ rk[22];
   	t3 = Te0[s3 >> 24] ^ Te1[(s0 >> 16) & 0xff] ^ Te2[(s1 >>  8) & 0xff] ^ Te3[s2 & 0xff] ^ rk[23];
   	/* round 6: */
   	s0 = Te0[t0 >> 24] ^ Te1[(t1 >> 16) & 0xff] ^ Te2[(t2 >>  8) & 0xff] ^ Te3[t3 & 0xff] ^ rk[24];
   	s1 = Te0[t1 >> 24] ^ Te1[(t2 >> 16) & 0xff] ^ Te2[(t3 >>  8) & 0xff] ^ Te3[t0 & 0xff] ^ rk[25];
   	s2 = Te0[t2 >> 24] ^ Te1[(t3 >> 16) & 0xff] ^ Te2[(t0 >>  8) & 0xff] ^ Te3[t1 & 0xff] ^ rk[26];
   	s3 = Te0[t3 >> 24] ^ Te1[(t0 >> 16) & 0xff] ^ Te2[(t1 >>  8) & 0xff] ^ Te3[t2 & 0xff] ^ rk[27];
    /* round 7: */
   	t0 = Te0[s0 >> 24] ^ Te1[(s1 >> 16) & 0xff] ^ Te2[(s2 >>  8) & 0xff] ^ Te3[s3 & 0xff] ^ rk[28];
   	t1 = Te0[s1 >> 24] ^ Te1[(s2 >> 16) & 0xff] ^ Te2[(s3 >>  8) & 0xff] ^ Te3[s0 & 0xff] ^ rk[29];
   	t2 = Te0[s2 >> 24] ^ Te1[(s3 >> 16) & 0xff] ^ Te2[(s0 >>  8) & 0xff] ^ Te3[s1 & 0xff] ^ rk[30];
   	t3 = Te0[s3 >> 24] ^ Te1[(s0 >> 16) & 0xff] ^ Te2[(s1 >>  8) & 0xff] ^ Te3[s2 & 0xff] ^ rk[31];
   	/* round 8: */
   	s0 = Te0[t0 >> 24] ^ Te1[(t1 >> 16) & 0xff] ^ Te2[(t2 >>  8) & 0xff] ^ Te3[t3 & 0xff] ^ rk[32];
   	s1 = Te0[t1 >> 24] ^ Te1[(t2 >> 16) & 0xff] ^ Te2[(t3 >>  8) & 0xff] ^ Te3[t0 & 0xff] ^ rk[33];
   	s2 = Te0[t2 >> 24] ^ Te1[(t3 >> 16) & 0xff] ^ Te2[(t0 >>  8) & 0xff] ^ Te3[t1 & 0xff] ^ rk[34];
   	s3 = Te0[t3 >> 24] ^ Te1[(t0 >> 16) & 0xff] ^ Te2[(t1 >>  8) & 0xff] ^ Te3[t2 & 0xff] ^ rk[35];
    /* round 9: */
   	t0 = Te0[s0 >> 24] ^ Te1[(s1 >> 16) & 0xff] ^ Te2[(s2 >>  8) & 0xff] ^ Te3[s3 & 0xff] ^ rk[36];
   	t1 = Te0[s1 >> 24] ^ Te1[(s2 >> 16) & 0xff] ^ Te2[(s3 >>  8) & 0xff] ^ Te3[s0 & 0xff] ^ rk[37];
   	t2 = Te0[s2 >> 24] ^ Te1[(s3 >> 16) & 0xff] ^ Te2[(s0 >>  8) & 0xff] ^ Te3[s1 & 0xff] ^ rk[38];
   	t3 = Te0[s3 >> 24] ^ Te1[(s0 >> 16) & 0xff] ^ Te2[(s1 >>  8) & 0xff] ^ Te3[s2 & 0xff] ^ rk[39];
    if (Nr > 10) {
        /* round 10: */
        s0 = Te0[t0 >> 24] ^ Te1[(t1 >> 16) & 0xff] ^ Te2[(t2 >>  8) & 0xff] ^ Te3[t3 & 0xff] ^ rk[40];
        s1 = Te0[t1 >> 24] ^ Te1[(t2 >> 16) & 0xff] ^ Te2[(t3 >>  8) & 0xff] ^ Te3[t0 & 0xff] ^ rk[41];
        s2 = Te0[t2 >> 24] ^ Te1[(t3 >> 16) & 0xff] ^ Te2[(t0 >>  8) & 0xff] ^ Te3[t1 & 0xff] ^ rk[42];
        s3 = Te0[t3 >> 24] ^ Te1[(t0 >> 16) & 0xff] ^ Te2[(t1 >>  8) & 0xff] ^ Te3[t2 & 0xff] ^ rk[43];
        /* round 11: */
        t0 = Te0[s0 >> 24] ^ Te1[(s1 >> 16) & 0xff] ^ Te2[(s2 >>  8) & 0xff] ^ Te3[s3 & 0xff] ^ rk[44];
        t1 = Te0[s1 >> 24] ^ Te1[(s2 >> 16) & 0xff] ^ Te2[(s3 >>  8) & 0xff] ^ Te3[s0 & 0xff] ^ rk[45];
        t2 = Te0[s2 >> 24] ^ Te1[(s3 >> 16) & 0xff] ^ Te2[(s0 >>  8) & 0xff] ^ Te3[s1 & 0xff] ^ rk[46];
        t3 = Te0[s3 >> 24] ^ Te1[(s0 >> 16) & 0xff] ^ Te2[(s1 >>  8) & 0xff] ^ Te3[s2 & 0xff] ^ rk[47];
        if (Nr > 12) {
            /* round 12: */
            s0 = Te0[t0 >> 24] ^ Te1[(t1 >> 16) & 0xff] ^ Te2[(t2 >>  8) & 0xff] ^ Te3[t3 & 0xff] ^ rk[48];
            s1 = Te0[t1 >> 24] ^ Te1[(t2 >> 16) & 0xff] ^ Te2[(t3 >>  8) & 0xff] ^ Te3[t0 & 0xff] ^ rk[49];
            s2 = Te0[t2 >> 24] ^ Te1[(t3 >> 16) & 0xff] ^ Te2[(t0 >>  8) & 0xff] ^ Te3[t1 & 0xff] ^ rk[50];
            s3 = Te0[t3 >> 24] ^ Te1[(t0 >> 16) & 0xff] ^ Te2[(t1 >>  8) & 0xff] ^ Te3[t2 & 0xff] ^ rk[51];
            /* round 13: */
            t0 = Te0[s0 >> 24] ^ Te1[(s1 >> 16) & 0xff] ^ Te2[(s2 >>  8) & 0xff] ^ Te3[s3 & 0xff] ^ rk[52];
            t1 = Te0[s1 >> 24] ^ Te1[(s2 >> 16) & 0xff] ^ Te2[(s3 >>  8) & 0xff] ^ Te3[s0 & 0xff] ^ rk[53];
            t2 = Te0[s2 >> 24] ^ Te1[(s3 >> 16) & 0xff] ^ Te2[(s0 >>  8) & 0xff] ^ Te3[s1 & 0xff] ^ rk[54];
            t3 = Te0[s3 >> 24] ^ Te1[(s0 >> 16) & 0xff] ^ Te2[(s1 >>  8) & 0xff] ^ Te3[s2 & 0xff] ^ rk[55];
        }
    }
    rk += Nr << 2;
#else  /* !FULL_UNROLL */
    /*
	 * Nr - 1 full rounds:
	 */
    r = Nr >> 1;
    for (;;) {
        t0 =
            Te0[(s0 >> 24)       ] ^
            Te1[(s1 >> 16) & 0xff] ^
            Te2[(s2 >>  8) & 0xff] ^
            Te3[(s3      ) & 0xff] ^
            rk[4];
        t1 =
            Te0[(s1 >> 24)       ] ^
            Te1[(s2 >> 16) & 0xff] ^
            Te2[(s3 >>  8) & 0xff] ^
            Te3[(s0      ) & 0xff] ^
            rk[5];
        t2 =
            Te0[(s2 >> 24)       ] ^
            Te1[(s3 >> 16) & 0xff] ^
            Te2[(s0 >>  8) & 0xff] ^
            Te3[(s1      ) & 0xff] ^
            rk[6];
        t3 =
            Te0[(s3 >> 24)       ] ^
            Te1[(s0 >> 16) & 0xff] ^
            Te2[(s1 >>  8) & 0xff] ^
            Te3[(s2      ) & 0xff] ^
            rk[7];

        rk += 8;
        if (--r == 0) {
            break;
        }

        s0 =
            Te0[(t0 >> 24)       ] ^
            Te1[(t1 >> 16) & 0xff] ^
            Te2[(t2 >>  8) & 0xff] ^
            Te3[(t3      ) & 0xff] ^
            rk[0];
        s1 =
            Te0[(t1 >> 24)       ] ^
            Te1[(t2 >> 16) & 0xff] ^
            Te2[(t3 >>  8) & 0xff] ^
            Te3[(t0      ) & 0xff] ^
            rk[1];
        s2 =
            Te0[(t2 >> 24)       ] ^
            Te1[(t3 >> 16) & 0xff] ^
            Te2[(t0 >>  8) & 0xff] ^
            Te3[(t1      ) & 0xff] ^
            rk[2];
        s3 =
            Te0[(t3 >> 24)       ] ^
            Te1[(t0 >> 16) & 0xff] ^
            Te2[(t1 >>  8) & 0xff] ^
            Te3[(t2      ) & 0xff] ^
            rk[3];
    }
#endif /* ?FULL_UNROLL */
    /*
	 * apply last round and
	 * map cipher state to byte array block:
	 */
	s0 =
		(Te4[(t0 >> 24)       ] & 0xff000000) ^
		(Te4[(t1 >> 16) & 0xff] & 0x00ff0000) ^
		(Te4[(t2 >>  8) & 0xff] & 0x0000ff00) ^
		(Te4[(t3      ) & 0xff] & 0x000000ff) ^
		rk[0];
	PUTU32(ct     , s0);
	s1 =
		(Te4[(t1 >> 24)       ] & 0xff000000) ^
		(Te4[(t2 >> 16) & 0xff] & 0x00ff0000) ^
		(Te4[(t3 >>  8) & 0xff] & 0x0000ff00) ^
		(Te4[(t0      ) & 0xff] & 0x000000ff) ^
		rk[1];
	PUTU32(ct +  4, s1);
	s2 =
		(Te4[(t2 >> 24)       ] & 0xff000000) ^
		(Te4[(t3 >> 16) & 0xff] & 0x00ff0000) ^
		(Te4[(t0 >>  8) & 0xff] & 0x0000ff00) ^
		(Te4[(t1      ) & 0xff] & 0x000000ff) ^
		rk[2];
	PUTU32(ct +  8, s2);
	s3 =
		(Te4[(t3 >> 24)       ] & 0xff000000) ^
		(Te4[(t0 >> 16) & 0xff] & 0x00ff0000) ^
		(Te4[(t1 >>  8) & 0xff] & 0x0000ff00) ^
		(Te4[(t2      ) & 0xff] & 0x000000ff) ^
		rk[3];
	PUTU32(ct + 12, s3);
}

void rijndaelDecrypt(u32 rk[/*4*(Nr + 1)*/], int Nr, const u8 ct[16], u8 pt[16]) {
	u32 s0, s1, s2, s3, t0, t1, t2, t3;
#ifndef FULL_UNROLL
    int r;
#endif /* ?FULL_UNROLL */

    /*
	 * map byte array block to cipher state
	 * and add initial round key:
	 */
    s0 = GETU32(ct     ) ^ rk[0];
    s1 = GETU32(ct +  4) ^ rk[1];
    s2 = GETU32(ct +  8) ^ rk[2];
    s3 = GETU32(ct + 12) ^ rk[3];
#ifdef FULL_UNROLL
    /* round 1: */
    t0 = Td0[s0 >> 24] ^ Td1[(s3 >> 16) & 0xff] ^ Td2[(s2 >>  8) & 0xff] ^ Td3[s1 & 0xff] ^ rk[ 4];
    t1 = Td0[s1 >> 24] ^ Td1[(s0 >> 16) & 0xff] ^ Td2[(s3 >>  8) & 0xff] ^ Td3[s2 & 0xff] ^ rk[ 5];
    t2 = Td0[s2 >> 24] ^ Td1[(s1 >> 16) & 0xff] ^ Td2[(s0 >>  8) & 0xff] ^ Td3[s3 & 0xff] ^ rk[ 6];
    t3 = Td0[s3 >> 24] ^ Td1[(s2 >> 16) & 0xff] ^ Td2[(s1 >>  8) & 0xff] ^ Td3[s0 & 0xff] ^ rk[ 7];
    /* round 2: */
    s0 = Td0[t0 >> 24] ^ Td1[(t3 >> 16) & 0xff] ^ Td2[(t2 >>  8) & 0xff] ^ Td3[t1 & 0xff] ^ rk[ 8];
    s1 = Td0[t1 >> 24] ^ Td1[(t0 >> 16) & 0xff] ^ Td2[(t3 >>  8) & 0xff] ^ Td3[t2 & 0xff] ^ rk[ 9];
    s2 = Td0[t2 >> 24] ^ Td1[(t1 >> 16) & 0xff] ^ Td2[(t0 >>  8) & 0xff] ^ Td3[t3 & 0xff] ^ rk[10];
    s3 = Td0[t3 >> 24] ^ Td1[(t2 >> 16) & 0xff] ^ Td2[(t1 >>  8) & 0xff] ^ Td3[t0 & 0xff] ^ rk[11];
    /* round 3: */
    t0 = Td0[s0 >> 24] ^ Td1[(s3 >> 16) & 0xff] ^ Td2[(s2 >>  8) & 0xff] ^ Td3[s1 & 0xff] ^ rk[12];
    t1 = Td0[s1 >> 24] ^ Td1[(s0 >> 16) & 0xff] ^ Td2[(s3 >>  8) & 0xff] ^ Td3[s2 & 0xff] ^ rk[13];
    t2 = Td0[s2 >> 24] ^ Td1[(s1 >> 16) & 0xff] ^ Td2[(s0 >>  8) & 0xff] ^ Td3[s3 & 0xff] ^ rk[14];
    t3 = Td0[s3 >> 24] ^ Td1[(s2 >> 16) & 0xff] ^ Td2[(s1 >>  8) & 0xff] ^ Td3[s0 & 0xff] ^ rk[15];
    /* round 4: */
    s0 = Td0[t0 >> 24] ^ Td1[(t3 >> 16) & 0xff] ^ Td2[(t2 >>  8) & 0xff] ^ Td3[t1 & 0xff] ^ rk[16];
    s1 = Td0[t1 >> 24] ^ Td1[(t0 >> 16) & 0xff] ^ Td2[(t3 >>  8) & 0xff] ^ Td3[t2 & 0xff] ^ rk[17];
    s2 = Td0[t2 >> 24] ^ Td1[(t1 >> 16) & 0xff] ^ Td2[(t0 >>  8) & 0xff] ^ Td3[t3 & 0xff] ^ rk[18];
    s3 = Td0[t3 >> 24] ^ Td1[(t2 >> 16) & 0xff] ^ Td2[(t1 >>  8) & 0xff] ^ Td3[t0 & 0xff] ^ rk[19];
    /* round 5: */
    t0 = Td0[s0 >> 24] ^ Td1[(s3 >> 16) & 0xff] ^ Td2[(s2 >>  8) & 0xff] ^ Td3[s1 & 0xff] ^ rk[20];
    t1 = Td0[s1 >> 24] ^ Td1[(s0 >> 16) & 0xff] ^ Td2[(s3 >>  8) & 0xff] ^ Td3[s2 & 0xff] ^ rk[21];
    t2 = Td0[s2 >> 24] ^ Td1[(s1 >> 16) & 0xff] ^ Td2[(s0 >>  8) & 0xff] ^ Td3[s3 & 0xff] ^ rk[22];
    t3 = Td0[s3 >> 24] ^ Td1[(s2 >> 16) & 0xff] ^ Td2[(s1 >>  8) & 0xff] ^ Td3[s0 & 0xff] ^ rk[23];
    /* round 6: */
    s0 = Td0[t0 >> 24] ^ Td1[(t3 >> 16) & 0xff] ^ Td2[(t2 >>  8) & 0xff] ^ Td3[t1 & 0xff] ^ rk[24];
    s1 = Td0[t1 >> 24] ^ Td1[(t0 >> 16) & 0xff] ^ Td2[(t3 >>  8) & 0xff] ^ Td3[t2 & 0xff] ^ rk[25];
    s2 = Td0[t2 >> 24] ^ Td1[(t1 >> 16) & 0xff] ^ Td2[(t0 >>  8) & 0xff] ^ Td3[t3 & 0xff] ^ rk[26];
    s3 = Td0[t3 >> 24] ^ Td1[(t2 >> 16) & 0xff] ^ Td2[(t1 >>  8) & 0xff] ^ Td3[t0 & 0xff] ^ rk[27];
    /* round 7: */
    t0 = Td0[s0 >> 24] ^ Td1[(s3 >> 16) & 0xff] ^ Td2[(s2 >>  8) & 0xff] ^ Td3[s1 & 0xff] ^ rk[28];
    t1 = Td0[s1 >> 24] ^ Td1[(s0 >> 16) & 0xff] ^ Td2[(s3 >>  8) & 0xff] ^ Td3[s2 & 0xff] ^ rk[29];
    t2 = Td0[s2 >> 24] ^ Td1[(s1 >> 16) & 0xff] ^ Td2[(s0 >>  8) & 0xff] ^ Td3[s3 & 0xff] ^ rk[30];
    t3 = Td0[s3 >> 24] ^ Td1[(s2 >> 16) & 0xff] ^ Td2[(s1 >>  8) & 0xff] ^ Td3[s0 & 0xff] ^ rk[31];
    /* round 8: */
    s0 = Td0[t0 >> 24] ^ Td1[(t3 >> 16) & 0xff] ^ Td2[(t2 >>  8) & 0xff] ^ Td3[t1 & 0xff] ^ rk[32];
    s1 = Td0[t1 >> 24] ^ Td1[(t0 >> 16) & 0xff] ^ Td2[(t3 >>  8) & 0xff] ^ Td3[t2 & 0xff] ^ rk[33];
    s2 = Td0[t2 >> 24] ^ Td1[(t1 >> 16) & 0xff] ^ Td2[(t0 >>  8) & 0xff] ^ Td3[t3 & 0xff] ^ rk[34];
    s3 = Td0[t3 >> 24] ^ Td1[(t2 >> 16) & 0xff] ^ Td2[(t1 >>  8) & 0xff] ^ Td3[t0 & 0xff] ^ rk[35];
    /* round 9: */
    t0 = Td0[s0 >> 24] ^ Td1[(s3 >> 16) & 0xff] ^ Td2[(s2 >>  8) & 0xff] ^ Td3[s1 & 0xff] ^ rk[36];
    t1 = Td0[s1 >> 24] ^ Td1[(s0 >> 16) & 0xff] ^ Td2[(s3 >>  8) & 0xff] ^ Td3[s2 & 0xff] ^ rk[37];
    t2 = Td0[s2 >> 24] ^ Td1[(s1 >> 16) & 0xff] ^ Td2[(s0 >>  8) & 0xff] ^ Td3[s3 & 0xff] ^ rk[38];
    t3 = Td0[s3 >> 24] ^ Td1[(s2 >> 16) & 0xff] ^ Td2[(s1 >>  8) & 0xff] ^ Td3[s0 & 0xff] ^ rk[39];
    if (Nr > 10) {
        /* round 10: */
        s0 = Td0[t0 >> 24] ^ Td1[(t3 >> 16) & 0xff] ^ Td2[(t2 >>  8) & 0xff] ^ Td3[t1 & 0xff] ^ rk[40];
        s1 = Td0[t1 >> 24] ^ Td1[(t0 >> 16) & 0xff] ^ Td2[(t3 >>  8) & 0xff] ^ Td3[t2 & 0xff] ^ rk[41];
        s2 = Td0[t2 >> 24] ^ Td1[(t1 >> 16) & 0xff] ^ Td2[(t0 >>  8) & 0xff] ^ Td3[t3 & 0xff] ^ rk[42];
        s3 = Td0[t3 >> 24] ^ Td1[(t2 >> 16) & 0xff] ^ Td2[(t1 >>  8) & 0xff] ^ Td3[t0 & 0xff] ^ rk[43];
        /* round 11: */
        t0 = Td0[s0 >> 24] ^ Td1[(s3 >> 16) & 0xff] ^ Td2[(s2 >>  8) & 0xff] ^ Td3[s1 & 0xff] ^ rk[44];
        t1 = Td0[s1 >> 24] ^ Td1[(s0 >> 16) & 0xff] ^ Td2[(s3 >>  8) & 0xff] ^ Td3[s2 & 0xff] ^ rk[45];
        t2 = Td0[s2 >> 24] ^ Td1[(s1 >> 16) & 0xff] ^ Td2[(s0 >>  8) & 0xff] ^ Td3[s3 & 0xff] ^ rk[46];
        t3 = Td0[s3 >> 24] ^ Td1[(s2 >> 16) & 0xff] ^ Td2[(s1 >>  8) & 0xff] ^ Td3[s0 & 0xff] ^ rk[47];
        if (Nr > 12) {
            /* round 12: */
            s0 = Td0[t0 >> 24] ^ Td1[(t3 >> 16) & 0xff] ^ Td2[(t2 >>  8) & 0xff] ^ Td3[t1 & 0xff] ^ rk[48];
            s1 = Td0[t1 >> 24] ^ Td1[(t0 >> 16) & 0xff] ^ Td2[(t3 >>  8) & 0xff] ^ Td3[t2 & 0xff] ^ rk[49];
            s2 = Td0[t2 >> 24] ^ Td1[(t1 >> 16) & 0xff] ^ Td2[(t0 >>  8) & 0xff] ^ Td3[t3 & 0xff] ^ rk[50];
            s3 = Td0[t3 >> 24] ^ Td1[(t2 >> 16) & 0xff] ^ Td2[(t1 >>  8) & 0xff] ^ Td3[t0 & 0xff] ^ rk[51];
            /* round 13: */
            t0 = Td0[s0 >> 24] ^ Td1[(s3 >> 16) & 0xff] ^ Td2[(s2 >>  8) & 0xff] ^ Td3[s1 & 0xff] ^ rk[52];
            t1 = Td0[s1 >> 24] ^ Td1[(s0 >> 16) & 0xff] ^ Td2[(s3 >>  8) & 0xff] ^ Td3[s2 & 0xff] ^ rk[53];
            t2 = Td0[s2 >> 24] ^ Td1[(s1 >> 16) & 0xff] ^ Td2[(s0 >>  8) & 0xff] ^ Td3[s3 & 0xff] ^ rk[54];
            t3 = Td0[s3 >> 24] ^ Td1[(s2 >> 16) & 0xff] ^ Td2[(s1 >>  8) & 0xff] ^ Td3[s0 & 0xff] ^ rk[55];
        }
    }
	rk += Nr << 2;
#else  /* !FULL_UNROLL */
    /*
     * Nr - 1 full rounds: