88 #define do16bit_ror(rot, op1, op2, T0, T1, tmp1, tmp2, ab, dst) \  argument
89 	movzbl ab ## bl,		tmp2 ## d; \
90 	movzbl ab ## bh,		tmp1 ## d; \
91 	rorq $(rot),			ab; \
99 #define g1g2_3(ab, cd, Tx0, Tx1, Tx2, Tx3, Ty0, Ty1, Ty2, Ty3, x, y) \  argument
101 	do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 0, ab ## 0, x ## 0); \
102 	do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 0, ab ## 0, y ## 0); \
104 	do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 1, ab ## 1, x ## 1); \
105 	do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 1, ab ## 1, y ## 1); \
107 	do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 2, ab ## 2, x ## 2); \
108 	do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 2, ab ## 2, y ## 2); \
111 	do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 0, x ## 0); \
112 	do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 0, y ## 0); \
113 	xchgq cd ## 0, ab ## 0; \
115 	do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 1, x ## 1); \
116 	do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 1, y ## 1); \
117 	xchgq cd ## 1, ab ## 1; \
119 	do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 2, x ## 2); \
120 	do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 2, y ## 2); \
121 	xchgq cd ## 2, ab ## 2;
123 #define enc_round_end(ab, x, y, n) \  argument
127 	xorl ab ## d,			x ## d; \
129 	shrq $32,			ab; \
130 	roll $1,			ab ## d; \
131 	xorl y ## d,			ab ## d; \
132 	shlq $32,			ab; \
134 	orq x,				ab;
149 #define encrypt_round3(ab, cd, n) \  argument
150 	g1g2_3(ab, cd, s0, s1, s2, s3, s0, s1, s2, s3, RX, RY); \
152 	enc_round_end(ab ## 0, RX0, RY0, n); \
153 	enc_round_end(ab ## 1, RX1, RY1, n); \
154 	enc_round_end(ab ## 2, RX2, RY2, n);
163 #define encrypt_cycle3(ab, cd, n) \  argument
164 	encrypt_round3(ab, cd, n*2); \
165 	encrypt_round3(ab, cd, (n*2)+1);