23 #define store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
24 	vmovdqu x0, (0*32)(dst); \
25 	vmovdqu x1, (1*32)(dst); \
26 	vmovdqu x2, (2*32)(dst); \
27 	vmovdqu x3, (3*32)(dst); \
28 	vmovdqu x4, (4*32)(dst); \
29 	vmovdqu x5, (5*32)(dst); \
30 	vmovdqu x6, (6*32)(dst); \
31 	vmovdqu x7, (7*32)(dst);
33 #define store_cbc_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7, t0) \  argument
44 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
93 #define store_ctr_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
102 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
123 #define load_xts_16way(iv, src, dst, x0, x1, x2, x3, x4, x5, x6, x7, tiv, \  argument
136 	vmovdqu tiv, (0*32)(dst); \
141 	vmovdqu tiv, (1*32)(dst); \
145 	vmovdqu tiv, (2*32)(dst); \
149 	vmovdqu tiv, (3*32)(dst); \
153 	vmovdqu tiv, (4*32)(dst); \
157 	vmovdqu tiv, (5*32)(dst); \
161 	vmovdqu tiv, (6*32)(dst); \
165 	vmovdqu tiv, (7*32)(dst); \
171 #define store_xts_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
172 	vpxor (0*32)(dst), x0, x0; \
173 	vpxor (1*32)(dst), x1, x1; \
174 	vpxor (2*32)(dst), x2, x2; \
175 	vpxor (3*32)(dst), x3, x3; \
176 	vpxor (4*32)(dst), x4, x4; \
177 	vpxor (5*32)(dst), x5, x5; \
178 	vpxor (6*32)(dst), x6, x6; \
179 	vpxor (7*32)(dst), x7, x7; \
180 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);