8 #define load_16way(src, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
12 	vmovdqu (3*32)(src), x3; \
18 #define store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
22 	vmovdqu x3, (3*32)(dst); \
28 #define store_cbc_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7, t0) \  argument
34 	vpxor (2*32+16)(src), x3, x3; \
39 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
55 #define load_ctr_16way(iv, bswap, x0, x1, x2, x3, x4, x5, x6, x7, t0, t0x, t1, \  argument
75 	vpshufb t1, t2, x3; \
88 #define store_ctr_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
92 	vpxor (3*32)(src), x3, x3; \
97 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
118 #define load_xts_16way(iv, src, dst, x0, x1, x2, x3, x4, x5, x6, x7, tiv, \  argument
143 	vpxor (3*32)(src), tiv, x3; \
166 #define store_xts_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
170 	vpxor (3*32)(dst), x3, x3; \
175 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);