8 #define load_16way(src, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
11 	vmovdqu (2*32)(src), x2; \
18 #define store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
21 	vmovdqu x2, (2*32)(dst); \
28 #define store_cbc_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7, t0) \  argument
33 	vpxor (1*32+16)(src), x2, x2; \
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
73 	vpshufb t1, t2, x2; \
88 #define store_ctr_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
91 	vpxor (2*32)(src), x2, x2; \
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
139 	vpxor (2*32)(src), tiv, x2; \
166 #define store_xts_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
169 	vpxor (2*32)(dst), x2, x2; \
175 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);