13 #define load_16way(src, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
21 	vmovdqu (7*32)(src), x7;
23 #define store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
31 	vmovdqu x7, (7*32)(dst);
33 #define store_cbc_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7, t0) \  argument
43 	vpxor (6*32+16)(src), x7, x7; \
44 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
60 #define load_ctr_16way(iv, bswap, x0, x1, x2, x3, x4, x5, x6, x7, t0, t0x, t1, \  argument
88 	vpshufb t1, t2, x7; \
93 #define store_ctr_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
101 	vpxor (7*32)(src), x7, x7; \
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
164 	vpxor (7*32)(src), tiv, x7; \
171 #define store_xts_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
179 	vpxor (7*32)(dst), x7, x7; \
180 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);