13 #define load_16way(src, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
16 	vmovdqu (2*32)(src), x2; \
23 #define store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
26 	vmovdqu x2, (2*32)(dst); \
33 #define store_cbc_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7, t0) \  argument
38 	vpxor (1*32+16)(src), x2, x2; \
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
78 	vpshufb t1, t2, x2; \
93 #define store_ctr_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
96 	vpxor (2*32)(src), x2, x2; \
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
144 	vpxor (2*32)(src), tiv, x2; \
171 #define store_xts_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
174 	vpxor (2*32)(dst), x2, x2; \
180 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);