13 #define load_16way(src, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
18 	vmovdqu (4*32)(src), x4; \
23 #define store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
28 	vmovdqu x4, (4*32)(dst); \
33 #define store_cbc_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7, t0) \  argument
40 	vpxor (3*32+16)(src), x4, x4; \
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
82 	vpshufb t1, t2, x4; \
93 #define store_ctr_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
98 	vpxor (4*32)(src), x4, x4; \
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
152 	vpxor (4*32)(src), tiv, x4; \
171 #define store_xts_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
176 	vpxor (4*32)(dst), x4, x4; \
180 	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);