8 #define load_8way(src, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
13 	vmovdqu (4*16)(src), x4; \
18 #define store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
23 	vmovdqu x4, (4*16)(dst); \
28 #define store_cbc_8way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
32 	vpxor (3*16)(src), x4, x4; \
36 	store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
44 #define load_ctr_8way(iv, bswap, x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2) \  argument
61 	vpshufb t1, x7, x4; \
72 #define store_ctr_8way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
77 	vpxor (4*16)(src), x4, x4; \
81 	store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
90 #define load_xts_8way(iv, src, dst, x0, x1, x2, x3, x4, x5, x6, x7, tiv, t0, \  argument
113 	vpxor (4*16)(src), tiv, x4; \
131 #define store_xts_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
136 	vpxor (4*16)(dst), x4, x4; \
140 	store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7);