18 #define load_8way(src, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
26 	vmovdqu (7*16)(src), x7;
28 #define store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
36 	vmovdqu x7, (7*16)(dst);
38 #define store_cbc_8way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
45 	vpxor (6*16)(src), x7, x7; \
46 	store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
54 #define load_ctr_8way(iv, bswap, x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2) \  argument
60 	vmovdqu (iv), x7; \
61 	vpshufb t1, x7, x0; \
64 	inc_le128(x7, t0, t2); \
65 	vpshufb t1, x7, x1; \
66 	inc_le128(x7, t0, t2); \
67 	vpshufb t1, x7, x2; \
68 	inc_le128(x7, t0, t2); \
69 	vpshufb t1, x7, x3; \
70 	inc_le128(x7, t0, t2); \
71 	vpshufb t1, x7, x4; \
72 	inc_le128(x7, t0, t2); \
73 	vpshufb t1, x7, x5; \
74 	inc_le128(x7, t0, t2); \
75 	vpshufb t1, x7, x6; \
76 	inc_le128(x7, t0, t2); \
77 	vmovdqa x7, t2; \
78 	vpshufb t1, x7, x7; \
82 #define store_ctr_8way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
90 	vpxor (7*16)(src), x7, x7; \
91 	store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
100 #define load_xts_8way(iv, src, dst, x0, x1, x2, x3, x4, x5, x6, x7, tiv, t0, \  argument
135 	vpxor (7*16)(src), tiv, x7; \
141 #define store_xts_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
149 	vpxor (7*16)(dst), x7, x7; \
150 	store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7);