18 #define load_8way(src, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
19 	vmovdqu (0*16)(src), x0; \
20 	vmovdqu (1*16)(src), x1; \
21 	vmovdqu (2*16)(src), x2; \
22 	vmovdqu (3*16)(src), x3; \
23 	vmovdqu (4*16)(src), x4; \
24 	vmovdqu (5*16)(src), x5; \
25 	vmovdqu (6*16)(src), x6; \
26 	vmovdqu (7*16)(src), x7;
38 #define store_cbc_8way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
39 	vpxor (0*16)(src), x1, x1; \
40 	vpxor (1*16)(src), x2, x2; \
41 	vpxor (2*16)(src), x3, x3; \
42 	vpxor (3*16)(src), x4, x4; \
43 	vpxor (4*16)(src), x5, x5; \
44 	vpxor (5*16)(src), x6, x6; \
45 	vpxor (6*16)(src), x7, x7; \
82 #define store_ctr_8way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \  argument
83 	vpxor (0*16)(src), x0, x0; \
84 	vpxor (1*16)(src), x1, x1; \
85 	vpxor (2*16)(src), x2, x2; \
86 	vpxor (3*16)(src), x3, x3; \
87 	vpxor (4*16)(src), x4, x4; \
88 	vpxor (5*16)(src), x5, x5; \
89 	vpxor (6*16)(src), x6, x6; \
90 	vpxor (7*16)(src), x7, x7; \
100 #define load_xts_8way(iv, src, dst, x0, x1, x2, x3, x4, x5, x6, x7, tiv, t0, \  argument
106 	vpxor (0*16)(src), tiv, x0; \
111 	vpxor (1*16)(src), tiv, x1; \
115 	vpxor (2*16)(src), tiv, x2; \
119 	vpxor (3*16)(src), tiv, x3; \
123 	vpxor (4*16)(src), tiv, x4; \
127 	vpxor (5*16)(src), tiv, x5; \
131 	vpxor (6*16)(src), tiv, x6; \
135 	vpxor (7*16)(src), tiv, x7; \