1/*
2   This file is adapted from amd64-51/fe25519_invert.c:
3   Loops of squares are replaced by nsquares for better performance.
4*/
5
6#include "fe51.h"
7
8#ifdef HAVE_AVX_ASM
9
10#define fe51_square(x, y) fe51_nsquare(x, y, 1)
11
12void
13fe51_invert(fe51 *r, const fe51 *x)
14{
15    fe51 z2;
16    fe51 z9;
17    fe51 z11;
18    fe51 z2_5_0;
19    fe51 z2_10_0;
20    fe51 z2_20_0;
21    fe51 z2_50_0;
22    fe51 z2_100_0;
23    fe51 t;
24
25    /* 2 */ fe51_square(&z2,x);
26    /* 4 */ fe51_square(&t,&z2);
27    /* 8 */ fe51_square(&t,&t);
28    /* 9 */ fe51_mul(&z9,&t,x);
29    /* 11 */ fe51_mul(&z11,&z9,&z2);
30    /* 22 */ fe51_square(&t,&z11);
31    /* 2^5 - 2^0 = 31 */ fe51_mul(&z2_5_0,&t,&z9);
32
33    /* 2^10 - 2^5 */ fe51_nsquare(&t,&z2_5_0, 5);
34    /* 2^10 - 2^0 */ fe51_mul(&z2_10_0,&t,&z2_5_0);
35
36    /* 2^20 - 2^10 */ fe51_nsquare(&t,&z2_10_0, 10);
37    /* 2^20 - 2^0 */ fe51_mul(&z2_20_0,&t,&z2_10_0);
38
39    /* 2^40 - 2^20 */ fe51_nsquare(&t,&z2_20_0, 20);
40    /* 2^40 - 2^0 */ fe51_mul(&t,&t,&z2_20_0);
41
42    /* 2^50 - 2^10 */ fe51_nsquare(&t,&t,10);
43    /* 2^50 - 2^0 */ fe51_mul(&z2_50_0,&t,&z2_10_0);
44
45    /* 2^100 - 2^50 */ fe51_nsquare(&t,&z2_50_0, 50);
46    /* 2^100 - 2^0 */ fe51_mul(&z2_100_0,&t,&z2_50_0);
47
48    /* 2^200 - 2^100 */ fe51_nsquare(&t,&z2_100_0, 100);
49    /* 2^200 - 2^0 */ fe51_mul(&t,&t,&z2_100_0);
50
51    /* 2^250 - 2^50 */ fe51_nsquare(&t,&t, 50);
52    /* 2^250 - 2^0 */ fe51_mul(&t,&t,&z2_50_0);
53
54    /* 2^255 - 2^5 */ fe51_nsquare(&t,&t,5);
55    /* 2^255 - 21 */ fe51_mul(r,&t,&z11);
56}
57
58#endif
59