Debian Codegraph @

crypto_kem/6960119pcf/vec/bm.c

raw
1 /*
2 This file is for implementating the inversion-free Berlekamp-Massey algorithm
3 see https://ieeexplore.ieee.org/document/87857
4
5 For the implementation strategy, see
6 https://eprint.iacr.org/2017/793.pdf
7 */
8 // 20240805 djb: more use of cryptoint
9 // 20240503 djb: use crypto_*_mask functions
10 // 20221230 djb: add linker lines
11
12 // linker define bm
13 // linker use vec_mul
14
15 #include "bm.h"
16
17 #include "gf.h"
18 #include "crypto_uint64.h"
19 #include "crypto_int64.h"
20
21 static inline void vec_cmov(vec * out, vec * in, uint16_t mask)
22 {
23 int i;
24
25 vec m0, m1;
26
27 m0 = vec_set1_16b(mask);
28 m1 = ~m0;
29
30 for (i = 0; i < GFBITS; i++)
31 {
32 out[i] = (in[i] & m0) | (out[i] & m1);
33 out[i] = (in[i] & m0) | (out[i] & m1);
34 }
35 }
36
37 static inline void interleave(vec *in, int idx0, int idx1, vec *mask, int b)
38 {
39 int s = 1 << b;
40
41 vec x, y;
42
43 x = (in[idx0] & mask[0]) | ((in[idx1] & mask[0]) << s);
44 y = ((in[idx0] & mask[1]) >> s) | (in[idx1] & mask[1]);
45
46 in[idx0] = x;
47 in[idx1] = y;
48 }
49
50 /* input: in, field elements in bitsliced form */
51 /* output: out, field elements in non-bitsliced form */
52 static inline void get_coefs(gf *out, vec *in)
53 {
54 int i, k;
55
56 vec mask[4][2];
57 vec buf[16];
58
59 for (i = 0; i < 13; i++) buf[i] = in[i];
60 for (i = 13; i < 16; i++) buf[i] = 0;
61
62 mask[0][0] = vec_set1_16b(0x5555);
63 mask[0][1] = vec_set1_16b(0xAAAA);
64 mask[1][0] = vec_set1_16b(0x3333);
65 mask[1][1] = vec_set1_16b(0xCCCC);
66 mask[2][0] = vec_set1_16b(0x0F0F);
67 mask[2][1] = vec_set1_16b(0xF0F0);
68 mask[3][0] = vec_set1_16b(0x00FF);
69 mask[3][1] = vec_set1_16b(0xFF00);
70
71 interleave(buf, 0, 8, mask[3], 3);
72 interleave(buf, 1, 9, mask[3], 3);
73 interleave(buf, 2, 10, mask[3], 3);
74 interleave(buf, 3, 11, mask[3], 3);
75 interleave(buf, 4, 12, mask[3], 3);
76 interleave(buf, 5, 13, mask[3], 3);
77 interleave(buf, 6, 14, mask[3], 3);
78 interleave(buf, 7, 15, mask[3], 3);
79
80 interleave(buf, 0, 4, mask[2], 2);
81 interleave(buf, 1, 5, mask[2], 2);
82 interleave(buf, 2, 6, mask[2], 2);
83 interleave(buf, 3, 7, mask[2], 2);
84 interleave(buf, 8, 12, mask[2], 2);
85 interleave(buf, 9, 13, mask[2], 2);
86 interleave(buf, 10, 14, mask[2], 2);
87 interleave(buf, 11, 15, mask[2], 2);
88
89 interleave(buf, 0, 2, mask[1], 1);
90 interleave(buf, 1, 3, mask[1], 1);
91 interleave(buf, 4, 6, mask[1], 1);
92 interleave(buf, 5, 7, mask[1], 1);
93 interleave(buf, 8, 10, mask[1], 1);
94 interleave(buf, 9, 11, mask[1], 1);
95 interleave(buf, 12, 14, mask[1], 1);
96 interleave(buf, 13, 15, mask[1], 1);
97
98 interleave(buf, 0, 1, mask[0], 0);
99 interleave(buf, 2, 3, mask[0], 0);
100 interleave(buf, 4, 5, mask[0], 0);
101 interleave(buf, 6, 7, mask[0], 0);
102 interleave(buf, 8, 9, mask[0], 0);
103 interleave(buf, 10, 11, mask[0], 0);
104 interleave(buf, 12, 13, mask[0], 0);
105 interleave(buf, 14, 15, mask[0], 0);
106
107 for (i = 0; i < 16; i++)
108 for (k = 0; k < 4; k++)
109 out[ k*16 + i ] = (buf[i] >> (k*16)) & GFMASK;
110 }
111
112 static void update(vec in[][GFBITS], const gf e)
113 {
114 int i;
115 vec tmp;
116
117 for (i = 0; i < GFBITS; i++)
118 {
119 tmp = crypto_int64_bitmod_01(e, i);
120
121 in[0][i] = (in[0][i] >> 1) | (in[1][i] << 63);
122 in[1][i] = (in[1][i] >> 1) | (tmp << 63);
123 }
124 }
125
126 static inline gf vec_reduce(vec in[][GFBITS])
127 {
128 int i;
129 vec tmp;
130 gf ret = 0;
131
132 for (i = GFBITS-1; i >= 0; i--)
133 {
134 tmp = in[0][i] ^ in[1][i];
135
136 tmp ^= tmp >> 32;
137 tmp ^= tmp >> 16;
138 tmp ^= tmp >> 8;
139 tmp ^= tmp >> 4;
140 tmp ^= tmp >> 2;
141 tmp ^= tmp >> 1;
142
143 ret <<= 1;
144 ret |= crypto_int64_bottombit_01(tmp);
145 }
146
147 return ret;
148 }
149
150 /* input: in, sequence of field elements */
151 /* output: out, minimal polynomial of in */
152 //void bm(vec out[][ GFBITS ], vec in[][ GFBITS ])
153 void bm(vec out[][GFBITS], vec in[][ GFBITS ])
154 {
155 int i;
156 uint16_t N, L;
157 uint16_t mask;
158 uint64_t one = 1;
159
160 vec prod[2][GFBITS];
161 vec interval[2][GFBITS];
162 vec dd[2][GFBITS], bb[2][GFBITS];
163 vec B[2][GFBITS], C[2][GFBITS];
164 vec B_tmp[2][GFBITS], C_tmp[2][GFBITS];
165
166 gf d, b;
167 gf coefs[256];
168
169 // initialization
170
171 get_coefs(&coefs[ 0], in[0]);
172 get_coefs(&coefs[ 64], in[1]);
173 get_coefs(&coefs[128], in[2]);
174 get_coefs(&coefs[192], in[3]);
175
176 C[0][0] = 0;
177 C[1][0] = one << 63;
178 B[0][0] = 0;
179 B[1][0] = one << 62;
180
181 for (i = 1; i < GFBITS; i++)
182 C[0][i] = C[1][i] = B[0][i] = B[1][i] = 0;
183
184 b = 1;
185 L = 0;
186
187 //
188
189 for (i = 0; i < GFBITS; i++)
190 interval[0][i] = interval[1][i] = 0;
191
192 for (N = 0; N < SYS_T*2; N++)
193 {
194 update(interval, coefs[N]);
195
196 vec_mul(prod[0], C[0], interval[0]);
197 vec_mul(prod[1], C[1], interval[1]);
198
199 d = vec_reduce(prod);
200
201 mask = crypto_uint64_nonzero_mask(d) & crypto_uint64_leq_mask(L*2, N);
202
203 for (i = 0; i < GFBITS; i++)
204 {
205 dd[0][i] = dd[1][i] = vec_setbits(crypto_int64_bitmod_01(d, i));
206 bb[0][i] = bb[1][i] = vec_setbits(crypto_int64_bitmod_01(b, i));
207 }
208
209 vec_mul(B_tmp[0], dd[0], B[0]);
210 vec_mul(B_tmp[1], dd[1], B[1]);
211 vec_mul(C_tmp[0], bb[0], C[0]);
212 vec_mul(C_tmp[1], bb[1], C[1]);
213
214 vec_cmov(B[0], C[0], mask);
215 vec_cmov(B[1], C[1], mask);
216 update(B, 0);
217
218 for (i = 0; i < GFBITS; i++)
219 {
220 C[0][i] = B_tmp[0][i] ^ C_tmp[0][i];
221 C[1][i] = B_tmp[1][i] ^ C_tmp[1][i];
222 }
223
224 b = (d & mask) | (b & ~mask);
225 L = ((N+1-L) & mask) | (L & ~mask);
226 }
227
228 for (i = 0; i < GFBITS; i++)
229 {
230 out[0][i] = (C[0][i] >> 8) | (C[1][i] << 56);
231 out[1][i] = C[1][i] >> 8;
232 }
233 }
234