| 1 |
/* |
| 2 |
This file is for Benes network related functions |
| 3 |
|
| 4 |
For the implementation strategy, see |
| 5 |
https://eprint.iacr.org/2017/793.pdf |
| 6 |
*/ |
| 7 |
// 20221231 djb: remove unused counter increment; tnx thom wiggers |
| 8 |
// 20221230 djb: add linker lines |
| 9 |
|
| 10 |
// linker define load_bits benes |
| 11 |
// linker use transpose_64x128_sp_asm |
| 12 |
|
| 13 |
#include "benes.h" |
| 14 |
|
| 15 |
#include "util.h" |
| 16 |
#include "transpose.h" |
| 17 |
#include "params.h" |
| 18 |
|
| 19 |
static void layer_x(vec128 * data, vec128 * bits) |
| 20 |
{ |
| 21 |
int i; |
| 22 |
vec128 v0, v1; |
| 23 |
vec128 d; |
| 24 |
|
| 25 |
for (i = 0; i < 64; i += 2) |
| 26 |
{ |
| 27 |
v0 = vec128_unpack_low(data[i+0], data[i+1]); |
| 28 |
v1 = vec128_unpack_high(data[i+0], data[i+1]); |
| 29 |
|
| 30 |
d = vec128_xor(v0, v1); |
| 31 |
d = vec128_and(d, *bits++); |
| 32 |
v0 = vec128_xor(v0, d); |
| 33 |
v1 = vec128_xor(v1, d); |
| 34 |
|
| 35 |
data[i+0] = vec128_unpack_low(v0, v1); |
| 36 |
data[i+1] = vec128_unpack_high(v0, v1); |
| 37 |
} |
| 38 |
} |
| 39 |
|
| 40 |
static void layer_0(vec128 *bs, vec128 *cond) |
| 41 |
{ |
| 42 |
int x; |
| 43 |
vec128 diff; |
| 44 |
|
| 45 |
for (x = 0; x < (1 << 6); x += 2) |
| 46 |
{ |
| 47 |
diff = vec128_xor(bs[ x ], bs[ x+1 ]); |
| 48 |
diff = vec128_and(diff, *cond++); |
| 49 |
bs[ x ] = vec128_xor(bs[ x ], diff); |
| 50 |
bs[ x+1 ] = vec128_xor(bs[ x+1 ], diff); |
| 51 |
} |
| 52 |
} |
| 53 |
|
| 54 |
static void layer_1(vec128 *bs, vec128 *cond) |
| 55 |
{ |
| 56 |
int x; |
| 57 |
vec128 diff; |
| 58 |
|
| 59 |
for (x = 0; x < (1 << 6); x += 4) |
| 60 |
{ |
| 61 |
diff = vec128_xor(bs[ x+0 ], bs[ x+2 ]); |
| 62 |
diff = vec128_and(diff, cond[0]); |
| 63 |
bs[ x+0 ] = vec128_xor(bs[ x+0 ], diff); |
| 64 |
bs[ x+2 ] = vec128_xor(bs[ x+2 ], diff); |
| 65 |
|
| 66 |
diff = vec128_xor(bs[ x+1 ], bs[ x+3 ]); |
| 67 |
diff = vec128_and(diff, cond[1]); |
| 68 |
bs[ x+1 ] = vec128_xor(bs[ x+1 ], diff); |
| 69 |
bs[ x+3 ] = vec128_xor(bs[ x+3 ], diff); |
| 70 |
|
| 71 |
cond += 2; |
| 72 |
} |
| 73 |
} |
| 74 |
|
| 75 |
static void layer_2(vec128 *bs, vec128 *cond) |
| 76 |
{ |
| 77 |
int x; |
| 78 |
vec128 diff; |
| 79 |
|
| 80 |
for (x = 0; x < (1 << 6); x += 8) |
| 81 |
{ |
| 82 |
diff = vec128_xor(bs[ x+0 ], bs[ x+4 ]); |
| 83 |
diff = vec128_and(diff, cond[0]); |
| 84 |
bs[ x+0 ] = vec128_xor(bs[ x+0 ], diff); |
| 85 |
bs[ x+4 ] = vec128_xor(bs[ x+4 ], diff); |
| 86 |
|
| 87 |
diff = vec128_xor(bs[ x+1 ], bs[ x+5 ]); |
| 88 |
diff = vec128_and(diff, cond[1]); |
| 89 |
bs[ x+1 ] = vec128_xor(bs[ x+1 ], diff); |
| 90 |
bs[ x+5 ] = vec128_xor(bs[ x+5 ], diff); |
| 91 |
|
| 92 |
diff = vec128_xor(bs[ x+2 ], bs[ x+6 ]); |
| 93 |
diff = vec128_and(diff, cond[2]); |
| 94 |
bs[ x+2 ] = vec128_xor(bs[ x+2 ], diff); |
| 95 |
bs[ x+6 ] = vec128_xor(bs[ x+6 ], diff); |
| 96 |
|
| 97 |
diff = vec128_xor(bs[ x+3 ], bs[ x+7 ]); |
| 98 |
diff = vec128_and(diff, cond[3]); |
| 99 |
bs[ x+3 ] = vec128_xor(bs[ x+3 ], diff); |
| 100 |
bs[ x+7 ] = vec128_xor(bs[ x+7 ], diff); |
| 101 |
|
| 102 |
cond += 4; |
| 103 |
} |
| 104 |
} |
| 105 |
|
| 106 |
static void layer_3(vec128 *bs, vec128 *cond) |
| 107 |
{ |
| 108 |
int x, s; |
| 109 |
vec128 diff; |
| 110 |
|
| 111 |
for (x = 0; x < (1 << 6); x += 16) |
| 112 |
for (s = x; s < x + 8; s += 4) |
| 113 |
{ |
| 114 |
diff = vec128_xor(bs[ s+0 ], bs[ s+8 ]); |
| 115 |
diff = vec128_and(diff, cond[0]); |
| 116 |
bs[ s+0 ] = vec128_xor(bs[ s+0 ], diff); |
| 117 |
bs[ s+8 ] = vec128_xor(bs[ s+8 ], diff); |
| 118 |
|
| 119 |
diff = vec128_xor(bs[ s+1 ], bs[ s+9 ]); |
| 120 |
diff = vec128_and(diff, cond[1]); |
| 121 |
bs[ s+1 ] = vec128_xor(bs[ s+1 ], diff); |
| 122 |
bs[ s+9 ] = vec128_xor(bs[ s+9 ], diff); |
| 123 |
|
| 124 |
diff = vec128_xor(bs[ s+2 ], bs[ s+10 ]); |
| 125 |
diff = vec128_and(diff, cond[2]); |
| 126 |
bs[ s+2 ] = vec128_xor(bs[ s+2 ], diff); |
| 127 |
bs[ s+10 ] = vec128_xor(bs[ s+10 ], diff); |
| 128 |
|
| 129 |
diff = vec128_xor(bs[ s+3 ], bs[ s+11 ]); |
| 130 |
diff = vec128_and(diff, cond[3]); |
| 131 |
bs[ s+3 ] = vec128_xor(bs[ s+3 ], diff); |
| 132 |
bs[ s+11 ] = vec128_xor(bs[ s+11 ], diff); |
| 133 |
|
| 134 |
cond += 4; |
| 135 |
} |
| 136 |
} |
| 137 |
|
| 138 |
static void layer_4(vec128 *bs, vec128 *cond) |
| 139 |
{ |
| 140 |
int x, s; |
| 141 |
vec128 diff; |
| 142 |
|
| 143 |
for (x = 0; x < (1 << 6); x += 32) |
| 144 |
for (s = x; s < x + 16; s += 4) |
| 145 |
{ |
| 146 |
diff = vec128_xor(bs[ s+0 ], bs[ s+16 ]); |
| 147 |
diff = vec128_and(diff, cond[0]); |
| 148 |
bs[ s+0 ] = vec128_xor(bs[ s+0 ], diff); |
| 149 |
bs[ s+16 ] = vec128_xor(bs[ s+16 ], diff); |
| 150 |
|
| 151 |
diff = vec128_xor(bs[ s+1 ], bs[ s+17 ]); |
| 152 |
diff = vec128_and(diff, cond[1]); |
| 153 |
bs[ s+1 ] = vec128_xor(bs[ s+1 ], diff); |
| 154 |
bs[ s+17 ] = vec128_xor(bs[ s+17 ], diff); |
| 155 |
|
| 156 |
diff = vec128_xor(bs[ s+2 ], bs[ s+18 ]); |
| 157 |
diff = vec128_and(diff, cond[2]); |
| 158 |
bs[ s+2 ] = vec128_xor(bs[ s+2 ], diff); |
| 159 |
bs[ s+18 ] = vec128_xor(bs[ s+18 ], diff); |
| 160 |
|
| 161 |
diff = vec128_xor(bs[ s+3 ], bs[ s+19 ]); |
| 162 |
diff = vec128_and(diff, cond[3]); |
| 163 |
bs[ s+3 ] = vec128_xor(bs[ s+3 ], diff); |
| 164 |
bs[ s+19 ] = vec128_xor(bs[ s+19 ], diff); |
| 165 |
|
| 166 |
cond += 4; |
| 167 |
} |
| 168 |
} |
| 169 |
|
| 170 |
static void layer_5(vec128 *bs, vec128 *cond) |
| 171 |
{ |
| 172 |
int x, s; |
| 173 |
vec128 diff; |
| 174 |
|
| 175 |
for (x = 0; x < (1 << 6); x += 64) |
| 176 |
for (s = x; s < x + 32; s += 4) |
| 177 |
{ |
| 178 |
diff = vec128_xor(bs[ s+0 ], bs[ s+32 ]); |
| 179 |
diff = vec128_and(diff, cond[0]); |
| 180 |
bs[ s+0 ] = vec128_xor(bs[ s+0 ], diff); |
| 181 |
bs[ s+32 ] = vec128_xor(bs[ s+32 ], diff); |
| 182 |
|
| 183 |
diff = vec128_xor(bs[ s+1 ], bs[ s+33 ]); |
| 184 |
diff = vec128_and(diff, cond[1]); |
| 185 |
bs[ s+1 ] = vec128_xor(bs[ s+1 ], diff); |
| 186 |
bs[ s+33 ] = vec128_xor(bs[ s+33 ], diff); |
| 187 |
|
| 188 |
diff = vec128_xor(bs[ s+2 ], bs[ s+34 ]); |
| 189 |
diff = vec128_and(diff, cond[2]); |
| 190 |
bs[ s+2 ] = vec128_xor(bs[ s+2 ], diff); |
| 191 |
bs[ s+34 ] = vec128_xor(bs[ s+34 ], diff); |
| 192 |
|
| 193 |
diff = vec128_xor(bs[ s+3 ], bs[ s+35 ]); |
| 194 |
diff = vec128_and(diff, cond[3]); |
| 195 |
bs[ s+3 ] = vec128_xor(bs[ s+3 ], diff); |
| 196 |
bs[ s+35 ] = vec128_xor(bs[ s+35 ], diff); |
| 197 |
|
| 198 |
cond += 4; |
| 199 |
} |
| 200 |
} |
| 201 |
|
| 202 |
/* input: bits, control bits as array of bytes */ |
| 203 |
/* output: bits_int, control bits as array of 128-bit vectors */ |
| 204 |
void load_bits(vec128 bits_int[][32], const unsigned char * bits) |
| 205 |
{ |
| 206 |
int i, j; |
| 207 |
const unsigned char *ptr = bits; |
| 208 |
|
| 209 |
vec128 buf[64]; |
| 210 |
|
| 211 |
for (i = 0; i <= 5; i += 2) |
| 212 |
{ |
| 213 |
for (j = 0; j < 64; j++) |
| 214 |
{ |
| 215 |
buf[j] = vec128_set2x(load8(ptr), load8(ptr + 512)); |
| 216 |
ptr += 8; |
| 217 |
} |
| 218 |
|
| 219 |
transpose_64x128_sp( buf ); |
| 220 |
|
| 221 |
for (j = 0; j < 32; j++) |
| 222 |
{ |
| 223 |
bits_int[i+0][j] = vec128_unpack_low(buf[j], buf[j+32]); |
| 224 |
bits_int[i+1][j] = vec128_unpack_high(buf[j], buf[j+32]); |
| 225 |
} |
| 226 |
|
| 227 |
ptr += 512; |
| 228 |
} |
| 229 |
|
| 230 |
for (i = 6; i <= 18; i++) |
| 231 |
for (j = 0; j < 32; j++) |
| 232 |
{ |
| 233 |
bits_int[i][j] = load16(ptr); ptr += 16; |
| 234 |
} |
| 235 |
|
| 236 |
for (i = 19; i < 25; i += 2) |
| 237 |
{ |
| 238 |
for (j = 0; j < 64; j++) |
| 239 |
{ |
| 240 |
buf[j] = vec128_set2x(load8(ptr), load8(ptr + 512)); |
| 241 |
ptr += 8; |
| 242 |
} |
| 243 |
|
| 244 |
transpose_64x128_sp( buf ); |
| 245 |
|
| 246 |
for (j = 0; j < 32; j++) |
| 247 |
{ |
| 248 |
bits_int[i+0][j] = vec128_unpack_low(buf[j], buf[j+32]); |
| 249 |
bits_int[i+1][j] = vec128_unpack_high(buf[j], buf[j+32]); |
| 250 |
} |
| 251 |
|
| 252 |
ptr += 512; |
| 253 |
} |
| 254 |
} |
| 255 |
|
| 256 |
/* input: r, sequence of bits to be permuted */ |
| 257 |
/* b, control bits as array of 128-bit vectors */ |
| 258 |
/* rev, 0 for normal application; !0 for inverse */ |
| 259 |
/* output: r, permuted bits */ |
| 260 |
void benes(vec128 * r, vec128 b[][32], int rev) |
| 261 |
{ |
| 262 |
int inc; |
| 263 |
|
| 264 |
vec128 *b_ptr = b[0]; |
| 265 |
|
| 266 |
if (rev == 0) { inc = 32; b_ptr = b[ 0]; } |
| 267 |
else { inc = -32; b_ptr = b[24]; } |
| 268 |
|
| 269 |
// |
| 270 |
|
| 271 |
transpose_64x128_sp( r ); |
| 272 |
|
| 273 |
layer_0(r, b_ptr); b_ptr += inc; |
| 274 |
layer_1(r, b_ptr); b_ptr += inc; |
| 275 |
layer_2(r, b_ptr); b_ptr += inc; |
| 276 |
layer_3(r, b_ptr); b_ptr += inc; |
| 277 |
layer_4(r, b_ptr); b_ptr += inc; |
| 278 |
layer_5(r, b_ptr); b_ptr += inc; |
| 279 |
|
| 280 |
transpose_64x128_sp( r ); |
| 281 |
|
| 282 |
layer_x(r, b_ptr); b_ptr += inc; |
| 283 |
layer_0(r, b_ptr); b_ptr += inc; |
| 284 |
layer_1(r, b_ptr); b_ptr += inc; |
| 285 |
layer_2(r, b_ptr); b_ptr += inc; |
| 286 |
layer_3(r, b_ptr); b_ptr += inc; |
| 287 |
layer_4(r, b_ptr); b_ptr += inc; |
| 288 |
layer_5(r, b_ptr); b_ptr += inc; |
| 289 |
layer_4(r, b_ptr); b_ptr += inc; |
| 290 |
layer_3(r, b_ptr); b_ptr += inc; |
| 291 |
layer_2(r, b_ptr); b_ptr += inc; |
| 292 |
layer_1(r, b_ptr); b_ptr += inc; |
| 293 |
layer_0(r, b_ptr); b_ptr += inc; |
| 294 |
layer_x(r, b_ptr); b_ptr += inc; |
| 295 |
|
| 296 |
transpose_64x128_sp( r ); |
| 297 |
|
| 298 |
layer_5(r, b_ptr); b_ptr += inc; |
| 299 |
layer_4(r, b_ptr); b_ptr += inc; |
| 300 |
layer_3(r, b_ptr); b_ptr += inc; |
| 301 |
layer_2(r, b_ptr); b_ptr += inc; |
| 302 |
layer_1(r, b_ptr); b_ptr += inc; |
| 303 |
layer_0(r, b_ptr); |
| 304 |
|
| 305 |
transpose_64x128_sp( r ); |
| 306 |
} |
| 307 |
|