| 1 |
/* |
| 2 |
* (C) 2025 Jack Lloyd |
| 3 |
* |
| 4 |
* Botan is released under the Simplified BSD License (see license.txt) |
| 5 |
*/ |
| 6 |
|
| 7 |
#include <botan/internal/sha1.h> |
| 8 |
|
| 9 |
#include <botan/internal/isa_extn.h> |
| 10 |
#include <botan/internal/sha1_f.h> |
| 11 |
#include <botan/internal/simd_avx2.h> |
| 12 |
#include <immintrin.h> |
| 13 |
|
| 14 |
namespace Botan { |
| 15 |
|
| 16 |
namespace { |
| 17 |
|
| 18 |
/* |
| 19 |
* This is exactly the same approach as used in sha1_simd.cpp, just done |
| 20 |
* twice in the two AVX2 "lanes" - remember that alignr and slli/srli |
| 21 |
* here are working not across the entire register but instead as if |
| 22 |
* there were two smaller vectors. |
| 23 |
*/ |
| 24 |
BOTAN_FN_ISA_AVX2_BMI2 BOTAN_FORCE_INLINE SIMD_8x32 sha1_avx2_next_w(SIMD_8x32& XW0, |
| 25 |
SIMD_8x32 XW1, |
| 26 |
SIMD_8x32 XW2, |
| 27 |
SIMD_8x32 XW3) { |
| 28 |
SIMD_8x32 T0 = XW0; // W[t-16..t-13] |
| 29 |
T0 ^= SIMD_8x32(_mm256_alignr_epi8(XW1.raw(), XW0.raw(), 8)); |
| 30 |
T0 ^= XW2; // W[t-8..t-5] |
| 31 |
T0 ^= SIMD_8x32(_mm256_srli_si256(XW3.raw(), 4)); // W[t-3..t-1] || 0 |
| 32 |
|
| 33 |
/* unrotated W[t]..W[t+2] in T0 ... still need W[t+3] */ |
| 34 |
|
| 35 |
// Extract w[t+0] into T2 |
| 36 |
auto T2 = SIMD_8x32(_mm256_slli_si256(T0.raw(), 3 * 4)); |
| 37 |
|
| 38 |
// Main rotation |
| 39 |
T0 = T0.rotl<1>(); |
| 40 |
|
| 41 |
// Rotation of W[t+3] has rot by 2 to account for us working on non-rotated words |
| 42 |
T2 = T2.rotl<2>(); |
| 43 |
|
| 44 |
// Merge rol(W[t+0], 1) into W[t+3] |
| 45 |
T0 ^= T2; |
| 46 |
|
| 47 |
XW0 = T0; |
| 48 |
return T0; |
| 49 |
} |
| 50 |
|
| 51 |
/* |
| 52 |
* Helper for word permutation with zeroing because AVX2 is awful |
| 53 |
* |
| 54 |
* Clang and GCC both compile this to a couple of stored constants plus |
| 55 |
* a vpermd/vpand pair. |
| 56 |
*/ |
| 57 |
template <int I0, int I1, int I2, int I3, int I4, int I5, int I6, int I7> |
| 58 |
BOTAN_FN_ISA_AVX2_BMI2 BOTAN_FORCE_INLINE SIMD_8x32 permute_words(SIMD_8x32 v) { |
| 59 |
const __m256i tbl = _mm256_setr_epi32(I0, I1, I2, I3, I4, I5, I6, I7); |
| 60 |
const __m256i mask = _mm256_setr_epi32(I0 >= 0 ? 0xFFFFFFFF : 0, |
| 61 |
I1 >= 0 ? 0xFFFFFFFF : 0, |
| 62 |
I2 >= 0 ? 0xFFFFFFFF : 0, |
| 63 |
I3 >= 0 ? 0xFFFFFFFF : 0, |
| 64 |
I4 >= 0 ? 0xFFFFFFFF : 0, |
| 65 |
I5 >= 0 ? 0xFFFFFFFF : 0, |
| 66 |
I6 >= 0 ? 0xFFFFFFFF : 0, |
| 67 |
I7 >= 0 ? 0xFFFFFFFF : 0); |
| 68 |
|
| 69 |
return SIMD_8x32(_mm256_and_si256(mask, _mm256_permutevar8x32_epi32(v.raw(), tbl))); |
| 70 |
} |
| 71 |
|
| 72 |
/* |
| 73 |
This is the same approach as the (single buffer) SHA-1 expansion in sha1_simd.cpp |
| 74 |
except unrolled further; instead of computing 4 words of W at once, we compute 8. |
| 75 |
|
| 76 |
However this is complicated both by the SHA-1 recurrence and AVX2 |
| 77 |
limitations; it is faster than what's done in sha1_simd.cpp but only just barely. |
| 78 |
|
| 79 |
The basic idea here is that when computing this (8x per message block): |
| 80 |
|
| 81 |
W[j + 0] = rotl<1>(W[j - 3] ^ W[j - 8] ^ W[j - 14] ^ W[j - 16]); |
| 82 |
W[j + 1] = rotl<1>(W[j - 2] ^ W[j - 7] ^ W[j - 13] ^ W[j - 15]); |
| 83 |
W[j + 2] = rotl<1>(W[j - 1] ^ W[j - 6] ^ W[j - 12] ^ W[j - 14]); |
| 84 |
W[j + 3] = rotl<1>(W[j ] ^ W[j - 5] ^ W[j - 11] ^ W[j - 13]); |
| 85 |
W[j + 4] = rotl<1>(W[j + 1] ^ W[j - 4] ^ W[j - 10] ^ W[j - 12]); |
| 86 |
W[j + 5] = rotl<1>(W[j + 2] ^ W[j - 3] ^ W[j - 9] ^ W[j - 11]); |
| 87 |
W[j + 6] = rotl<1>(W[j + 3] ^ W[j - 2] ^ W[j - 8] ^ W[j - 10]); |
| 88 |
W[j + 7] = rotl<1>(W[j + 4] ^ W[j - 1] ^ W[j - 7] ^ W[j - 9]); |
| 89 |
|
| 90 |
We instead compute a partial expansion: |
| 91 |
|
| 92 |
W[j + 0] = rotl<1>(W[j - 3] ^ W[j - 8] ^ W[j - 14] ^ W[j - 16]); |
| 93 |
W[j + 1] = rotl<1>(W[j - 2] ^ W[j - 7] ^ W[j - 13] ^ W[j - 15]); |
| 94 |
W[j + 2] = rotl<1>(W[j - 1] ^ W[j - 6] ^ W[j - 12] ^ W[j - 14]); |
| 95 |
W[j + 3] = rotl<1>( W[j - 5] ^ W[j - 11] ^ W[j - 13]); |
| 96 |
W[j + 4] = rotl<1>( W[j - 4] ^ W[j - 10] ^ W[j - 12]); |
| 97 |
W[j + 5] = rotl<1>( W[j - 3] ^ W[j - 9] ^ W[j - 11]); |
| 98 |
W[j + 6] = rotl<1>( W[j - 2] ^ W[j - 8] ^ W[j - 10]); |
| 99 |
W[j + 7] = rotl<1>( W[j - 1] ^ W[j - 7] ^ W[j - 9]); |
| 100 |
|
| 101 |
Then update it with values that were not available until the first expansion is |
| 102 |
completed: |
| 103 |
|
| 104 |
W[j + 3] ^= rotl<1>(W[j ]); |
| 105 |
W[j + 4] ^= rotl<1>(W[j + 1]); |
| 106 |
W[j + 5] ^= rotl<1>(W[j + 2]); |
| 107 |
|
| 108 |
And then update again with values not available until the second expansion step |
| 109 |
is completed: |
| 110 |
|
| 111 |
W[j + 6] ^= rotl<1>(W[j + 3]); |
| 112 |
W[j + 7] ^= rotl<1>(W[j + 4]); |
| 113 |
*/ |
| 114 |
|
| 115 |
BOTAN_FN_ISA_AVX2_BMI2 BOTAN_FORCE_INLINE SIMD_8x32 sha1_avx2_next_w2(SIMD_8x32& W0, SIMD_8x32 W2) { |
| 116 |
// W[j-16..j-9] ^ W[j-8...j-1] |
| 117 |
auto WN = W0 ^ W2; |
| 118 |
|
| 119 |
// XOR in W[j-3..j-1] || 0 || 0 || 0 || W[j-8...j-7] |
| 120 |
WN ^= permute_words<5, 6, 7, -1, -1, -1, 0, 1>(W2); |
| 121 |
|
| 122 |
// XOR in W[j-14...j-9] || 0 || 0 |
| 123 |
WN ^= permute_words<2, 3, 4, 5, 6, 7, -1, -1>(W0); |
| 124 |
|
| 125 |
// Extract W[j...j+2], rotate, and XOR into W[j+3...j+5] |
| 126 |
auto T0 = permute_words<-1, -1, -1, 0, 1, 2, -1, -1>(WN).rotl<2>(); |
| 127 |
WN = WN.rotl<1>(); // main block rotation |
| 128 |
|
| 129 |
WN ^= T0; |
| 130 |
|
| 131 |
// Extract W[j+3...j+4], rotate, and XOR into W[j+6...j+7] |
| 132 |
WN ^= permute_words<-1, -1, -1, -1, -1, -1, 3, 4>(WN).rotl<1>(); |
| 133 |
|
| 134 |
W0 = WN; |
| 135 |
return WN; |
| 136 |
} |
| 137 |
|
| 138 |
} // namespace |
| 139 |
|
| 140 |
/* |
| 141 |
* SHA-1 Compression Function using SIMD for message expansion |
| 142 |
*/ |
| 143 |
//static |
| 144 |
void BOTAN_FN_ISA_AVX2_BMI2 SHA_1::avx2_compress_n(digest_type& digest, std::span<const uint8_t> input, size_t blocks) { |
| 145 |
using namespace SHA1_F; |
| 146 |
|
| 147 |
const SIMD_8x32 K11 = SIMD_8x32::splat(K1); |
| 148 |
const SIMD_8x32 K22 = SIMD_8x32::splat(K2); |
| 149 |
const SIMD_8x32 K33 = SIMD_8x32::splat(K3); |
| 150 |
const SIMD_8x32 K44 = SIMD_8x32::splat(K4); |
| 151 |
|
| 152 |
const SIMD_8x32 K12(K1, K1, K1, K1, K2, K2, K2, K2); |
| 153 |
const SIMD_8x32 K34(K3, K3, K3, K3, K4, K4, K4, K4); |
| 154 |
|
| 155 |
uint32_t A = digest[0]; |
| 156 |
uint32_t B = digest[1]; |
| 157 |
uint32_t C = digest[2]; |
| 158 |
uint32_t D = digest[3]; |
| 159 |
uint32_t E = digest[4]; |
| 160 |
|
| 161 |
BufferSlicer in(input); |
| 162 |
|
| 163 |
while(blocks >= 2) { |
| 164 |
const auto block = in.take(2 * block_bytes); |
| 165 |
blocks -= 2; |
| 166 |
|
| 167 |
uint32_t W2[80] = {0}; |
| 168 |
|
| 169 |
uint32_t PT[4]; |
| 170 |
|
| 171 |
// NOLINTNEXTLINE(*-container-data-pointer) |
| 172 |
SIMD_8x32 XW0 = SIMD_8x32::load_be128(&block[0], &block[64]); |
| 173 |
SIMD_8x32 XW1 = SIMD_8x32::load_be128(&block[16], &block[80]); |
| 174 |
SIMD_8x32 XW2 = SIMD_8x32::load_be128(&block[32], &block[96]); |
| 175 |
SIMD_8x32 XW3 = SIMD_8x32::load_be128(&block[48], &block[112]); |
| 176 |
|
| 177 |
SIMD_8x32 P0 = XW0 + SIMD_8x32::splat(K1); |
| 178 |
SIMD_8x32 P1 = XW1 + SIMD_8x32::splat(K1); |
| 179 |
SIMD_8x32 P2 = XW2 + SIMD_8x32::splat(K1); |
| 180 |
SIMD_8x32 P3 = XW3 + SIMD_8x32::splat(K1); |
| 181 |
|
| 182 |
// NOLINTBEGIN(readability-suspicious-call-argument) XW rotation |
| 183 |
|
| 184 |
P0.store_le128(PT, &W2[0]); |
| 185 |
P0 = sha1_avx2_next_w(XW0, XW1, XW2, XW3) + SIMD_8x32::splat(K1); |
| 186 |
F1(A, B, C, D, E, PT[0]); |
| 187 |
F1(E, A, B, C, D, PT[1]); |
| 188 |
F1(D, E, A, B, C, PT[2]); |
| 189 |
F1(C, D, E, A, B, PT[3]); |
| 190 |
|
| 191 |
P1.store_le128(PT, &W2[4]); |
| 192 |
P1 = sha1_avx2_next_w(XW1, XW2, XW3, XW0) + SIMD_8x32::splat(K2); |
| 193 |
F1(B, C, D, E, A, PT[0]); |
| 194 |
F1(A, B, C, D, E, PT[1]); |
| 195 |
F1(E, A, B, C, D, PT[2]); |
| 196 |
F1(D, E, A, B, C, PT[3]); |
| 197 |
|
| 198 |
P2.store_le128(PT, &W2[8]); |
| 199 |
P2 = sha1_avx2_next_w(XW2, XW3, XW0, XW1) + SIMD_8x32::splat(K2); |
| 200 |
F1(C, D, E, A, B, PT[0]); |
| 201 |
F1(B, C, D, E, A, PT[1]); |
| 202 |
F1(A, B, C, D, E, PT[2]); |
| 203 |
F1(E, A, B, C, D, PT[3]); |
| 204 |
|
| 205 |
P3.store_le128(PT, &W2[12]); |
| 206 |
P3 = sha1_avx2_next_w(XW3, XW0, XW1, XW2) + SIMD_8x32::splat(K2); |
| 207 |
F1(D, E, A, B, C, PT[0]); |
| 208 |
F1(C, D, E, A, B, PT[1]); |
| 209 |
F1(B, C, D, E, A, PT[2]); |
| 210 |
F1(A, B, C, D, E, PT[3]); |
| 211 |
|
| 212 |
P0.store_le128(PT, &W2[16]); |
| 213 |
P0 = sha1_avx2_next_w(XW0, XW1, XW2, XW3) + SIMD_8x32::splat(K2); |
| 214 |
F1(E, A, B, C, D, PT[0]); |
| 215 |
F1(D, E, A, B, C, PT[1]); |
| 216 |
F1(C, D, E, A, B, PT[2]); |
| 217 |
F1(B, C, D, E, A, PT[3]); |
| 218 |
|
| 219 |
P1.store_le128(PT, &W2[20]); |
| 220 |
P1 = sha1_avx2_next_w(XW1, XW2, XW3, XW0) + SIMD_8x32::splat(K2); |
| 221 |
F2(A, B, C, D, E, PT[0]); |
| 222 |
F2(E, A, B, C, D, PT[1]); |
| 223 |
F2(D, E, A, B, C, PT[2]); |
| 224 |
F2(C, D, E, A, B, PT[3]); |
| 225 |
|
| 226 |
P2.store_le128(PT, &W2[24]); |
| 227 |
P2 = sha1_avx2_next_w(XW2, XW3, XW0, XW1) + SIMD_8x32::splat(K3); |
| 228 |
F2(B, C, D, E, A, PT[0]); |
| 229 |
F2(A, B, C, D, E, PT[1]); |
| 230 |
F2(E, A, B, C, D, PT[2]); |
| 231 |
F2(D, E, A, B, C, PT[3]); |
| 232 |
|
| 233 |
P3.store_le128(PT, &W2[28]); |
| 234 |
P3 = sha1_avx2_next_w(XW3, XW0, XW1, XW2) + SIMD_8x32::splat(K3); |
| 235 |
F2(C, D, E, A, B, PT[0]); |
| 236 |
F2(B, C, D, E, A, PT[1]); |
| 237 |
F2(A, B, C, D, E, PT[2]); |
| 238 |
F2(E, A, B, C, D, PT[3]); |
| 239 |
|
| 240 |
P0.store_le128(PT, &W2[32]); |
| 241 |
P0 = sha1_avx2_next_w(XW0, XW1, XW2, XW3) + SIMD_8x32::splat(K3); |
| 242 |
F2(D, E, A, B, C, PT[0]); |
| 243 |
F2(C, D, E, A, B, PT[1]); |
| 244 |
F2(B, C, D, E, A, PT[2]); |
| 245 |
F2(A, B, C, D, E, PT[3]); |
| 246 |
|
| 247 |
P1.store_le128(PT, &W2[36]); |
| 248 |
P1 = sha1_avx2_next_w(XW1, XW2, XW3, XW0) + SIMD_8x32::splat(K3); |
| 249 |
F2(E, A, B, C, D, PT[0]); |
| 250 |
F2(D, E, A, B, C, PT[1]); |
| 251 |
F2(C, D, E, A, B, PT[2]); |
| 252 |
F2(B, C, D, E, A, PT[3]); |
| 253 |
|
| 254 |
P2.store_le128(PT, &W2[40]); |
| 255 |
P2 = sha1_avx2_next_w(XW2, XW3, XW0, XW1) + SIMD_8x32::splat(K3); |
| 256 |
F3(A, B, C, D, E, PT[0]); |
| 257 |
F3(E, A, B, C, D, PT[1]); |
| 258 |
F3(D, E, A, B, C, PT[2]); |
| 259 |
F3(C, D, E, A, B, PT[3]); |
| 260 |
|
| 261 |
P3.store_le128(PT, &W2[44]); |
| 262 |
P3 = sha1_avx2_next_w(XW3, XW0, XW1, XW2) + SIMD_8x32::splat(K4); |
| 263 |
F3(B, C, D, E, A, PT[0]); |
| 264 |
F3(A, B, C, D, E, PT[1]); |
| 265 |
F3(E, A, B, C, D, PT[2]); |
| 266 |
F3(D, E, A, B, C, PT[3]); |
| 267 |
|
| 268 |
P0.store_le128(PT, &W2[48]); |
| 269 |
P0 = sha1_avx2_next_w(XW0, XW1, XW2, XW3) + SIMD_8x32::splat(K4); |
| 270 |
F3(C, D, E, A, B, PT[0]); |
| 271 |
F3(B, C, D, E, A, PT[1]); |
| 272 |
F3(A, B, C, D, E, PT[2]); |
| 273 |
F3(E, A, B, C, D, PT[3]); |
| 274 |
|
| 275 |
P1.store_le128(PT, &W2[52]); |
| 276 |
P1 = sha1_avx2_next_w(XW1, XW2, XW3, XW0) + SIMD_8x32::splat(K4); |
| 277 |
F3(D, E, A, B, C, PT[0]); |
| 278 |
F3(C, D, E, A, B, PT[1]); |
| 279 |
F3(B, C, D, E, A, PT[2]); |
| 280 |
F3(A, B, C, D, E, PT[3]); |
| 281 |
|
| 282 |
P2.store_le128(PT, &W2[56]); |
| 283 |
P2 = sha1_avx2_next_w(XW2, XW3, XW0, XW1) + SIMD_8x32::splat(K4); |
| 284 |
F3(E, A, B, C, D, PT[0]); |
| 285 |
F3(D, E, A, B, C, PT[1]); |
| 286 |
F3(C, D, E, A, B, PT[2]); |
| 287 |
F3(B, C, D, E, A, PT[3]); |
| 288 |
|
| 289 |
P3.store_le128(PT, &W2[60]); |
| 290 |
P3 = sha1_avx2_next_w(XW3, XW0, XW1, XW2) + SIMD_8x32::splat(K4); |
| 291 |
F4(A, B, C, D, E, PT[0]); |
| 292 |
F4(E, A, B, C, D, PT[1]); |
| 293 |
F4(D, E, A, B, C, PT[2]); |
| 294 |
F4(C, D, E, A, B, PT[3]); |
| 295 |
|
| 296 |
P0.store_le128(PT, &W2[64]); |
| 297 |
F4(B, C, D, E, A, PT[0]); |
| 298 |
F4(A, B, C, D, E, PT[1]); |
| 299 |
F4(E, A, B, C, D, PT[2]); |
| 300 |
F4(D, E, A, B, C, PT[3]); |
| 301 |
|
| 302 |
P1.store_le128(PT, &W2[68]); |
| 303 |
F4(C, D, E, A, B, PT[0]); |
| 304 |
F4(B, C, D, E, A, PT[1]); |
| 305 |
F4(A, B, C, D, E, PT[2]); |
| 306 |
F4(E, A, B, C, D, PT[3]); |
| 307 |
|
| 308 |
P2.store_le128(PT, &W2[72]); |
| 309 |
F4(D, E, A, B, C, PT[0]); |
| 310 |
F4(C, D, E, A, B, PT[1]); |
| 311 |
F4(B, C, D, E, A, PT[2]); |
| 312 |
F4(A, B, C, D, E, PT[3]); |
| 313 |
|
| 314 |
P3.store_le128(PT, &W2[76]); |
| 315 |
F4(E, A, B, C, D, PT[0]); |
| 316 |
F4(D, E, A, B, C, PT[1]); |
| 317 |
F4(C, D, E, A, B, PT[2]); |
| 318 |
F4(B, C, D, E, A, PT[3]); |
| 319 |
|
| 320 |
// NOLINTEND(readability-suspicious-call-argument) |
| 321 |
|
| 322 |
A = (digest[0] += A); |
| 323 |
B = (digest[1] += B); |
| 324 |
C = (digest[2] += C); |
| 325 |
D = (digest[3] += D); |
| 326 |
E = (digest[4] += E); |
| 327 |
|
| 328 |
// Second block with pre-expanded message |
| 329 |
F1(A, B, C, D, E, W2[0]); |
| 330 |
F1(E, A, B, C, D, W2[1]); |
| 331 |
F1(D, E, A, B, C, W2[2]); |
| 332 |
F1(C, D, E, A, B, W2[3]); |
| 333 |
F1(B, C, D, E, A, W2[4]); |
| 334 |
F1(A, B, C, D, E, W2[5]); |
| 335 |
F1(E, A, B, C, D, W2[6]); |
| 336 |
F1(D, E, A, B, C, W2[7]); |
| 337 |
F1(C, D, E, A, B, W2[8]); |
| 338 |
F1(B, C, D, E, A, W2[9]); |
| 339 |
F1(A, B, C, D, E, W2[10]); |
| 340 |
F1(E, A, B, C, D, W2[11]); |
| 341 |
F1(D, E, A, B, C, W2[12]); |
| 342 |
F1(C, D, E, A, B, W2[13]); |
| 343 |
F1(B, C, D, E, A, W2[14]); |
| 344 |
F1(A, B, C, D, E, W2[15]); |
| 345 |
F1(E, A, B, C, D, W2[16]); |
| 346 |
F1(D, E, A, B, C, W2[17]); |
| 347 |
F1(C, D, E, A, B, W2[18]); |
| 348 |
F1(B, C, D, E, A, W2[19]); |
| 349 |
F2(A, B, C, D, E, W2[20]); |
| 350 |
F2(E, A, B, C, D, W2[21]); |
| 351 |
F2(D, E, A, B, C, W2[22]); |
| 352 |
F2(C, D, E, A, B, W2[23]); |
| 353 |
F2(B, C, D, E, A, W2[24]); |
| 354 |
F2(A, B, C, D, E, W2[25]); |
| 355 |
F2(E, A, B, C, D, W2[26]); |
| 356 |
F2(D, E, A, B, C, W2[27]); |
| 357 |
F2(C, D, E, A, B, W2[28]); |
| 358 |
F2(B, C, D, E, A, W2[29]); |
| 359 |
F2(A, B, C, D, E, W2[30]); |
| 360 |
F2(E, A, B, C, D, W2[31]); |
| 361 |
F2(D, E, A, B, C, W2[32]); |
| 362 |
F2(C, D, E, A, B, W2[33]); |
| 363 |
F2(B, C, D, E, A, W2[34]); |
| 364 |
F2(A, B, C, D, E, W2[35]); |
| 365 |
F2(E, A, B, C, D, W2[36]); |
| 366 |
F2(D, E, A, B, C, W2[37]); |
| 367 |
F2(C, D, E, A, B, W2[38]); |
| 368 |
F2(B, C, D, E, A, W2[39]); |
| 369 |
F3(A, B, C, D, E, W2[40]); |
| 370 |
F3(E, A, B, C, D, W2[41]); |
| 371 |
F3(D, E, A, B, C, W2[42]); |
| 372 |
F3(C, D, E, A, B, W2[43]); |
| 373 |
F3(B, C, D, E, A, W2[44]); |
| 374 |
F3(A, B, C, D, E, W2[45]); |
| 375 |
F3(E, A, B, C, D, W2[46]); |
| 376 |
F3(D, E, A, B, C, W2[47]); |
| 377 |
F3(C, D, E, A, B, W2[48]); |
| 378 |
F3(B, C, D, E, A, W2[49]); |
| 379 |
F3(A, B, C, D, E, W2[50]); |
| 380 |
F3(E, A, B, C, D, W2[51]); |
| 381 |
F3(D, E, A, B, C, W2[52]); |
| 382 |
F3(C, D, E, A, B, W2[53]); |
| 383 |
F3(B, C, D, E, A, W2[54]); |
| 384 |
F3(A, B, C, D, E, W2[55]); |
| 385 |
F3(E, A, B, C, D, W2[56]); |
| 386 |
F3(D, E, A, B, C, W2[57]); |
| 387 |
F3(C, D, E, A, B, W2[58]); |
| 388 |
F3(B, C, D, E, A, W2[59]); |
| 389 |
F4(A, B, C, D, E, W2[60]); |
| 390 |
F4(E, A, B, C, D, W2[61]); |
| 391 |
F4(D, E, A, B, C, W2[62]); |
| 392 |
F4(C, D, E, A, B, W2[63]); |
| 393 |
F4(B, C, D, E, A, W2[64]); |
| 394 |
F4(A, B, C, D, E, W2[65]); |
| 395 |
F4(E, A, B, C, D, W2[66]); |
| 396 |
F4(D, E, A, B, C, W2[67]); |
| 397 |
F4(C, D, E, A, B, W2[68]); |
| 398 |
F4(B, C, D, E, A, W2[69]); |
| 399 |
F4(A, B, C, D, E, W2[70]); |
| 400 |
F4(E, A, B, C, D, W2[71]); |
| 401 |
F4(D, E, A, B, C, W2[72]); |
| 402 |
F4(C, D, E, A, B, W2[73]); |
| 403 |
F4(B, C, D, E, A, W2[74]); |
| 404 |
F4(A, B, C, D, E, W2[75]); |
| 405 |
F4(E, A, B, C, D, W2[76]); |
| 406 |
F4(D, E, A, B, C, W2[77]); |
| 407 |
F4(C, D, E, A, B, W2[78]); |
| 408 |
F4(B, C, D, E, A, W2[79]); |
| 409 |
|
| 410 |
A = (digest[0] += A); |
| 411 |
B = (digest[1] += B); |
| 412 |
C = (digest[2] += C); |
| 413 |
D = (digest[3] += D); |
| 414 |
E = (digest[4] += E); |
| 415 |
} |
| 416 |
|
| 417 |
for(size_t i = 0; i != blocks; ++i) { |
| 418 |
uint32_t PT[8]; |
| 419 |
|
| 420 |
const auto block = in.take(block_bytes); |
| 421 |
|
| 422 |
SIMD_8x32 W0 = SIMD_8x32::load_be(&block[0]); // NOLINT(*-container-data-pointer) |
| 423 |
SIMD_8x32 W2 = SIMD_8x32::load_be(&block[32]); |
| 424 |
|
| 425 |
SIMD_8x32 P0 = W0 + K11; |
| 426 |
SIMD_8x32 P2 = W2 + K11; |
| 427 |
|
| 428 |
P0.store_le(PT); |
| 429 |
P0 = sha1_avx2_next_w2(W0, W2) + K12; |
| 430 |
|
| 431 |
F1(A, B, C, D, E, PT[0]); |
| 432 |
F1(E, A, B, C, D, PT[1]); |
| 433 |
F1(D, E, A, B, C, PT[2]); |
| 434 |
F1(C, D, E, A, B, PT[3]); |
| 435 |
F1(B, C, D, E, A, PT[4]); |
| 436 |
F1(A, B, C, D, E, PT[5]); |
| 437 |
F1(E, A, B, C, D, PT[6]); |
| 438 |
F1(D, E, A, B, C, PT[7]); |
| 439 |
|
| 440 |
P2.store_le(PT); |
| 441 |
P2 = sha1_avx2_next_w2(W2, W0) + K22; |
| 442 |
|
| 443 |
F1(C, D, E, A, B, PT[0]); |
| 444 |
F1(B, C, D, E, A, PT[1]); |
| 445 |
F1(A, B, C, D, E, PT[2]); |
| 446 |
F1(E, A, B, C, D, PT[3]); |
| 447 |
F1(D, E, A, B, C, PT[4]); |
| 448 |
F1(C, D, E, A, B, PT[5]); |
| 449 |
F1(B, C, D, E, A, PT[6]); |
| 450 |
F1(A, B, C, D, E, PT[7]); |
| 451 |
|
| 452 |
P0.store_le(PT); |
| 453 |
P0 = sha1_avx2_next_w2(W0, W2) + K22; |
| 454 |
|
| 455 |
F1(E, A, B, C, D, PT[0]); |
| 456 |
F1(D, E, A, B, C, PT[1]); |
| 457 |
F1(C, D, E, A, B, PT[2]); |
| 458 |
F1(B, C, D, E, A, PT[3]); |
| 459 |
F2(A, B, C, D, E, PT[4]); |
| 460 |
F2(E, A, B, C, D, PT[5]); |
| 461 |
F2(D, E, A, B, C, PT[6]); |
| 462 |
F2(C, D, E, A, B, PT[7]); |
| 463 |
|
| 464 |
P2.store_le(PT); |
| 465 |
P2 = sha1_avx2_next_w2(W2, W0) + K33; |
| 466 |
|
| 467 |
F2(B, C, D, E, A, PT[0]); |
| 468 |
F2(A, B, C, D, E, PT[1]); |
| 469 |
F2(E, A, B, C, D, PT[2]); |
| 470 |
F2(D, E, A, B, C, PT[3]); |
| 471 |
F2(C, D, E, A, B, PT[4]); |
| 472 |
F2(B, C, D, E, A, PT[5]); |
| 473 |
F2(A, B, C, D, E, PT[6]); |
| 474 |
F2(E, A, B, C, D, PT[7]); |
| 475 |
|
| 476 |
P0.store_le(PT); |
| 477 |
P0 = sha1_avx2_next_w2(W0, W2) + K33; |
| 478 |
|
| 479 |
F2(D, E, A, B, C, PT[0]); |
| 480 |
F2(C, D, E, A, B, PT[1]); |
| 481 |
F2(B, C, D, E, A, PT[2]); |
| 482 |
F2(A, B, C, D, E, PT[3]); |
| 483 |
F2(E, A, B, C, D, PT[4]); |
| 484 |
F2(D, E, A, B, C, PT[5]); |
| 485 |
F2(C, D, E, A, B, PT[6]); |
| 486 |
F2(B, C, D, E, A, PT[7]); |
| 487 |
|
| 488 |
P2.store_le(PT); |
| 489 |
P2 = sha1_avx2_next_w2(W2, W0) + K34; |
| 490 |
|
| 491 |
F3(A, B, C, D, E, PT[0]); |
| 492 |
F3(E, A, B, C, D, PT[1]); |
| 493 |
F3(D, E, A, B, C, PT[2]); |
| 494 |
F3(C, D, E, A, B, PT[3]); |
| 495 |
F3(B, C, D, E, A, PT[4]); |
| 496 |
F3(A, B, C, D, E, PT[5]); |
| 497 |
F3(E, A, B, C, D, PT[6]); |
| 498 |
F3(D, E, A, B, C, PT[7]); |
| 499 |
|
| 500 |
P0.store_le(PT); |
| 501 |
P0 = sha1_avx2_next_w2(W0, W2) + K44; |
| 502 |
|
| 503 |
F3(C, D, E, A, B, PT[0]); |
| 504 |
F3(B, C, D, E, A, PT[1]); |
| 505 |
F3(A, B, C, D, E, PT[2]); |
| 506 |
F3(E, A, B, C, D, PT[3]); |
| 507 |
F3(D, E, A, B, C, PT[4]); |
| 508 |
F3(C, D, E, A, B, PT[5]); |
| 509 |
F3(B, C, D, E, A, PT[6]); |
| 510 |
F3(A, B, C, D, E, PT[7]); |
| 511 |
|
| 512 |
P2.store_le(PT); |
| 513 |
P2 = sha1_avx2_next_w2(W2, W0) + K44; |
| 514 |
|
| 515 |
F3(E, A, B, C, D, PT[0]); |
| 516 |
F3(D, E, A, B, C, PT[1]); |
| 517 |
F3(C, D, E, A, B, PT[2]); |
| 518 |
F3(B, C, D, E, A, PT[3]); |
| 519 |
F4(A, B, C, D, E, PT[4]); |
| 520 |
F4(E, A, B, C, D, PT[5]); |
| 521 |
F4(D, E, A, B, C, PT[6]); |
| 522 |
F4(C, D, E, A, B, PT[7]); |
| 523 |
|
| 524 |
P0.store_le(PT); |
| 525 |
|
| 526 |
F4(B, C, D, E, A, PT[0]); |
| 527 |
F4(A, B, C, D, E, PT[1]); |
| 528 |
F4(E, A, B, C, D, PT[2]); |
| 529 |
F4(D, E, A, B, C, PT[3]); |
| 530 |
F4(C, D, E, A, B, PT[4]); |
| 531 |
F4(B, C, D, E, A, PT[5]); |
| 532 |
F4(A, B, C, D, E, PT[6]); |
| 533 |
F4(E, A, B, C, D, PT[7]); |
| 534 |
|
| 535 |
P2.store_le(PT); |
| 536 |
|
| 537 |
F4(D, E, A, B, C, PT[0]); |
| 538 |
F4(C, D, E, A, B, PT[1]); |
| 539 |
F4(B, C, D, E, A, PT[2]); |
| 540 |
F4(A, B, C, D, E, PT[3]); |
| 541 |
F4(E, A, B, C, D, PT[4]); |
| 542 |
F4(D, E, A, B, C, PT[5]); |
| 543 |
F4(C, D, E, A, B, PT[6]); |
| 544 |
F4(B, C, D, E, A, PT[7]); |
| 545 |
|
| 546 |
A = (digest[0] += A); |
| 547 |
B = (digest[1] += B); |
| 548 |
C = (digest[2] += C); |
| 549 |
D = (digest[3] += D); |
| 550 |
E = (digest[4] += E); |
| 551 |
} |
| 552 |
} |
| 553 |
|
| 554 |
} // namespace Botan |