| 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/shacal2.h> |
| 8 |
|
| 9 |
#include <botan/internal/simd_avx2.h> |
| 10 |
#include <botan/internal/simd_avx512.h> |
| 11 |
|
| 12 |
namespace Botan { |
| 13 |
|
| 14 |
namespace SHACAL2_AVX512_F { |
| 15 |
|
| 16 |
namespace { |
| 17 |
|
| 18 |
// NOLINTBEGIN(portability-simd-intrinsics) |
| 19 |
|
| 20 |
/* |
| 21 |
* 8x16 Transpose |
| 22 |
* |
| 23 |
* Convert from |
| 24 |
* |
| 25 |
* A00 B00 C00 ... H00 |
| 26 |
* A01 B01 C01 ... H01 |
| 27 |
* .. |
| 28 |
* A15 B15 C15 ... H15 |
| 29 |
* |
| 30 |
* with two blocks stored in each register, into |
| 31 |
* |
| 32 |
* A00 A01 ... A15 |
| 33 |
* B00 B01 ... B15 |
| 34 |
* ... |
| 35 |
* H00 H01 ... H15 |
| 36 |
*/ |
| 37 |
BOTAN_FN_ISA_AVX512 |
| 38 |
void transpose_in(SIMD_16x32& B0, |
| 39 |
SIMD_16x32& B1, |
| 40 |
SIMD_16x32& B2, |
| 41 |
SIMD_16x32& B3, |
| 42 |
SIMD_16x32& B4, |
| 43 |
SIMD_16x32& B5, |
| 44 |
SIMD_16x32& B6, |
| 45 |
SIMD_16x32& B7) { |
| 46 |
auto t0 = _mm512_unpacklo_epi32(B0.raw(), B1.raw()); |
| 47 |
auto t1 = _mm512_unpackhi_epi32(B0.raw(), B1.raw()); |
| 48 |
auto t2 = _mm512_unpacklo_epi32(B2.raw(), B3.raw()); |
| 49 |
auto t3 = _mm512_unpackhi_epi32(B2.raw(), B3.raw()); |
| 50 |
auto t4 = _mm512_unpacklo_epi32(B4.raw(), B5.raw()); |
| 51 |
auto t5 = _mm512_unpackhi_epi32(B4.raw(), B5.raw()); |
| 52 |
auto t6 = _mm512_unpacklo_epi32(B6.raw(), B7.raw()); |
| 53 |
auto t7 = _mm512_unpackhi_epi32(B6.raw(), B7.raw()); |
| 54 |
|
| 55 |
auto r0 = _mm512_unpacklo_epi64(t0, t2); |
| 56 |
auto r1 = _mm512_unpackhi_epi64(t0, t2); |
| 57 |
auto r2 = _mm512_unpacklo_epi64(t1, t3); |
| 58 |
auto r3 = _mm512_unpackhi_epi64(t1, t3); |
| 59 |
auto r4 = _mm512_unpacklo_epi64(t4, t6); |
| 60 |
auto r5 = _mm512_unpackhi_epi64(t4, t6); |
| 61 |
auto r6 = _mm512_unpacklo_epi64(t5, t7); |
| 62 |
auto r7 = _mm512_unpackhi_epi64(t5, t7); |
| 63 |
|
| 64 |
const __m512i tbl0 = _mm512_set_epi32(27, 19, 26, 18, 25, 17, 24, 16, 11, 3, 10, 2, 9, 1, 8, 0); |
| 65 |
const __m512i tbl1 = _mm512_add_epi32(tbl0, _mm512_set1_epi32(4)); |
| 66 |
B0 = SIMD_16x32(_mm512_permutex2var_epi32(r0, tbl0, r4)); |
| 67 |
B1 = SIMD_16x32(_mm512_permutex2var_epi32(r1, tbl0, r5)); |
| 68 |
B2 = SIMD_16x32(_mm512_permutex2var_epi32(r2, tbl0, r6)); |
| 69 |
B3 = SIMD_16x32(_mm512_permutex2var_epi32(r3, tbl0, r7)); |
| 70 |
B4 = SIMD_16x32(_mm512_permutex2var_epi32(r0, tbl1, r4)); |
| 71 |
B5 = SIMD_16x32(_mm512_permutex2var_epi32(r1, tbl1, r5)); |
| 72 |
B6 = SIMD_16x32(_mm512_permutex2var_epi32(r2, tbl1, r6)); |
| 73 |
B7 = SIMD_16x32(_mm512_permutex2var_epi32(r3, tbl1, r7)); |
| 74 |
} |
| 75 |
|
| 76 |
BOTAN_FN_ISA_AVX512 |
| 77 |
void transpose_out(SIMD_16x32& B0, |
| 78 |
SIMD_16x32& B1, |
| 79 |
SIMD_16x32& B2, |
| 80 |
SIMD_16x32& B3, |
| 81 |
SIMD_16x32& B4, |
| 82 |
SIMD_16x32& B5, |
| 83 |
SIMD_16x32& B6, |
| 84 |
SIMD_16x32& B7) { |
| 85 |
auto t0 = _mm512_unpacklo_epi32(B0.raw(), B1.raw()); |
| 86 |
auto t1 = _mm512_unpackhi_epi32(B0.raw(), B1.raw()); |
| 87 |
auto t2 = _mm512_unpacklo_epi32(B2.raw(), B3.raw()); |
| 88 |
auto t3 = _mm512_unpackhi_epi32(B2.raw(), B3.raw()); |
| 89 |
auto t4 = _mm512_unpacklo_epi32(B4.raw(), B5.raw()); |
| 90 |
auto t5 = _mm512_unpackhi_epi32(B4.raw(), B5.raw()); |
| 91 |
auto t6 = _mm512_unpacklo_epi32(B6.raw(), B7.raw()); |
| 92 |
auto t7 = _mm512_unpackhi_epi32(B6.raw(), B7.raw()); |
| 93 |
|
| 94 |
auto r0 = _mm512_unpacklo_epi64(t0, t2); |
| 95 |
auto r1 = _mm512_unpackhi_epi64(t0, t2); |
| 96 |
auto r2 = _mm512_unpacklo_epi64(t1, t3); |
| 97 |
auto r3 = _mm512_unpackhi_epi64(t1, t3); |
| 98 |
auto r4 = _mm512_unpacklo_epi64(t4, t6); |
| 99 |
auto r5 = _mm512_unpackhi_epi64(t4, t6); |
| 100 |
auto r6 = _mm512_unpacklo_epi64(t5, t7); |
| 101 |
auto r7 = _mm512_unpackhi_epi64(t5, t7); |
| 102 |
|
| 103 |
const __m512i tbl0 = _mm512_set_epi32(23, 22, 21, 20, 7, 6, 5, 4, 19, 18, 17, 16, 3, 2, 1, 0); |
| 104 |
const __m512i tbl1 = _mm512_add_epi32(tbl0, _mm512_set1_epi32(8)); |
| 105 |
|
| 106 |
auto s0 = _mm512_permutex2var_epi32(r0, tbl0, r4); |
| 107 |
auto s1 = _mm512_permutex2var_epi32(r1, tbl0, r5); |
| 108 |
auto s2 = _mm512_permutex2var_epi32(r2, tbl0, r6); |
| 109 |
auto s3 = _mm512_permutex2var_epi32(r3, tbl0, r7); |
| 110 |
auto s4 = _mm512_permutex2var_epi32(r0, tbl1, r4); |
| 111 |
auto s5 = _mm512_permutex2var_epi32(r1, tbl1, r5); |
| 112 |
auto s6 = _mm512_permutex2var_epi32(r2, tbl1, r6); |
| 113 |
auto s7 = _mm512_permutex2var_epi32(r3, tbl1, r7); |
| 114 |
|
| 115 |
B0 = SIMD_16x32(_mm512_shuffle_i32x4(s0, s1, 0b01000100)); |
| 116 |
B1 = SIMD_16x32(_mm512_shuffle_i32x4(s2, s3, 0b01000100)); |
| 117 |
B2 = SIMD_16x32(_mm512_shuffle_i32x4(s0, s1, 0b11101110)); |
| 118 |
B3 = SIMD_16x32(_mm512_shuffle_i32x4(s2, s3, 0b11101110)); |
| 119 |
B4 = SIMD_16x32(_mm512_shuffle_i32x4(s4, s5, 0b01000100)); |
| 120 |
B5 = SIMD_16x32(_mm512_shuffle_i32x4(s6, s7, 0b01000100)); |
| 121 |
B6 = SIMD_16x32(_mm512_shuffle_i32x4(s4, s5, 0b11101110)); |
| 122 |
B7 = SIMD_16x32(_mm512_shuffle_i32x4(s6, s7, 0b11101110)); |
| 123 |
} |
| 124 |
|
| 125 |
// NOLINTEND(portability-simd-intrinsics) |
| 126 |
|
| 127 |
template <typename SimdT> |
| 128 |
void BOTAN_FORCE_INLINE BOTAN_FN_ISA_AVX512 SHACAL2_Fwd(const SimdT& A, |
| 129 |
const SimdT& B, |
| 130 |
const SimdT& C, |
| 131 |
SimdT& D, |
| 132 |
const SimdT& E, |
| 133 |
const SimdT& F, |
| 134 |
const SimdT& G, |
| 135 |
SimdT& H, |
| 136 |
uint32_t RK) { |
| 137 |
H += E.sigma1() + SimdT::choose(E, F, G) + SimdT::splat(RK); |
| 138 |
D += H; |
| 139 |
H += A.sigma0() + SimdT::majority(A, B, C); |
| 140 |
} |
| 141 |
|
| 142 |
template <typename SimdT> |
| 143 |
void BOTAN_FORCE_INLINE BOTAN_FN_ISA_AVX512 SHACAL2_Rev(const SimdT& A, |
| 144 |
const SimdT& B, |
| 145 |
const SimdT& C, |
| 146 |
SimdT& D, |
| 147 |
const SimdT& E, |
| 148 |
const SimdT& F, |
| 149 |
const SimdT& G, |
| 150 |
SimdT& H, |
| 151 |
uint32_t RK) { |
| 152 |
H -= A.sigma0() + SimdT::majority(A, B, C); |
| 153 |
D -= H; |
| 154 |
H -= E.sigma1() + SimdT::choose(E, F, G) + SimdT::splat(RK); |
| 155 |
} |
| 156 |
|
| 157 |
} // namespace |
| 158 |
|
| 159 |
} // namespace SHACAL2_AVX512_F |
| 160 |
|
| 161 |
size_t BOTAN_FN_ISA_AVX512 SHACAL2::avx512_encrypt_blocks(const uint8_t in[], uint8_t out[], size_t blocks) const { |
| 162 |
using namespace SHACAL2_AVX512_F; |
| 163 |
|
| 164 |
size_t consumed = 0; |
| 165 |
|
| 166 |
while(blocks >= 16) { |
| 167 |
SIMD_16x32 A = SIMD_16x32::load_be(in + 64 * 0); |
| 168 |
SIMD_16x32 B = SIMD_16x32::load_be(in + 64 * 1); |
| 169 |
SIMD_16x32 C = SIMD_16x32::load_be(in + 64 * 2); |
| 170 |
SIMD_16x32 D = SIMD_16x32::load_be(in + 64 * 3); |
| 171 |
SIMD_16x32 E = SIMD_16x32::load_be(in + 64 * 4); |
| 172 |
SIMD_16x32 F = SIMD_16x32::load_be(in + 64 * 5); |
| 173 |
SIMD_16x32 G = SIMD_16x32::load_be(in + 64 * 6); |
| 174 |
SIMD_16x32 H = SIMD_16x32::load_be(in + 64 * 7); |
| 175 |
|
| 176 |
transpose_in(A, B, C, D, E, F, G, H); |
| 177 |
|
| 178 |
for(size_t r = 0; r != 64; r += 8) { |
| 179 |
SHACAL2_Fwd(A, B, C, D, E, F, G, H, m_RK[r + 0]); |
| 180 |
SHACAL2_Fwd(H, A, B, C, D, E, F, G, m_RK[r + 1]); |
| 181 |
SHACAL2_Fwd(G, H, A, B, C, D, E, F, m_RK[r + 2]); |
| 182 |
SHACAL2_Fwd(F, G, H, A, B, C, D, E, m_RK[r + 3]); |
| 183 |
SHACAL2_Fwd(E, F, G, H, A, B, C, D, m_RK[r + 4]); |
| 184 |
SHACAL2_Fwd(D, E, F, G, H, A, B, C, m_RK[r + 5]); |
| 185 |
SHACAL2_Fwd(C, D, E, F, G, H, A, B, m_RK[r + 6]); |
| 186 |
SHACAL2_Fwd(B, C, D, E, F, G, H, A, m_RK[r + 7]); |
| 187 |
} |
| 188 |
|
| 189 |
transpose_out(A, B, C, D, E, F, G, H); |
| 190 |
|
| 191 |
A.store_be(out + 64 * 0); |
| 192 |
B.store_be(out + 64 * 1); |
| 193 |
C.store_be(out + 64 * 2); |
| 194 |
D.store_be(out + 64 * 3); |
| 195 |
E.store_be(out + 64 * 4); |
| 196 |
F.store_be(out + 64 * 5); |
| 197 |
G.store_be(out + 64 * 6); |
| 198 |
H.store_be(out + 64 * 7); |
| 199 |
|
| 200 |
in += 16 * BLOCK_SIZE; |
| 201 |
out += 16 * BLOCK_SIZE; |
| 202 |
blocks -= 16; |
| 203 |
consumed += 16; |
| 204 |
} |
| 205 |
|
| 206 |
while(blocks >= 8) { |
| 207 |
SIMD_8x32 A = SIMD_8x32::load_be(in + 32 * 0); |
| 208 |
SIMD_8x32 B = SIMD_8x32::load_be(in + 32 * 1); |
| 209 |
SIMD_8x32 C = SIMD_8x32::load_be(in + 32 * 2); |
| 210 |
SIMD_8x32 D = SIMD_8x32::load_be(in + 32 * 3); |
| 211 |
SIMD_8x32 E = SIMD_8x32::load_be(in + 32 * 4); |
| 212 |
SIMD_8x32 F = SIMD_8x32::load_be(in + 32 * 5); |
| 213 |
SIMD_8x32 G = SIMD_8x32::load_be(in + 32 * 6); |
| 214 |
SIMD_8x32 H = SIMD_8x32::load_be(in + 32 * 7); |
| 215 |
|
| 216 |
SIMD_8x32::transpose(A, B, C, D, E, F, G, H); |
| 217 |
|
| 218 |
for(size_t r = 0; r != 64; r += 8) { |
| 219 |
SHACAL2_Fwd(A, B, C, D, E, F, G, H, m_RK[r + 0]); |
| 220 |
SHACAL2_Fwd(H, A, B, C, D, E, F, G, m_RK[r + 1]); |
| 221 |
SHACAL2_Fwd(G, H, A, B, C, D, E, F, m_RK[r + 2]); |
| 222 |
SHACAL2_Fwd(F, G, H, A, B, C, D, E, m_RK[r + 3]); |
| 223 |
SHACAL2_Fwd(E, F, G, H, A, B, C, D, m_RK[r + 4]); |
| 224 |
SHACAL2_Fwd(D, E, F, G, H, A, B, C, m_RK[r + 5]); |
| 225 |
SHACAL2_Fwd(C, D, E, F, G, H, A, B, m_RK[r + 6]); |
| 226 |
SHACAL2_Fwd(B, C, D, E, F, G, H, A, m_RK[r + 7]); |
| 227 |
} |
| 228 |
|
| 229 |
SIMD_8x32::transpose(A, B, C, D, E, F, G, H); |
| 230 |
|
| 231 |
A.store_be(out + 32 * 0); |
| 232 |
B.store_be(out + 32 * 1); |
| 233 |
C.store_be(out + 32 * 2); |
| 234 |
D.store_be(out + 32 * 3); |
| 235 |
E.store_be(out + 32 * 4); |
| 236 |
F.store_be(out + 32 * 5); |
| 237 |
G.store_be(out + 32 * 6); |
| 238 |
H.store_be(out + 32 * 7); |
| 239 |
|
| 240 |
in += 8 * BLOCK_SIZE; |
| 241 |
out += 8 * BLOCK_SIZE; |
| 242 |
blocks -= 8; |
| 243 |
consumed += 8; |
| 244 |
} |
| 245 |
|
| 246 |
return consumed; |
| 247 |
} |
| 248 |
|
| 249 |
size_t BOTAN_FN_ISA_AVX512 SHACAL2::avx512_decrypt_blocks(const uint8_t in[], uint8_t out[], size_t blocks) const { |
| 250 |
using namespace SHACAL2_AVX512_F; |
| 251 |
|
| 252 |
size_t consumed = 0; |
| 253 |
|
| 254 |
while(blocks >= 16) { |
| 255 |
SIMD_16x32 A = SIMD_16x32::load_be(in + 64 * 0); |
| 256 |
SIMD_16x32 B = SIMD_16x32::load_be(in + 64 * 1); |
| 257 |
SIMD_16x32 C = SIMD_16x32::load_be(in + 64 * 2); |
| 258 |
SIMD_16x32 D = SIMD_16x32::load_be(in + 64 * 3); |
| 259 |
SIMD_16x32 E = SIMD_16x32::load_be(in + 64 * 4); |
| 260 |
SIMD_16x32 F = SIMD_16x32::load_be(in + 64 * 5); |
| 261 |
SIMD_16x32 G = SIMD_16x32::load_be(in + 64 * 6); |
| 262 |
SIMD_16x32 H = SIMD_16x32::load_be(in + 64 * 7); |
| 263 |
|
| 264 |
transpose_in(A, B, C, D, E, F, G, H); |
| 265 |
|
| 266 |
for(size_t r = 0; r != 64; r += 8) { |
| 267 |
SHACAL2_Rev(B, C, D, E, F, G, H, A, m_RK[63 - r]); |
| 268 |
SHACAL2_Rev(C, D, E, F, G, H, A, B, m_RK[62 - r]); |
| 269 |
SHACAL2_Rev(D, E, F, G, H, A, B, C, m_RK[61 - r]); |
| 270 |
SHACAL2_Rev(E, F, G, H, A, B, C, D, m_RK[60 - r]); |
| 271 |
SHACAL2_Rev(F, G, H, A, B, C, D, E, m_RK[59 - r]); |
| 272 |
SHACAL2_Rev(G, H, A, B, C, D, E, F, m_RK[58 - r]); |
| 273 |
SHACAL2_Rev(H, A, B, C, D, E, F, G, m_RK[57 - r]); |
| 274 |
SHACAL2_Rev(A, B, C, D, E, F, G, H, m_RK[56 - r]); |
| 275 |
} |
| 276 |
|
| 277 |
transpose_out(A, B, C, D, E, F, G, H); |
| 278 |
|
| 279 |
A.store_be(out + 64 * 0); |
| 280 |
B.store_be(out + 64 * 1); |
| 281 |
C.store_be(out + 64 * 2); |
| 282 |
D.store_be(out + 64 * 3); |
| 283 |
E.store_be(out + 64 * 4); |
| 284 |
F.store_be(out + 64 * 5); |
| 285 |
G.store_be(out + 64 * 6); |
| 286 |
H.store_be(out + 64 * 7); |
| 287 |
|
| 288 |
in += 16 * BLOCK_SIZE; |
| 289 |
out += 16 * BLOCK_SIZE; |
| 290 |
blocks -= 16; |
| 291 |
consumed += 16; |
| 292 |
} |
| 293 |
|
| 294 |
while(blocks >= 8) { |
| 295 |
SIMD_8x32 A = SIMD_8x32::load_be(in + 32 * 0); |
| 296 |
SIMD_8x32 B = SIMD_8x32::load_be(in + 32 * 1); |
| 297 |
SIMD_8x32 C = SIMD_8x32::load_be(in + 32 * 2); |
| 298 |
SIMD_8x32 D = SIMD_8x32::load_be(in + 32 * 3); |
| 299 |
SIMD_8x32 E = SIMD_8x32::load_be(in + 32 * 4); |
| 300 |
SIMD_8x32 F = SIMD_8x32::load_be(in + 32 * 5); |
| 301 |
SIMD_8x32 G = SIMD_8x32::load_be(in + 32 * 6); |
| 302 |
SIMD_8x32 H = SIMD_8x32::load_be(in + 32 * 7); |
| 303 |
|
| 304 |
SIMD_8x32::transpose(A, B, C, D, E, F, G, H); |
| 305 |
|
| 306 |
for(size_t r = 0; r != 64; r += 8) { |
| 307 |
SHACAL2_Rev(B, C, D, E, F, G, H, A, m_RK[63 - r]); |
| 308 |
SHACAL2_Rev(C, D, E, F, G, H, A, B, m_RK[62 - r]); |
| 309 |
SHACAL2_Rev(D, E, F, G, H, A, B, C, m_RK[61 - r]); |
| 310 |
SHACAL2_Rev(E, F, G, H, A, B, C, D, m_RK[60 - r]); |
| 311 |
SHACAL2_Rev(F, G, H, A, B, C, D, E, m_RK[59 - r]); |
| 312 |
SHACAL2_Rev(G, H, A, B, C, D, E, F, m_RK[58 - r]); |
| 313 |
SHACAL2_Rev(H, A, B, C, D, E, F, G, m_RK[57 - r]); |
| 314 |
SHACAL2_Rev(A, B, C, D, E, F, G, H, m_RK[56 - r]); |
| 315 |
} |
| 316 |
|
| 317 |
SIMD_8x32::transpose(A, B, C, D, E, F, G, H); |
| 318 |
|
| 319 |
A.store_be(out + 32 * 0); |
| 320 |
B.store_be(out + 32 * 1); |
| 321 |
C.store_be(out + 32 * 2); |
| 322 |
D.store_be(out + 32 * 3); |
| 323 |
E.store_be(out + 32 * 4); |
| 324 |
F.store_be(out + 32 * 5); |
| 325 |
G.store_be(out + 32 * 6); |
| 326 |
H.store_be(out + 32 * 7); |
| 327 |
|
| 328 |
in += 8 * BLOCK_SIZE; |
| 329 |
out += 8 * BLOCK_SIZE; |
| 330 |
blocks -= 8; |
| 331 |
consumed += 8; |
| 332 |
} |
| 333 |
|
| 334 |
return consumed; |
| 335 |
} |
| 336 |
|
| 337 |
} // namespace Botan |