CoCalc -- chacha

GitHub Repository: folium-app/Folium
Path: blob/a-new-beginning/SharedDependencies/Sources/cryptopp/chacha_avx.cpp
² views
1
// chacha_avx.cpp - written and placed in the public domain by
2
//                  Jack Lloyd and Jeffrey Walton
3
//
4
//    This source file uses intrinsics and built-ins to gain access to
5
//    AVX2 instructions. A separate source file is needed because
6
//    additional CXXFLAGS are required to enable the appropriate
7
//    instructions sets in some build configurations.
8
//
9
//    AVX2 implementation based on Botan's chacha_avx.cpp. Many thanks
10
//    to Jack Lloyd and the Botan team for allowing us to use it.
11
//
12
//    Here are some relative numbers for ChaCha8:
13
//    * Intel Skylake,   3.0 GHz: AVX2 at 4411 MB/s; 0.57 cpb.
14
//    * Intel Broadwell, 2.3 GHz: AVX2 at 3828 MB/s; 0.58 cpb.
15
//    * AMD Bulldozer,   3.3 GHz: AVX2 at 1680 MB/s; 1.47 cpb.
16

17
#include "pch.h"
18
#include "config.h"
19

20
#include "chacha.h"
21
#include "misc.h"
22

23
#if defined(CRYPTOPP_AVX2_AVAILABLE)
24
# include <xmmintrin.h>
25
# include <emmintrin.h>
26
# include <immintrin.h>
27
#endif
28

29
// Squash MS LNK4221 and libtool warnings
30
extern const char CHACHA_AVX_FNAME[] = __FILE__;
31

32
// Sun Studio 12.4 OK, 12.5 and 12.6 compile error.
33
#if (__SUNPRO_CC >= 0x5140) && (__SUNPRO_CC <= 0x5150)
34
# define MAYBE_CONST
35
#else
36
# define MAYBE_CONST const
37
#endif
38

39
// VS2017 and global optimization bug. Also see
40
// https://github.com/weidai11/cryptopp/issues/649 and
41
// https://github.com/weidai11/cryptopp/issues/735. The
42
// 649 issue affects AES but it is the same here. The 735
43
// issue is ChaCha AVX2 cut-in where it surfaced again.
44
#if (CRYPTOPP_MSC_VERSION >= 1910) && (CRYPTOPP_MSC_VERSION <= 1916)
45
# ifndef CRYPTOPP_DEBUG
46
#  pragma optimize("", off)
47
#  pragma optimize("ts", on)
48
# endif
49
#endif
50

51
// The data is aligned, but Clang issues warning based on type
52
// and not the actual alignment of the variable and data.
53
#if CRYPTOPP_GCC_DIAGNOSTIC_AVAILABLE
54
# pragma GCC diagnostic ignored "-Wcast-align"
55
#endif
56

57
ANONYMOUS_NAMESPACE_BEGIN
58

59
#if (CRYPTOPP_AVX2_AVAILABLE)
60

61
template <unsigned int R>
62
inline __m256i RotateLeft(const __m256i val)
63
{
64
    return _mm256_or_si256(_mm256_slli_epi32(val, R), _mm256_srli_epi32(val, 32-R));
65
}
66

67
template <>
68
inline __m256i RotateLeft<8>(const __m256i val)
69
{
70
    const __m256i mask = _mm256_set_epi8(14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3,
71
                                         14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3);
72
    return _mm256_shuffle_epi8(val, mask);
73
}
74

75
template <>
76
inline __m256i RotateLeft<16>(const __m256i val)
77
{
78
    const __m256i mask = _mm256_set_epi8(13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2,
79
                                         13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2);
80
    return _mm256_shuffle_epi8(val, mask);
81
}
82

83
#endif  // CRYPTOPP_AVX2_AVAILABLE
84

85
ANONYMOUS_NAMESPACE_END
86

87
NAMESPACE_BEGIN(CryptoPP)
88

89
#if (CRYPTOPP_AVX2_AVAILABLE)
90

91
void ChaCha_OperateKeystream_AVX2(const word32 *state, const byte* input, byte *output, unsigned int rounds)
92
{
93
    const __m256i state0 = _mm256_broadcastsi128_si256(
94
        _mm_loadu_si128(reinterpret_cast<const __m128i*>(state+0*4)));
95
    const __m256i state1 = _mm256_broadcastsi128_si256(
96
        _mm_loadu_si128(reinterpret_cast<const __m128i*>(state+1*4)));
97
    const __m256i state2 = _mm256_broadcastsi128_si256(
98
        _mm_loadu_si128(reinterpret_cast<const __m128i*>(state+2*4)));
99
    const __m256i state3 = _mm256_broadcastsi128_si256(
100
        _mm_loadu_si128(reinterpret_cast<const __m128i*>(state+3*4)));
101

102
    const word32 C = 0xFFFFFFFFu - state[12];
103
    const __m256i CTR0 = _mm256_set_epi32(0, 0,     0, 0, 0, 0, C < 4, 4);
104
    const __m256i CTR1 = _mm256_set_epi32(0, 0, C < 1, 1, 0, 0, C < 5, 5);
105
    const __m256i CTR2 = _mm256_set_epi32(0, 0, C < 2, 2, 0, 0, C < 6, 6);
106
    const __m256i CTR3 = _mm256_set_epi32(0, 0, C < 3, 3, 0, 0, C < 7, 7);
107

108
    __m256i X0_0 = state0;
109
    __m256i X0_1 = state1;
110
    __m256i X0_2 = state2;
111
    __m256i X0_3 = _mm256_add_epi32(state3, CTR0);
112

113
    __m256i X1_0 = state0;
114
    __m256i X1_1 = state1;
115
    __m256i X1_2 = state2;
116
    __m256i X1_3 = _mm256_add_epi32(state3, CTR1);
117

118
    __m256i X2_0 = state0;
119
    __m256i X2_1 = state1;
120
    __m256i X2_2 = state2;
121
    __m256i X2_3 = _mm256_add_epi32(state3, CTR2);
122

123
    __m256i X3_0 = state0;
124
    __m256i X3_1 = state1;
125
    __m256i X3_2 = state2;
126
    __m256i X3_3 = _mm256_add_epi32(state3, CTR3);
127

128
    for (int i = static_cast<int>(rounds); i > 0; i -= 2)
129
    {
130
        X0_0 = _mm256_add_epi32(X0_0, X0_1);
131
        X1_0 = _mm256_add_epi32(X1_0, X1_1);
132
        X2_0 = _mm256_add_epi32(X2_0, X2_1);
133
        X3_0 = _mm256_add_epi32(X3_0, X3_1);
134

135
        X0_3 = _mm256_xor_si256(X0_3, X0_0);
136
        X1_3 = _mm256_xor_si256(X1_3, X1_0);
137
        X2_3 = _mm256_xor_si256(X2_3, X2_0);
138
        X3_3 = _mm256_xor_si256(X3_3, X3_0);
139

140
        X0_3 = RotateLeft<16>(X0_3);
141
        X1_3 = RotateLeft<16>(X1_3);
142
        X2_3 = RotateLeft<16>(X2_3);
143
        X3_3 = RotateLeft<16>(X3_3);
144

145
        X0_2 = _mm256_add_epi32(X0_2, X0_3);
146
        X1_2 = _mm256_add_epi32(X1_2, X1_3);
147
        X2_2 = _mm256_add_epi32(X2_2, X2_3);
148
        X3_2 = _mm256_add_epi32(X3_2, X3_3);
149

150
        X0_1 = _mm256_xor_si256(X0_1, X0_2);
151
        X1_1 = _mm256_xor_si256(X1_1, X1_2);
152
        X2_1 = _mm256_xor_si256(X2_1, X2_2);
153
        X3_1 = _mm256_xor_si256(X3_1, X3_2);
154

155
        X0_1 = RotateLeft<12>(X0_1);
156
        X1_1 = RotateLeft<12>(X1_1);
157
        X2_1 = RotateLeft<12>(X2_1);
158
        X3_1 = RotateLeft<12>(X3_1);
159

160
        X0_0 = _mm256_add_epi32(X0_0, X0_1);
161
        X1_0 = _mm256_add_epi32(X1_0, X1_1);
162
        X2_0 = _mm256_add_epi32(X2_0, X2_1);
163
        X3_0 = _mm256_add_epi32(X3_0, X3_1);
164

165
        X0_3 = _mm256_xor_si256(X0_3, X0_0);
166
        X1_3 = _mm256_xor_si256(X1_3, X1_0);
167
        X2_3 = _mm256_xor_si256(X2_3, X2_0);
168
        X3_3 = _mm256_xor_si256(X3_3, X3_0);
169

170
        X0_3 = RotateLeft<8>(X0_3);
171
        X1_3 = RotateLeft<8>(X1_3);
172
        X2_3 = RotateLeft<8>(X2_3);
173
        X3_3 = RotateLeft<8>(X3_3);
174

175
        X0_2 = _mm256_add_epi32(X0_2, X0_3);
176
        X1_2 = _mm256_add_epi32(X1_2, X1_3);
177
        X2_2 = _mm256_add_epi32(X2_2, X2_3);
178
        X3_2 = _mm256_add_epi32(X3_2, X3_3);
179

180
        X0_1 = _mm256_xor_si256(X0_1, X0_2);
181
        X1_1 = _mm256_xor_si256(X1_1, X1_2);
182
        X2_1 = _mm256_xor_si256(X2_1, X2_2);
183
        X3_1 = _mm256_xor_si256(X3_1, X3_2);
184

185
        X0_1 = RotateLeft<7>(X0_1);
186
        X1_1 = RotateLeft<7>(X1_1);
187
        X2_1 = RotateLeft<7>(X2_1);
188
        X3_1 = RotateLeft<7>(X3_1);
189

190
        X0_1 = _mm256_shuffle_epi32(X0_1, _MM_SHUFFLE(0, 3, 2, 1));
191
        X0_2 = _mm256_shuffle_epi32(X0_2, _MM_SHUFFLE(1, 0, 3, 2));
192
        X0_3 = _mm256_shuffle_epi32(X0_3, _MM_SHUFFLE(2, 1, 0, 3));
193

194
        X1_1 = _mm256_shuffle_epi32(X1_1, _MM_SHUFFLE(0, 3, 2, 1));
195
        X1_2 = _mm256_shuffle_epi32(X1_2, _MM_SHUFFLE(1, 0, 3, 2));
196
        X1_3 = _mm256_shuffle_epi32(X1_3, _MM_SHUFFLE(2, 1, 0, 3));
197

198
        X2_1 = _mm256_shuffle_epi32(X2_1, _MM_SHUFFLE(0, 3, 2, 1));
199
        X2_2 = _mm256_shuffle_epi32(X2_2, _MM_SHUFFLE(1, 0, 3, 2));
200
        X2_3 = _mm256_shuffle_epi32(X2_3, _MM_SHUFFLE(2, 1, 0, 3));
201

202
        X3_1 = _mm256_shuffle_epi32(X3_1, _MM_SHUFFLE(0, 3, 2, 1));
203
        X3_2 = _mm256_shuffle_epi32(X3_2, _MM_SHUFFLE(1, 0, 3, 2));
204
        X3_3 = _mm256_shuffle_epi32(X3_3, _MM_SHUFFLE(2, 1, 0, 3));
205

206
        X0_0 = _mm256_add_epi32(X0_0, X0_1);
207
        X1_0 = _mm256_add_epi32(X1_0, X1_1);
208
        X2_0 = _mm256_add_epi32(X2_0, X2_1);
209
        X3_0 = _mm256_add_epi32(X3_0, X3_1);
210

211
        X0_3 = _mm256_xor_si256(X0_3, X0_0);
212
        X1_3 = _mm256_xor_si256(X1_3, X1_0);
213
        X2_3 = _mm256_xor_si256(X2_3, X2_0);
214
        X3_3 = _mm256_xor_si256(X3_3, X3_0);
215

216
        X0_3 = RotateLeft<16>(X0_3);
217
        X1_3 = RotateLeft<16>(X1_3);
218
        X2_3 = RotateLeft<16>(X2_3);
219
        X3_3 = RotateLeft<16>(X3_3);
220

221
        X0_2 = _mm256_add_epi32(X0_2, X0_3);
222
        X1_2 = _mm256_add_epi32(X1_2, X1_3);
223
        X2_2 = _mm256_add_epi32(X2_2, X2_3);
224
        X3_2 = _mm256_add_epi32(X3_2, X3_3);
225

226
        X0_1 = _mm256_xor_si256(X0_1, X0_2);
227
        X1_1 = _mm256_xor_si256(X1_1, X1_2);
228
        X2_1 = _mm256_xor_si256(X2_1, X2_2);
229
        X3_1 = _mm256_xor_si256(X3_1, X3_2);
230

231
        X0_1 = RotateLeft<12>(X0_1);
232
        X1_1 = RotateLeft<12>(X1_1);
233
        X2_1 = RotateLeft<12>(X2_1);
234
        X3_1 = RotateLeft<12>(X3_1);
235

236
        X0_0 = _mm256_add_epi32(X0_0, X0_1);
237
        X1_0 = _mm256_add_epi32(X1_0, X1_1);
238
        X2_0 = _mm256_add_epi32(X2_0, X2_1);
239
        X3_0 = _mm256_add_epi32(X3_0, X3_1);
240

241
        X0_3 = _mm256_xor_si256(X0_3, X0_0);
242
        X1_3 = _mm256_xor_si256(X1_3, X1_0);
243
        X2_3 = _mm256_xor_si256(X2_3, X2_0);
244
        X3_3 = _mm256_xor_si256(X3_3, X3_0);
245

246
        X0_3 = RotateLeft<8>(X0_3);
247
        X1_3 = RotateLeft<8>(X1_3);
248
        X2_3 = RotateLeft<8>(X2_3);
249
        X3_3 = RotateLeft<8>(X3_3);
250

251
        X0_2 = _mm256_add_epi32(X0_2, X0_3);
252
        X1_2 = _mm256_add_epi32(X1_2, X1_3);
253
        X2_2 = _mm256_add_epi32(X2_2, X2_3);
254
        X3_2 = _mm256_add_epi32(X3_2, X3_3);
255

256
        X0_1 = _mm256_xor_si256(X0_1, X0_2);
257
        X1_1 = _mm256_xor_si256(X1_1, X1_2);
258
        X2_1 = _mm256_xor_si256(X2_1, X2_2);
259
        X3_1 = _mm256_xor_si256(X3_1, X3_2);
260

261
        X0_1 = RotateLeft<7>(X0_1);
262
        X1_1 = RotateLeft<7>(X1_1);
263
        X2_1 = RotateLeft<7>(X2_1);
264
        X3_1 = RotateLeft<7>(X3_1);
265

266
        X0_1 = _mm256_shuffle_epi32(X0_1, _MM_SHUFFLE(2, 1, 0, 3));
267
        X0_2 = _mm256_shuffle_epi32(X0_2, _MM_SHUFFLE(1, 0, 3, 2));
268
        X0_3 = _mm256_shuffle_epi32(X0_3, _MM_SHUFFLE(0, 3, 2, 1));
269

270
        X1_1 = _mm256_shuffle_epi32(X1_1, _MM_SHUFFLE(2, 1, 0, 3));
271
        X1_2 = _mm256_shuffle_epi32(X1_2, _MM_SHUFFLE(1, 0, 3, 2));
272
        X1_3 = _mm256_shuffle_epi32(X1_3, _MM_SHUFFLE(0, 3, 2, 1));
273

274
        X2_1 = _mm256_shuffle_epi32(X2_1, _MM_SHUFFLE(2, 1, 0, 3));
275
        X2_2 = _mm256_shuffle_epi32(X2_2, _MM_SHUFFLE(1, 0, 3, 2));
276
        X2_3 = _mm256_shuffle_epi32(X2_3, _MM_SHUFFLE(0, 3, 2, 1));
277

278
        X3_1 = _mm256_shuffle_epi32(X3_1, _MM_SHUFFLE(2, 1, 0, 3));
279
        X3_2 = _mm256_shuffle_epi32(X3_2, _MM_SHUFFLE(1, 0, 3, 2));
280
        X3_3 = _mm256_shuffle_epi32(X3_3, _MM_SHUFFLE(0, 3, 2, 1));
281
    }
282

283
    X0_0 = _mm256_add_epi32(X0_0, state0);
284
    X0_1 = _mm256_add_epi32(X0_1, state1);
285
    X0_2 = _mm256_add_epi32(X0_2, state2);
286
    X0_3 = _mm256_add_epi32(X0_3, state3);
287
    X0_3 = _mm256_add_epi32(X0_3, CTR0);
288

289
    X1_0 = _mm256_add_epi32(X1_0, state0);
290
    X1_1 = _mm256_add_epi32(X1_1, state1);
291
    X1_2 = _mm256_add_epi32(X1_2, state2);
292
    X1_3 = _mm256_add_epi32(X1_3, state3);
293
    X1_3 = _mm256_add_epi32(X1_3, CTR1);
294

295
    X2_0 = _mm256_add_epi32(X2_0, state0);
296
    X2_1 = _mm256_add_epi32(X2_1, state1);
297
    X2_2 = _mm256_add_epi32(X2_2, state2);
298
    X2_3 = _mm256_add_epi32(X2_3, state3);
299
    X2_3 = _mm256_add_epi32(X2_3, CTR2);
300

301
    X3_0 = _mm256_add_epi32(X3_0, state0);
302
    X3_1 = _mm256_add_epi32(X3_1, state1);
303
    X3_2 = _mm256_add_epi32(X3_2, state2);
304
    X3_3 = _mm256_add_epi32(X3_3, state3);
305
    X3_3 = _mm256_add_epi32(X3_3, CTR3);
306

307
    if (input)
308
    {
309
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+0*32),
310
            _mm256_xor_si256(_mm256_permute2x128_si256(X0_0, X0_1, 1 + (3 << 4)),
311
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+0*32)))));
312
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+1*32),
313
            _mm256_xor_si256(_mm256_permute2x128_si256(X0_2, X0_3, 1 + (3 << 4)),
314
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+1*32)))));
315
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+2*32),
316
            _mm256_xor_si256(_mm256_permute2x128_si256(X1_0, X1_1, 1 + (3 << 4)),
317
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+2*32)))));
318
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+3*32),
319
            _mm256_xor_si256(_mm256_permute2x128_si256(X1_2, X1_3, 1 + (3 << 4)),
320
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+3*32)))));
321
    }
322
    else
323
    {
324
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+0*32),
325
            _mm256_permute2x128_si256(X0_0, X0_1, 1 + (3 << 4)));
326
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+1*32),
327
            _mm256_permute2x128_si256(X0_2, X0_3, 1 + (3 << 4)));
328
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+2*32),
329
            _mm256_permute2x128_si256(X1_0, X1_1, 1 + (3 << 4)));
330
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+3*32),
331
            _mm256_permute2x128_si256(X1_2, X1_3, 1 + (3 << 4)));
332
    }
333

334
    if (input)
335
    {
336
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+4*32),
337
            _mm256_xor_si256(_mm256_permute2x128_si256(X2_0, X2_1, 1 + (3 << 4)),
338
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+4*32)))));
339
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+5*32),
340
            _mm256_xor_si256(_mm256_permute2x128_si256(X2_2, X2_3, 1 + (3 << 4)),
341
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+5*32)))));
342
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+6*32),
343
            _mm256_xor_si256(_mm256_permute2x128_si256(X3_0, X3_1, 1 + (3 << 4)),
344
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+6*32)))));
345
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+7*32),
346
            _mm256_xor_si256(_mm256_permute2x128_si256(X3_2, X3_3, 1 + (3 << 4)),
347
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+7*32)))));
348
    }
349
    else
350
    {
351
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+4*32),
352
            _mm256_permute2x128_si256(X2_0, X2_1, 1 + (3 << 4)));
353
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+5*32),
354
            _mm256_permute2x128_si256(X2_2, X2_3, 1 + (3 << 4)));
355
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+6*32),
356
            _mm256_permute2x128_si256(X3_0, X3_1, 1 + (3 << 4)));
357
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+7*32),
358
            _mm256_permute2x128_si256(X3_2, X3_3, 1 + (3 << 4)));
359
    }
360

361
    if (input)
362
    {
363
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+ 8*32),
364
            _mm256_xor_si256(_mm256_permute2x128_si256(X0_0, X0_1, 0 + (2 << 4)),
365
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+8*32)))));
366
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+ 9*32),
367
            _mm256_xor_si256(_mm256_permute2x128_si256(X0_2, X0_3, 0 + (2 << 4)),
368
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+9*32)))));
369
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+10*32),
370
            _mm256_xor_si256(_mm256_permute2x128_si256(X1_0, X1_1, 0 + (2 << 4)),
371
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+10*32)))));
372
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+11*32),
373
            _mm256_xor_si256(_mm256_permute2x128_si256(X1_2, X1_3, 0 + (2 << 4)),
374
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+11*32)))));
375
    }
376
    else
377
    {
378
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+ 8*32),
379
            _mm256_permute2x128_si256(X0_0, X0_1, 0 + (2 << 4)));
380
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+ 9*32),
381
            _mm256_permute2x128_si256(X0_2, X0_3, 0 + (2 << 4)));
382
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+10*32),
383
            _mm256_permute2x128_si256(X1_0, X1_1, 0 + (2 << 4)));
384
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+11*32),
385
            _mm256_permute2x128_si256(X1_2, X1_3, 0 + (2 << 4)));
386
    }
387

388
    if (input)
389
    {
390
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+12*32),
391
            _mm256_xor_si256(_mm256_permute2x128_si256(X2_0, X2_1, 0 + (2 << 4)),
392
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+12*32)))));
393
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+13*32),
394
            _mm256_xor_si256(_mm256_permute2x128_si256(X2_2, X2_3, 0 + (2 << 4)),
395
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+13*32)))));
396
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+14*32),
397
            _mm256_xor_si256(_mm256_permute2x128_si256(X3_0, X3_1, 0 + (2 << 4)),
398
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+14*32)))));
399
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+15*32),
400
            _mm256_xor_si256(_mm256_permute2x128_si256(X3_2, X3_3, 0 + (2 << 4)),
401
            _mm256_loadu_si256(const_cast<MAYBE_CONST __m256i*>(reinterpret_cast<const __m256i*>(input+15*32)))));
402
    }
403
    else
404
    {
405
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+12*32),
406
            _mm256_permute2x128_si256(X2_0, X2_1, 0 + (2 << 4)));
407
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+13*32),
408
            _mm256_permute2x128_si256(X2_2, X2_3, 0 + (2 << 4)));
409
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+14*32),
410
            _mm256_permute2x128_si256(X3_0, X3_1, 0 + (2 << 4)));
411
        _mm256_storeu_si256(reinterpret_cast<__m256i*>(output+15*32),
412
            _mm256_permute2x128_si256(X3_2, X3_3, 0 + (2 << 4)));
413
    }
414

415
    // https://software.intel.com/en-us/articles/avoiding-avx-sse-transition-penalties
416
    _mm256_zeroupper();
417
}
418

419
#endif  // CRYPTOPP_AVX2_AVAILABLE
420

421
NAMESPACE_END
422

423
Product

Resources

Company
