forked from weidai11/cryptopp
-
Notifications
You must be signed in to change notification settings - Fork 2
/
chacha_avx.cpp
389 lines (323 loc) · 15.4 KB
/
chacha_avx.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
// chacha_avx.cpp - written and placed in the public domain by
// Jack Lloyd and Jeffrey Walton
//
// This source file uses intrinsics and built-ins to gain access to
// AVX2 instructions. A separate source file is needed because
// additional CXXFLAGS are required to enable the appropriate
// instructions sets in some build configurations.
//
// AVX2 implementation based on Botan's chacha_avx.cpp. Many thanks
// to Jack Lloyd and the Botan team for allowing us to use it.
//
// Here are some relative numbers for ChaCha8:
// * Intel Skylake, 3.0 GHz: AVX2 at 4411 MB/s; 0.57 cpb.
// * Intel Broadwell, 2.3 GHz: AVX2 at 3828 MB/s; 0.58 cpb.
// * AMD Bulldozer, 3.3 GHz: AVX2 at 1680 MB/s; 1.47 cpb.
#include "pch.h"
#include "config.h"
#include "chacha.h"
#include "misc.h"
#if defined(CRYPTOPP_AVX2_AVAILABLE)
# include <xmmintrin.h>
# include <emmintrin.h>
# include <immintrin.h>
#endif
// Squash MS LNK4221 and libtool warnings
extern const char CHACHA_AVX_FNAME[] = __FILE__;
// Sun Studio 12.4 OK, 12.5 and 12.6 compile error.
#if (__SUNPRO_CC >= 0x5140) && (__SUNPRO_CC <= 0x5150)
# define MAYBE_CONST
#else
# define MAYBE_CONST const
#endif
// VS2017 and global optimization bug. TODO, figure out when
// we can re-enable full optimizations for VS2017. Also see
// https://github.com/weidai11/cryptopp/issues/649 and
// https://github.com/weidai11/cryptopp/issues/735. The
// 649 issue affects AES but it is the same here. The 735
// issue is ChaCha AVX2 cut-in where it surfaced again.
#if (_MSC_VER >= 1910)
# ifndef CRYPTOPP_DEBUG
# pragma optimize("", off)
# pragma optimize("ts", on)
# endif
#endif
// The data is aligned, but Clang issues warning based on type
// and not the actual alignment of the variable and data.
#if CRYPTOPP_GCC_DIAGNOSTIC_AVAILABLE
# pragma GCC diagnostic ignored "-Wcast-align"
#endif
ANONYMOUS_NAMESPACE_BEGIN
#if (CRYPTOPP_AVX2_AVAILABLE)
template <unsigned int R>
inline __m256i RotateLeft(const __m256i val)
{
return _mm256_or_si256(_mm256_slli_epi32(val, R), _mm256_srli_epi32(val, 32-R));
}
template <>
inline __m256i RotateLeft<8>(const __m256i val)
{
const __m256i mask = _mm256_set_epi8(14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3,
14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3);
return _mm256_shuffle_epi8(val, mask);
}
template <>
inline __m256i RotateLeft<16>(const __m256i val)
{
const __m256i mask = _mm256_set_epi8(13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2,
13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2);
return _mm256_shuffle_epi8(val, mask);
}
#endif // CRYPTOPP_AVX2_AVAILABLE
ANONYMOUS_NAMESPACE_END
NAMESPACE_BEGIN(CryptoPP)
#if (CRYPTOPP_AVX2_AVAILABLE)
void ChaCha_OperateKeystream_AVX2(const word32 *state, const byte* input, byte *output, unsigned int rounds)
{
MAYBE_CONST __m128i* state_mm = (MAYBE_CONST __m128i*)(state);
MAYBE_CONST __m256i* input_mm = (MAYBE_CONST __m256i*)(input);
__m256i* output_mm = reinterpret_cast<__m256i*>(output);
const __m256i state0 = _mm256_broadcastsi128_si256(_mm_loadu_si128(state_mm + 0));
const __m256i state1 = _mm256_broadcastsi128_si256(_mm_loadu_si128(state_mm + 1));
const __m256i state2 = _mm256_broadcastsi128_si256(_mm_loadu_si128(state_mm + 2));
const __m256i state3 = _mm256_broadcastsi128_si256(_mm_loadu_si128(state_mm + 3));
const __m256i CTR0 = _mm256_set_epi32(0, 0, 0, 0, 0, 0, 0, 4);
const __m256i CTR1 = _mm256_set_epi32(0, 0, 0, 1, 0, 0, 0, 5);
const __m256i CTR2 = _mm256_set_epi32(0, 0, 0, 2, 0, 0, 0, 6);
const __m256i CTR3 = _mm256_set_epi32(0, 0, 0, 3, 0, 0, 0, 7);
__m256i X0_0 = state0;
__m256i X0_1 = state1;
__m256i X0_2 = state2;
__m256i X0_3 = _mm256_add_epi64(state3, CTR0);
__m256i X1_0 = state0;
__m256i X1_1 = state1;
__m256i X1_2 = state2;
__m256i X1_3 = _mm256_add_epi64(state3, CTR1);
__m256i X2_0 = state0;
__m256i X2_1 = state1;
__m256i X2_2 = state2;
__m256i X2_3 = _mm256_add_epi64(state3, CTR2);
__m256i X3_0 = state0;
__m256i X3_1 = state1;
__m256i X3_2 = state2;
__m256i X3_3 = _mm256_add_epi64(state3, CTR3);
for (int i = static_cast<int>(rounds); i > 0; i -= 2)
{
X0_0 = _mm256_add_epi32(X0_0, X0_1);
X1_0 = _mm256_add_epi32(X1_0, X1_1);
X2_0 = _mm256_add_epi32(X2_0, X2_1);
X3_0 = _mm256_add_epi32(X3_0, X3_1);
X0_3 = _mm256_xor_si256(X0_3, X0_0);
X1_3 = _mm256_xor_si256(X1_3, X1_0);
X2_3 = _mm256_xor_si256(X2_3, X2_0);
X3_3 = _mm256_xor_si256(X3_3, X3_0);
X0_3 = RotateLeft<16>(X0_3);
X1_3 = RotateLeft<16>(X1_3);
X2_3 = RotateLeft<16>(X2_3);
X3_3 = RotateLeft<16>(X3_3);
X0_2 = _mm256_add_epi32(X0_2, X0_3);
X1_2 = _mm256_add_epi32(X1_2, X1_3);
X2_2 = _mm256_add_epi32(X2_2, X2_3);
X3_2 = _mm256_add_epi32(X3_2, X3_3);
X0_1 = _mm256_xor_si256(X0_1, X0_2);
X1_1 = _mm256_xor_si256(X1_1, X1_2);
X2_1 = _mm256_xor_si256(X2_1, X2_2);
X3_1 = _mm256_xor_si256(X3_1, X3_2);
X0_1 = RotateLeft<12>(X0_1);
X1_1 = RotateLeft<12>(X1_1);
X2_1 = RotateLeft<12>(X2_1);
X3_1 = RotateLeft<12>(X3_1);
X0_0 = _mm256_add_epi32(X0_0, X0_1);
X1_0 = _mm256_add_epi32(X1_0, X1_1);
X2_0 = _mm256_add_epi32(X2_0, X2_1);
X3_0 = _mm256_add_epi32(X3_0, X3_1);
X0_3 = _mm256_xor_si256(X0_3, X0_0);
X1_3 = _mm256_xor_si256(X1_3, X1_0);
X2_3 = _mm256_xor_si256(X2_3, X2_0);
X3_3 = _mm256_xor_si256(X3_3, X3_0);
X0_3 = RotateLeft<8>(X0_3);
X1_3 = RotateLeft<8>(X1_3);
X2_3 = RotateLeft<8>(X2_3);
X3_3 = RotateLeft<8>(X3_3);
X0_2 = _mm256_add_epi32(X0_2, X0_3);
X1_2 = _mm256_add_epi32(X1_2, X1_3);
X2_2 = _mm256_add_epi32(X2_2, X2_3);
X3_2 = _mm256_add_epi32(X3_2, X3_3);
X0_1 = _mm256_xor_si256(X0_1, X0_2);
X1_1 = _mm256_xor_si256(X1_1, X1_2);
X2_1 = _mm256_xor_si256(X2_1, X2_2);
X3_1 = _mm256_xor_si256(X3_1, X3_2);
X0_1 = RotateLeft<7>(X0_1);
X1_1 = RotateLeft<7>(X1_1);
X2_1 = RotateLeft<7>(X2_1);
X3_1 = RotateLeft<7>(X3_1);
X0_1 = _mm256_shuffle_epi32(X0_1, _MM_SHUFFLE(0, 3, 2, 1));
X0_2 = _mm256_shuffle_epi32(X0_2, _MM_SHUFFLE(1, 0, 3, 2));
X0_3 = _mm256_shuffle_epi32(X0_3, _MM_SHUFFLE(2, 1, 0, 3));
X1_1 = _mm256_shuffle_epi32(X1_1, _MM_SHUFFLE(0, 3, 2, 1));
X1_2 = _mm256_shuffle_epi32(X1_2, _MM_SHUFFLE(1, 0, 3, 2));
X1_3 = _mm256_shuffle_epi32(X1_3, _MM_SHUFFLE(2, 1, 0, 3));
X2_1 = _mm256_shuffle_epi32(X2_1, _MM_SHUFFLE(0, 3, 2, 1));
X2_2 = _mm256_shuffle_epi32(X2_2, _MM_SHUFFLE(1, 0, 3, 2));
X2_3 = _mm256_shuffle_epi32(X2_3, _MM_SHUFFLE(2, 1, 0, 3));
X3_1 = _mm256_shuffle_epi32(X3_1, _MM_SHUFFLE(0, 3, 2, 1));
X3_2 = _mm256_shuffle_epi32(X3_2, _MM_SHUFFLE(1, 0, 3, 2));
X3_3 = _mm256_shuffle_epi32(X3_3, _MM_SHUFFLE(2, 1, 0, 3));
X0_0 = _mm256_add_epi32(X0_0, X0_1);
X1_0 = _mm256_add_epi32(X1_0, X1_1);
X2_0 = _mm256_add_epi32(X2_0, X2_1);
X3_0 = _mm256_add_epi32(X3_0, X3_1);
X0_3 = _mm256_xor_si256(X0_3, X0_0);
X1_3 = _mm256_xor_si256(X1_3, X1_0);
X2_3 = _mm256_xor_si256(X2_3, X2_0);
X3_3 = _mm256_xor_si256(X3_3, X3_0);
X0_3 = RotateLeft<16>(X0_3);
X1_3 = RotateLeft<16>(X1_3);
X2_3 = RotateLeft<16>(X2_3);
X3_3 = RotateLeft<16>(X3_3);
X0_2 = _mm256_add_epi32(X0_2, X0_3);
X1_2 = _mm256_add_epi32(X1_2, X1_3);
X2_2 = _mm256_add_epi32(X2_2, X2_3);
X3_2 = _mm256_add_epi32(X3_2, X3_3);
X0_1 = _mm256_xor_si256(X0_1, X0_2);
X1_1 = _mm256_xor_si256(X1_1, X1_2);
X2_1 = _mm256_xor_si256(X2_1, X2_2);
X3_1 = _mm256_xor_si256(X3_1, X3_2);
X0_1 = RotateLeft<12>(X0_1);
X1_1 = RotateLeft<12>(X1_1);
X2_1 = RotateLeft<12>(X2_1);
X3_1 = RotateLeft<12>(X3_1);
X0_0 = _mm256_add_epi32(X0_0, X0_1);
X1_0 = _mm256_add_epi32(X1_0, X1_1);
X2_0 = _mm256_add_epi32(X2_0, X2_1);
X3_0 = _mm256_add_epi32(X3_0, X3_1);
X0_3 = _mm256_xor_si256(X0_3, X0_0);
X1_3 = _mm256_xor_si256(X1_3, X1_0);
X2_3 = _mm256_xor_si256(X2_3, X2_0);
X3_3 = _mm256_xor_si256(X3_3, X3_0);
X0_3 = RotateLeft<8>(X0_3);
X1_3 = RotateLeft<8>(X1_3);
X2_3 = RotateLeft<8>(X2_3);
X3_3 = RotateLeft<8>(X3_3);
X0_2 = _mm256_add_epi32(X0_2, X0_3);
X1_2 = _mm256_add_epi32(X1_2, X1_3);
X2_2 = _mm256_add_epi32(X2_2, X2_3);
X3_2 = _mm256_add_epi32(X3_2, X3_3);
X0_1 = _mm256_xor_si256(X0_1, X0_2);
X1_1 = _mm256_xor_si256(X1_1, X1_2);
X2_1 = _mm256_xor_si256(X2_1, X2_2);
X3_1 = _mm256_xor_si256(X3_1, X3_2);
X0_1 = RotateLeft<7>(X0_1);
X1_1 = RotateLeft<7>(X1_1);
X2_1 = RotateLeft<7>(X2_1);
X3_1 = RotateLeft<7>(X3_1);
X0_1 = _mm256_shuffle_epi32(X0_1, _MM_SHUFFLE(2, 1, 0, 3));
X0_2 = _mm256_shuffle_epi32(X0_2, _MM_SHUFFLE(1, 0, 3, 2));
X0_3 = _mm256_shuffle_epi32(X0_3, _MM_SHUFFLE(0, 3, 2, 1));
X1_1 = _mm256_shuffle_epi32(X1_1, _MM_SHUFFLE(2, 1, 0, 3));
X1_2 = _mm256_shuffle_epi32(X1_2, _MM_SHUFFLE(1, 0, 3, 2));
X1_3 = _mm256_shuffle_epi32(X1_3, _MM_SHUFFLE(0, 3, 2, 1));
X2_1 = _mm256_shuffle_epi32(X2_1, _MM_SHUFFLE(2, 1, 0, 3));
X2_2 = _mm256_shuffle_epi32(X2_2, _MM_SHUFFLE(1, 0, 3, 2));
X2_3 = _mm256_shuffle_epi32(X2_3, _MM_SHUFFLE(0, 3, 2, 1));
X3_1 = _mm256_shuffle_epi32(X3_1, _MM_SHUFFLE(2, 1, 0, 3));
X3_2 = _mm256_shuffle_epi32(X3_2, _MM_SHUFFLE(1, 0, 3, 2));
X3_3 = _mm256_shuffle_epi32(X3_3, _MM_SHUFFLE(0, 3, 2, 1));
}
X0_0 = _mm256_add_epi32(X0_0, state0);
X0_1 = _mm256_add_epi32(X0_1, state1);
X0_2 = _mm256_add_epi32(X0_2, state2);
X0_3 = _mm256_add_epi32(X0_3, state3);
X0_3 = _mm256_add_epi64(X0_3, CTR0);
X1_0 = _mm256_add_epi32(X1_0, state0);
X1_1 = _mm256_add_epi32(X1_1, state1);
X1_2 = _mm256_add_epi32(X1_2, state2);
X1_3 = _mm256_add_epi32(X1_3, state3);
X1_3 = _mm256_add_epi64(X1_3, CTR1);
X2_0 = _mm256_add_epi32(X2_0, state0);
X2_1 = _mm256_add_epi32(X2_1, state1);
X2_2 = _mm256_add_epi32(X2_2, state2);
X2_3 = _mm256_add_epi32(X2_3, state3);
X2_3 = _mm256_add_epi64(X2_3, CTR2);
X3_0 = _mm256_add_epi32(X3_0, state0);
X3_1 = _mm256_add_epi32(X3_1, state1);
X3_2 = _mm256_add_epi32(X3_2, state2);
X3_3 = _mm256_add_epi32(X3_3, state3);
X3_3 = _mm256_add_epi64(X3_3, CTR3);
if (input_mm)
{
_mm256_storeu_si256(output_mm + 0, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 0),
_mm256_permute2x128_si256(X0_0, X0_1, 1 + (3 << 4))));
_mm256_storeu_si256(output_mm + 1, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 1),
_mm256_permute2x128_si256(X0_2, X0_3, 1 + (3 << 4))));
_mm256_storeu_si256(output_mm + 2, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 2),
_mm256_permute2x128_si256(X1_0, X1_1, 1 + (3 << 4))));
_mm256_storeu_si256(output_mm + 3, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 3),
_mm256_permute2x128_si256(X1_2, X1_3, 1 + (3 << 4))));
}
else
{
_mm256_storeu_si256(output_mm + 0, _mm256_permute2x128_si256(X0_0, X0_1, 1 + (3 << 4)));
_mm256_storeu_si256(output_mm + 1, _mm256_permute2x128_si256(X0_2, X0_3, 1 + (3 << 4)));
_mm256_storeu_si256(output_mm + 2, _mm256_permute2x128_si256(X1_0, X1_1, 1 + (3 << 4)));
_mm256_storeu_si256(output_mm + 3, _mm256_permute2x128_si256(X1_2, X1_3, 1 + (3 << 4)));
}
if (input_mm)
{
_mm256_storeu_si256(output_mm + 4, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 4),
_mm256_permute2x128_si256(X2_0, X2_1, 1 + (3 << 4))));
_mm256_storeu_si256(output_mm + 5, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 5),
_mm256_permute2x128_si256(X2_2, X2_3, 1 + (3 << 4))));
_mm256_storeu_si256(output_mm + 6, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 6),
_mm256_permute2x128_si256(X3_0, X3_1, 1 + (3 << 4))));
_mm256_storeu_si256(output_mm + 7, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 7),
_mm256_permute2x128_si256(X3_2, X3_3, 1 + (3 << 4))));
}
else
{
_mm256_storeu_si256(output_mm + 4, _mm256_permute2x128_si256(X2_0, X2_1, 1 + (3 << 4)));
_mm256_storeu_si256(output_mm + 5, _mm256_permute2x128_si256(X2_2, X2_3, 1 + (3 << 4)));
_mm256_storeu_si256(output_mm + 6, _mm256_permute2x128_si256(X3_0, X3_1, 1 + (3 << 4)));
_mm256_storeu_si256(output_mm + 7, _mm256_permute2x128_si256(X3_2, X3_3, 1 + (3 << 4)));
}
if (input_mm)
{
_mm256_storeu_si256(output_mm + 8, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 8),
_mm256_permute2x128_si256(X0_0, X0_1, 0 + (2 << 4))));
_mm256_storeu_si256(output_mm + 9, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 9),
_mm256_permute2x128_si256(X0_2, X0_3, 0 + (2 << 4))));
_mm256_storeu_si256(output_mm + 10, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 10),
_mm256_permute2x128_si256(X1_0, X1_1, 0 + (2 << 4))));
_mm256_storeu_si256(output_mm + 11, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 11),
_mm256_permute2x128_si256(X1_2, X1_3, 0 + (2 << 4))));
}
else
{
_mm256_storeu_si256(output_mm + 8, _mm256_permute2x128_si256(X0_0, X0_1, 0 + (2 << 4)));
_mm256_storeu_si256(output_mm + 9, _mm256_permute2x128_si256(X0_2, X0_3, 0 + (2 << 4)));
_mm256_storeu_si256(output_mm + 10, _mm256_permute2x128_si256(X1_0, X1_1, 0 + (2 << 4)));
_mm256_storeu_si256(output_mm + 11, _mm256_permute2x128_si256(X1_2, X1_3, 0 + (2 << 4)));
}
if (input_mm)
{
_mm256_storeu_si256(output_mm + 12, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 12),
_mm256_permute2x128_si256(X2_0, X2_1, 0 + (2 << 4))));
_mm256_storeu_si256(output_mm + 13, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 13),
_mm256_permute2x128_si256(X2_2, X2_3, 0 + (2 << 4))));
_mm256_storeu_si256(output_mm + 14, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 14),
_mm256_permute2x128_si256(X3_0, X3_1, 0 + (2 << 4))));
_mm256_storeu_si256(output_mm + 15, _mm256_xor_si256(_mm256_loadu_si256(input_mm + 15),
_mm256_permute2x128_si256(X3_2, X3_3, 0 + (2 << 4))));
}
else
{
_mm256_storeu_si256(output_mm + 12, _mm256_permute2x128_si256(X2_0, X2_1, 0 + (2 << 4)));
_mm256_storeu_si256(output_mm + 13, _mm256_permute2x128_si256(X2_2, X2_3, 0 + (2 << 4)));
_mm256_storeu_si256(output_mm + 14, _mm256_permute2x128_si256(X3_0, X3_1, 0 + (2 << 4)));
_mm256_storeu_si256(output_mm + 15, _mm256_permute2x128_si256(X3_2, X3_3, 0 + (2 << 4)));
}
// https://software.intel.com/en-us/articles/avoiding-avx-sse-transition-penalties
_mm256_zeroupper();
}
#endif // CRYPTOPP_AVX2_AVAILABLE
NAMESPACE_END