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
|
/*
* SHA-1 using SSE2
* Based on public domain code by Dean Gaudet
* (http://arctic.org/~dean/crypto/sha1.html)
* (C) 2009-2011 Jack Lloyd
*
* Botan is released under the Simplified BSD License (see license.txt)
*/
#include <botan/sha160.h>
#include <emmintrin.h>
namespace Botan {
namespace SHA1_SSE2_F {
namespace {
/*
* First 16 bytes just need byte swapping. Preparing just means
* adding in the round constants.
*/
#define prep00_15(P, W) \
do { \
W = _mm_shufflehi_epi16(W, _MM_SHUFFLE(2, 3, 0, 1)); \
W = _mm_shufflelo_epi16(W, _MM_SHUFFLE(2, 3, 0, 1)); \
W = _mm_or_si128(_mm_slli_epi16(W, 8), \
_mm_srli_epi16(W, 8)); \
P.u128 = _mm_add_epi32(W, K00_19); \
} while(0)
/*
For each multiple of 4, t, we want to calculate this:
W[t+0] = rol(W[t-3] ^ W[t-8] ^ W[t-14] ^ W[t-16], 1);
W[t+1] = rol(W[t-2] ^ W[t-7] ^ W[t-13] ^ W[t-15], 1);
W[t+2] = rol(W[t-1] ^ W[t-6] ^ W[t-12] ^ W[t-14], 1);
W[t+3] = rol(W[t] ^ W[t-5] ^ W[t-11] ^ W[t-13], 1);
we'll actually calculate this:
W[t+0] = rol(W[t-3] ^ W[t-8] ^ W[t-14] ^ W[t-16], 1);
W[t+1] = rol(W[t-2] ^ W[t-7] ^ W[t-13] ^ W[t-15], 1);
W[t+2] = rol(W[t-1] ^ W[t-6] ^ W[t-12] ^ W[t-14], 1);
W[t+3] = rol( 0 ^ W[t-5] ^ W[t-11] ^ W[t-13], 1);
W[t+3] ^= rol(W[t+0], 1);
the parameters are:
W0 = &W[t-16];
W1 = &W[t-12];
W2 = &W[t- 8];
W3 = &W[t- 4];
and on output:
prepared = W0 + K
W0 = W[t]..W[t+3]
*/
/* note that there is a step here where i want to do a rol by 1, which
* normally would look like this:
*
* r1 = psrld r0,$31
* r0 = pslld r0,$1
* r0 = por r0,r1
*
* but instead i do this:
*
* r1 = pcmpltd r0,zero
* r0 = paddd r0,r0
* r0 = psub r0,r1
*
* because pcmpltd and paddd are available in both MMX units on
* efficeon, pentium-m, and opteron but shifts are available in
* only one unit.
*/
#define prep(prep, XW0, XW1, XW2, XW3, K) \
do { \
__m128i r0, r1, r2, r3; \
\
/* load W[t-4] 16-byte aligned, and shift */ \
r3 = _mm_srli_si128((XW3), 4); \
r0 = (XW0); \
/* get high 64-bits of XW0 into low 64-bits */ \
r1 = _mm_shuffle_epi32((XW0), _MM_SHUFFLE(1,0,3,2)); \
/* load high 64-bits of r1 */ \
r1 = _mm_unpacklo_epi64(r1, (XW1)); \
r2 = (XW2); \
\
r0 = _mm_xor_si128(r1, r0); \
r2 = _mm_xor_si128(r3, r2); \
r0 = _mm_xor_si128(r2, r0); \
/* unrotated W[t]..W[t+2] in r0 ... still need W[t+3] */ \
\
r2 = _mm_slli_si128(r0, 12); \
r1 = _mm_cmplt_epi32(r0, _mm_setzero_si128()); \
r0 = _mm_add_epi32(r0, r0); /* shift left by 1 */ \
r0 = _mm_sub_epi32(r0, r1); /* r0 has W[t]..W[t+2] */ \
\
r3 = _mm_srli_epi32(r2, 30); \
r2 = _mm_slli_epi32(r2, 2); \
\
r0 = _mm_xor_si128(r0, r3); \
r0 = _mm_xor_si128(r0, r2); /* r0 now has W[t+3] */ \
\
(XW0) = r0; \
(prep).u128 = _mm_add_epi32(r0, K); \
} while(0)
/*
* SHA-160 F1 Function
*/
inline void F1(uint32_t A, uint32_t& B, uint32_t C, uint32_t D, uint32_t& E, uint32_t msg)
{
E += (D ^ (B & (C ^ D))) + msg + rotl<5>(A);
B = rotl<30>(B);
}
/*
* SHA-160 F2 Function
*/
inline void F2(uint32_t A, uint32_t& B, uint32_t C, uint32_t D, uint32_t& E, uint32_t msg)
{
E += (B ^ C ^ D) + msg + rotl<5>(A);
B = rotl<30>(B);
}
/*
* SHA-160 F3 Function
*/
inline void F3(uint32_t A, uint32_t& B, uint32_t C, uint32_t D, uint32_t& E, uint32_t msg)
{
E += ((B & C) | ((B | C) & D)) + msg + rotl<5>(A);
B = rotl<30>(B);
}
/*
* SHA-160 F4 Function
*/
inline void F4(uint32_t A, uint32_t& B, uint32_t C, uint32_t D, uint32_t& E, uint32_t msg)
{
E += (B ^ C ^ D) + msg + rotl<5>(A);
B = rotl<30>(B);
}
}
}
/*
* SHA-160 Compression Function using SSE for message expansion
*/
//static
BOTAN_FUNC_ISA("sse2")
void SHA_160::sse2_compress_n(secure_vector<uint32_t>& digest, const uint8_t input[], size_t blocks)
{
using namespace SHA1_SSE2_F;
const __m128i K00_19 = _mm_set1_epi32(0x5A827999);
const __m128i K20_39 = _mm_set1_epi32(0x6ED9EBA1);
const __m128i K40_59 = _mm_set1_epi32(0x8F1BBCDC);
const __m128i K60_79 = _mm_set1_epi32(0xCA62C1D6);
uint32_t A = digest[0],
B = digest[1],
C = digest[2],
D = digest[3],
E = digest[4];
const __m128i* input_mm = reinterpret_cast<const __m128i*>(input);
for(size_t i = 0; i != blocks; ++i)
{
union v4si {
uint32_t u32[4];
__m128i u128;
};
v4si P0, P1, P2, P3;
__m128i W0 = _mm_loadu_si128(&input_mm[0]);
prep00_15(P0, W0);
__m128i W1 = _mm_loadu_si128(&input_mm[1]);
prep00_15(P1, W1);
__m128i W2 = _mm_loadu_si128(&input_mm[2]);
prep00_15(P2, W2);
__m128i W3 = _mm_loadu_si128(&input_mm[3]);
prep00_15(P3, W3);
/*
Using SSE4; slower on Core2 and Nehalem
#define GET_P_32(P, i) _mm_extract_epi32(P.u128, i)
Much slower on all tested platforms
#define GET_P_32(P,i) _mm_cvtsi128_si32(_mm_srli_si128(P.u128, i*4))
*/
#define GET_P_32(P, i) P.u32[i]
F1(A, B, C, D, E, GET_P_32(P0, 0));
F1(E, A, B, C, D, GET_P_32(P0, 1));
F1(D, E, A, B, C, GET_P_32(P0, 2));
F1(C, D, E, A, B, GET_P_32(P0, 3));
prep(P0, W0, W1, W2, W3, K00_19);
F1(B, C, D, E, A, GET_P_32(P1, 0));
F1(A, B, C, D, E, GET_P_32(P1, 1));
F1(E, A, B, C, D, GET_P_32(P1, 2));
F1(D, E, A, B, C, GET_P_32(P1, 3));
prep(P1, W1, W2, W3, W0, K20_39);
F1(C, D, E, A, B, GET_P_32(P2, 0));
F1(B, C, D, E, A, GET_P_32(P2, 1));
F1(A, B, C, D, E, GET_P_32(P2, 2));
F1(E, A, B, C, D, GET_P_32(P2, 3));
prep(P2, W2, W3, W0, W1, K20_39);
F1(D, E, A, B, C, GET_P_32(P3, 0));
F1(C, D, E, A, B, GET_P_32(P3, 1));
F1(B, C, D, E, A, GET_P_32(P3, 2));
F1(A, B, C, D, E, GET_P_32(P3, 3));
prep(P3, W3, W0, W1, W2, K20_39);
F1(E, A, B, C, D, GET_P_32(P0, 0));
F1(D, E, A, B, C, GET_P_32(P0, 1));
F1(C, D, E, A, B, GET_P_32(P0, 2));
F1(B, C, D, E, A, GET_P_32(P0, 3));
prep(P0, W0, W1, W2, W3, K20_39);
F2(A, B, C, D, E, GET_P_32(P1, 0));
F2(E, A, B, C, D, GET_P_32(P1, 1));
F2(D, E, A, B, C, GET_P_32(P1, 2));
F2(C, D, E, A, B, GET_P_32(P1, 3));
prep(P1, W1, W2, W3, W0, K20_39);
F2(B, C, D, E, A, GET_P_32(P2, 0));
F2(A, B, C, D, E, GET_P_32(P2, 1));
F2(E, A, B, C, D, GET_P_32(P2, 2));
F2(D, E, A, B, C, GET_P_32(P2, 3));
prep(P2, W2, W3, W0, W1, K40_59);
F2(C, D, E, A, B, GET_P_32(P3, 0));
F2(B, C, D, E, A, GET_P_32(P3, 1));
F2(A, B, C, D, E, GET_P_32(P3, 2));
F2(E, A, B, C, D, GET_P_32(P3, 3));
prep(P3, W3, W0, W1, W2, K40_59);
F2(D, E, A, B, C, GET_P_32(P0, 0));
F2(C, D, E, A, B, GET_P_32(P0, 1));
F2(B, C, D, E, A, GET_P_32(P0, 2));
F2(A, B, C, D, E, GET_P_32(P0, 3));
prep(P0, W0, W1, W2, W3, K40_59);
F2(E, A, B, C, D, GET_P_32(P1, 0));
F2(D, E, A, B, C, GET_P_32(P1, 1));
F2(C, D, E, A, B, GET_P_32(P1, 2));
F2(B, C, D, E, A, GET_P_32(P1, 3));
prep(P1, W1, W2, W3, W0, K40_59);
F3(A, B, C, D, E, GET_P_32(P2, 0));
F3(E, A, B, C, D, GET_P_32(P2, 1));
F3(D, E, A, B, C, GET_P_32(P2, 2));
F3(C, D, E, A, B, GET_P_32(P2, 3));
prep(P2, W2, W3, W0, W1, K40_59);
F3(B, C, D, E, A, GET_P_32(P3, 0));
F3(A, B, C, D, E, GET_P_32(P3, 1));
F3(E, A, B, C, D, GET_P_32(P3, 2));
F3(D, E, A, B, C, GET_P_32(P3, 3));
prep(P3, W3, W0, W1, W2, K60_79);
F3(C, D, E, A, B, GET_P_32(P0, 0));
F3(B, C, D, E, A, GET_P_32(P0, 1));
F3(A, B, C, D, E, GET_P_32(P0, 2));
F3(E, A, B, C, D, GET_P_32(P0, 3));
prep(P0, W0, W1, W2, W3, K60_79);
F3(D, E, A, B, C, GET_P_32(P1, 0));
F3(C, D, E, A, B, GET_P_32(P1, 1));
F3(B, C, D, E, A, GET_P_32(P1, 2));
F3(A, B, C, D, E, GET_P_32(P1, 3));
prep(P1, W1, W2, W3, W0, K60_79);
F3(E, A, B, C, D, GET_P_32(P2, 0));
F3(D, E, A, B, C, GET_P_32(P2, 1));
F3(C, D, E, A, B, GET_P_32(P2, 2));
F3(B, C, D, E, A, GET_P_32(P2, 3));
prep(P2, W2, W3, W0, W1, K60_79);
F4(A, B, C, D, E, GET_P_32(P3, 0));
F4(E, A, B, C, D, GET_P_32(P3, 1));
F4(D, E, A, B, C, GET_P_32(P3, 2));
F4(C, D, E, A, B, GET_P_32(P3, 3));
prep(P3, W3, W0, W1, W2, K60_79);
F4(B, C, D, E, A, GET_P_32(P0, 0));
F4(A, B, C, D, E, GET_P_32(P0, 1));
F4(E, A, B, C, D, GET_P_32(P0, 2));
F4(D, E, A, B, C, GET_P_32(P0, 3));
F4(C, D, E, A, B, GET_P_32(P1, 0));
F4(B, C, D, E, A, GET_P_32(P1, 1));
F4(A, B, C, D, E, GET_P_32(P1, 2));
F4(E, A, B, C, D, GET_P_32(P1, 3));
F4(D, E, A, B, C, GET_P_32(P2, 0));
F4(C, D, E, A, B, GET_P_32(P2, 1));
F4(B, C, D, E, A, GET_P_32(P2, 2));
F4(A, B, C, D, E, GET_P_32(P2, 3));
F4(E, A, B, C, D, GET_P_32(P3, 0));
F4(D, E, A, B, C, GET_P_32(P3, 1));
F4(C, D, E, A, B, GET_P_32(P3, 2));
F4(B, C, D, E, A, GET_P_32(P3, 3));
A = (digest[0] += A);
B = (digest[1] += B);
C = (digest[2] += C);
D = (digest[3] += D);
E = (digest[4] += E);
input_mm += (64 / 16);
}
#undef GET_P_32
}
#undef prep00_15
#undef prep
}
|
