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
|
/*
* IDEA in SSE2
* (C) 2009 Jack Lloyd
*
* Distributed under the terms of the Botan license
*/
#include <botan/idea_sse2.h>
#include <emmintrin.h>
namespace Botan {
namespace {
inline __m128i mul(__m128i X, u16bit K_16)
{
const __m128i zeros = _mm_set1_epi16(0);
const __m128i ones = _mm_set1_epi16(1);
const __m128i high_bit = _mm_set1_epi16(0x8000);
const __m128i K = _mm_set1_epi16(K_16);
const __m128i X_is_zero = _mm_cmpeq_epi16(X, zeros);
const __m128i K_is_zero = _mm_cmpeq_epi16(K, zeros);
const __m128i mul_lo = _mm_mullo_epi16(X, K);
const __m128i mul_hi = _mm_mulhi_epu16(X, K);
__m128i T = _mm_sub_epi16(mul_lo, mul_hi);
// Unsigned compare; cmp = 1 if mul_lo < mul_hi else 0
const __m128i cmp = _mm_srli_epi16(_mm_cmpgt_epi16(
_mm_add_epi16(mul_hi, high_bit),
_mm_add_epi16(mul_lo, high_bit)),
15);
T = _mm_add_epi16(T, cmp);
/* Selection: if X[i] is zero then assign 1-K
if K is zero then assign 1-X[i]
Could if() off value of K_16 for the second, but this gives a
constant time implementation which is a nice bonus.
*/
T = _mm_or_si128(
_mm_andnot_si128(X_is_zero, T),
_mm_and_si128(_mm_sub_epi16(ones, K), X_is_zero));
T = _mm_or_si128(
_mm_andnot_si128(K_is_zero, T),
_mm_and_si128(_mm_sub_epi16(ones, X), K_is_zero));
return T;
}
/*
* 4x8 matrix transpose
*
* FIXME: why do I need the extra set of unpack_epi32 here? Inverse in
* transpose_out doesn't need it. Something with the shuffle? Removing
* that extra unpack could easily save 3-4 cycles per block, and would
* also help a lot with register pressure on 32-bit x86
*/
void transpose_in(__m128i& B0, __m128i& B1, __m128i& B2, __m128i& B3)
{
__m128i T0 = _mm_unpackhi_epi32(B0, B1);
__m128i T1 = _mm_unpacklo_epi32(B0, B1);
__m128i T2 = _mm_unpackhi_epi32(B2, B3);
__m128i T3 = _mm_unpacklo_epi32(B2, B3);
__m128i T4 = _mm_unpacklo_epi32(T0, T1);
__m128i T5 = _mm_unpackhi_epi32(T0, T1);
__m128i T6 = _mm_unpacklo_epi32(T2, T3);
__m128i T7 = _mm_unpackhi_epi32(T2, T3);
T0 = _mm_shufflehi_epi16(T4, _MM_SHUFFLE(1, 3, 0, 2));
T1 = _mm_shufflehi_epi16(T5, _MM_SHUFFLE(1, 3, 0, 2));
T2 = _mm_shufflehi_epi16(T6, _MM_SHUFFLE(1, 3, 0, 2));
T3 = _mm_shufflehi_epi16(T7, _MM_SHUFFLE(1, 3, 0, 2));
T0 = _mm_shufflelo_epi16(T0, _MM_SHUFFLE(1, 3, 0, 2));
T1 = _mm_shufflelo_epi16(T1, _MM_SHUFFLE(1, 3, 0, 2));
T2 = _mm_shufflelo_epi16(T2, _MM_SHUFFLE(1, 3, 0, 2));
T3 = _mm_shufflelo_epi16(T3, _MM_SHUFFLE(1, 3, 0, 2));
T0 = _mm_shuffle_epi32(T0, _MM_SHUFFLE(3, 1, 2, 0));
T1 = _mm_shuffle_epi32(T1, _MM_SHUFFLE(3, 1, 2, 0));
T2 = _mm_shuffle_epi32(T2, _MM_SHUFFLE(3, 1, 2, 0));
T3 = _mm_shuffle_epi32(T3, _MM_SHUFFLE(3, 1, 2, 0));
B0 = _mm_unpacklo_epi64(T0, T2);
B1 = _mm_unpackhi_epi64(T0, T2);
B2 = _mm_unpacklo_epi64(T1, T3);
B3 = _mm_unpackhi_epi64(T1, T3);
}
/*
* 4x8 matrix transpose (reverse)
*/
void transpose_out(__m128i& B0, __m128i& B1, __m128i& B2, __m128i& B3)
{
__m128i T0 = _mm_unpacklo_epi64(B0, B1);
__m128i T1 = _mm_unpacklo_epi64(B2, B3);
__m128i T2 = _mm_unpackhi_epi64(B0, B1);
__m128i T3 = _mm_unpackhi_epi64(B2, B3);
T0 = _mm_shuffle_epi32(T0, _MM_SHUFFLE(3, 1, 2, 0));
T1 = _mm_shuffle_epi32(T1, _MM_SHUFFLE(3, 1, 2, 0));
T2 = _mm_shuffle_epi32(T2, _MM_SHUFFLE(3, 1, 2, 0));
T3 = _mm_shuffle_epi32(T3, _MM_SHUFFLE(3, 1, 2, 0));
T0 = _mm_shufflehi_epi16(T0, _MM_SHUFFLE(3, 1, 2, 0));
T1 = _mm_shufflehi_epi16(T1, _MM_SHUFFLE(3, 1, 2, 0));
T2 = _mm_shufflehi_epi16(T2, _MM_SHUFFLE(3, 1, 2, 0));
T3 = _mm_shufflehi_epi16(T3, _MM_SHUFFLE(3, 1, 2, 0));
T0 = _mm_shufflelo_epi16(T0, _MM_SHUFFLE(3, 1, 2, 0));
T1 = _mm_shufflelo_epi16(T1, _MM_SHUFFLE(3, 1, 2, 0));
T2 = _mm_shufflelo_epi16(T2, _MM_SHUFFLE(3, 1, 2, 0));
T3 = _mm_shufflelo_epi16(T3, _MM_SHUFFLE(3, 1, 2, 0));
B0 = _mm_unpacklo_epi32(T0, T1);
B1 = _mm_unpackhi_epi32(T0, T1);
B2 = _mm_unpacklo_epi32(T2, T3);
B3 = _mm_unpackhi_epi32(T2, T3);
}
/*
* IDEA encryption/decryption in SSE2
*/
void idea_op_8(const byte in[64], byte out[64], const u16bit EK[52])
{
__m128i B0 = _mm_loadu_si128((const __m128i*)in);
__m128i B1 = _mm_loadu_si128((const __m128i*)in + 1);
__m128i B2 = _mm_loadu_si128((const __m128i*)in + 2);
__m128i B3 = _mm_loadu_si128((const __m128i*)in + 3);
transpose_in(B0, B1, B2, B3);
// byte swap
B0 = _mm_or_si128(_mm_slli_epi16(B0, 8), _mm_srli_epi16(B0, 8));
B1 = _mm_or_si128(_mm_slli_epi16(B1, 8), _mm_srli_epi16(B1, 8));
B2 = _mm_or_si128(_mm_slli_epi16(B2, 8), _mm_srli_epi16(B2, 8));
B3 = _mm_or_si128(_mm_slli_epi16(B3, 8), _mm_srli_epi16(B3, 8));
for(size_t i = 0; i != 8; ++i)
{
B0 = mul(B0, EK[6*i+0]);
B1 = _mm_add_epi16(B1, _mm_set1_epi16(EK[6*i+1]));
B2 = _mm_add_epi16(B2, _mm_set1_epi16(EK[6*i+2]));
B3 = mul(B3, EK[6*i+3]);
__m128i T0 = B2;
B2 = _mm_xor_si128(B2, B0);
B2 = mul(B2, EK[6*i+4]);
__m128i T1 = B1;
B1 = _mm_xor_si128(B1, B3);
B1 = _mm_add_epi16(B1, B2);
B1 = mul(B1, EK[6*i+5]);
B2 = _mm_add_epi16(B2, B1);
B0 = _mm_xor_si128(B0, B1);
B1 = _mm_xor_si128(B1, T0);
B3 = _mm_xor_si128(B3, B2);
B2 = _mm_xor_si128(B2, T1);
}
B0 = mul(B0, EK[48]);
B1 = _mm_add_epi16(B1, _mm_set1_epi16(EK[50]));
B2 = _mm_add_epi16(B2, _mm_set1_epi16(EK[49]));
B3 = mul(B3, EK[51]);
// byte swap
B0 = _mm_or_si128(_mm_slli_epi16(B0, 8), _mm_srli_epi16(B0, 8));
B1 = _mm_or_si128(_mm_slli_epi16(B1, 8), _mm_srli_epi16(B1, 8));
B2 = _mm_or_si128(_mm_slli_epi16(B2, 8), _mm_srli_epi16(B2, 8));
B3 = _mm_or_si128(_mm_slli_epi16(B3, 8), _mm_srli_epi16(B3, 8));
transpose_out(B0, B2, B1, B3);
_mm_storeu_si128((__m128i*)out, B0);
_mm_storeu_si128((__m128i*)out + 1, B2);
_mm_storeu_si128((__m128i*)out + 2, B1);
_mm_storeu_si128((__m128i*)out + 3, B3);
}
}
/*
* IDEA Encryption
*/
void IDEA_SSE2::encrypt_n(const byte in[], byte out[], size_t blocks) const
{
const u16bit* KS = &this->get_EK()[0];
while(blocks >= 8)
{
idea_op_8(in, out, KS);
in += 8 * BLOCK_SIZE;
out += 8 * BLOCK_SIZE;
blocks -= 8;
}
if(blocks)
IDEA::encrypt_n(in, out, blocks);
}
/*
* IDEA Decryption
*/
void IDEA_SSE2::decrypt_n(const byte in[], byte out[], size_t blocks) const
{
const u16bit* KS = &this->get_DK()[0];
while(blocks >= 8)
{
idea_op_8(in, out, KS);
in += 8 * BLOCK_SIZE;
out += 8 * BLOCK_SIZE;
blocks -= 8;
}
if(blocks)
IDEA::decrypt_n(in, out, blocks);
}
}
|
