/
md5-sse2.c
213 lines (197 loc) · 7.49 KB
/
md5-sse2.c
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
/* code was originally based off OpenSSL's implementation */
#if defined(__XOP__) || defined(__AVX512VL__)
# include <x86intrin.h>
#else
# include <emmintrin.h>
#endif
#define _mm(f) _mm_ ##f
#define _mmi(f) _mm_ ##f## _si128
#ifdef __AVX512VL__
# define F(b,c,d) _mm(ternarylogic_epi32)(b,c,d,0xCA) /*0b11001010*/
# define G(b,c,d) _mm(ternarylogic_epi32)(b,c,d,0xE4) /*0b11100100*/
# define H(b,c,d) _mm(ternarylogic_epi32)(b,c,d,0x96) /*0b10010110*/
# define I(b,c,d) _mm(ternarylogic_epi32)(b,c,d,0x39) /*0b00111001*/
# define ROTATE _mm(rol_epi32)
#else
# ifdef __XOP__
# define F(b,c,d) _mmi(cmov)((c), (d), (b))
# define G(b,c,d) _mmi(cmov)((b), (c), (d))
# else
# define F(b,c,d) _mmi(xor)(_mmi(and)(_mmi(xor)((c), (d)), (b)), (d))
/* using ANDNOT is likely faster: http://www.zorinaq.com/papers/md5-amd64.html */
# define G(b,c,d) _mmi(or)(_mmi(and)((d), (b)), _mmi(andnot)((d), (c)))
/*#define G(b,c,d) F(d, b, c)*/
# endif
# define H(b,c,d) _mmi(xor)(_mmi(xor)((d), (c)), (b))
# define I(b,c,d) _mmi(xor)(_mmi(or)(_mmi(xor)((d), _mm(set1_epi8(0xFF))), (b)), (c))
# if defined(__XOP__) && MD5_SIMD_NUM == 4
# define ROTATE _mm_roti_epi32
# else
// TODO: investigate with SSSE3 byte shuffle
# define ROTATE(a,n) (n == 16 ? \
_mm(shufflehi_epi16)(_mm(shufflelo_epi16)((a), 0xb1), 0xb1) \
: _mmi(or)(_mm(slli_epi32)((a), (n)), _mm(srli_epi32)((a), (32-(n)))) \
)
# endif
#endif
#define RX(f,a,b,c,d,k,s,t) { \
a=_mm(add_epi32)( \
_mm(add_epi32)( \
a, \
_mm(add_epi32)((k),_mm(set1_epi32)(t)) \
), \
f((b),(c),(d)) \
); \
a=ROTATE(a,s); \
a=_mm(add_epi32)(a, b); };
#define TRANSPOSE4(a, b, c, d) { \
MWORD T0 = _mm(unpacklo_epi32)((a), (b)); \
MWORD T1 = _mm(unpackhi_epi32)((a), (b)); \
MWORD T2 = _mm(unpacklo_epi32)((c), (d)); \
MWORD T3 = _mm(unpackhi_epi32)((c), (d)); \
\
(a) = _mm(unpacklo_epi64)(T0, T2); \
(b) = _mm(unpackhi_epi64)(T0, T2); \
(c) = _mm(unpacklo_epi64)(T1, T3); \
(d) = _mm(unpackhi_epi64)(T1, T3); \
}
void md5_update_sse(uint32_t *vals_, const void** data_, size_t num)
{
const MWORD *data0 = (MWORD*)data_[0];
const MWORD *data1 = (MWORD*)data_[1];
const MWORD *data2 = (MWORD*)data_[2];
const MWORD *data3 = (MWORD*)data_[3];
MWORD A, B, C, D;
MWORD oA, oB, oC, oD;
MWORD* vals = (MWORD*)vals_;
# if 1
/* some compilers don't optimise arrays well (i.e. register spills), so use local variables */
MWORD XX0, XX1, XX2, XX3, XX4, XX5, XX6, XX7,
XX8, XX9, XX10, XX11, XX12, XX13, XX14, XX15;
# define X(i) XX##i
# else
MWORD XX[(MD5_BLOCKSIZE/4)];
# define X(i) XX[i]
# endif
/* this may spill too much on 32-bit, consider 64-bit reads? */
/* TODO: enforce alignment? */
#if MD5_SIMD_NUM == 4
#define READ2(a, b)
#define READ4(a, b, c, d) \
X(a) = _mmi(loadu)(data0++); \
X(b) = _mmi(loadu)(data1++); \
X(c) = _mmi(loadu)(data2++); \
X(d) = _mmi(loadu)(data3++); \
\
do TRANSPOSE4(X(a), X(b), X(c), X(d)) while(0)
#elif MD5_SIMD_NUM == 2
#define READ2(a, b) \
X(a) = _mm_loadl_epi64(data0++); \
X(b) = _mm_loadl_epi64(data1++); \
\
X(a) = _mm_unpacklo_epi32(X(a), X(b)); \
X(b) = _mm_slli_si128(X(a), 8)
#define READ4(a, b, c, d)
#else
# error "not defined"
#endif
oA = _mmi(loadu)(vals +0);
oB = _mmi(loadu)(vals +1);
oC = _mmi(loadu)(vals +2);
oD = _mmi(loadu)(vals +3);
while (num--) {
A = oA;
B = oB;
C = oC;
D = oD;
READ4(0, 1, 2, 3);
READ2(0, 1);
/* Round 0 */
RX(F, A, B, C, D, X( 0), 7, 0xd76aa478L);
RX(F, D, A, B, C, X( 1), 12, 0xe8c7b756L);
READ2(2, 3);
RX(F, C, D, A, B, X( 2), 17, 0x242070dbL);
RX(F, B, C, D, A, X( 3), 22, 0xc1bdceeeL);
READ4(4, 5, 6, 7);
READ2(4, 5);
RX(F, A, B, C, D, X( 4), 7, 0xf57c0fafL);
RX(F, D, A, B, C, X( 5), 12, 0x4787c62aL);
READ2(6, 7);
RX(F, C, D, A, B, X( 6), 17, 0xa8304613L);
RX(F, B, C, D, A, X( 7), 22, 0xfd469501L);
READ4( 8, 9, 10, 11);
READ2( 8, 9);
RX(F, A, B, C, D, X( 8), 7, 0x698098d8L);
RX(F, D, A, B, C, X( 9), 12, 0x8b44f7afL);
READ2(10, 11);
RX(F, C, D, A, B, X(10), 17, 0xffff5bb1L);
RX(F, B, C, D, A, X(11), 22, 0x895cd7beL);
READ4(12, 13, 14, 15);
READ2(12, 13);
RX(F, A, B, C, D, X(12), 7, 0x6b901122L);
RX(F, D, A, B, C, X(13), 12, 0xfd987193L);
READ2(14, 15);
RX(F, C, D, A, B, X(14), 17, 0xa679438eL);
RX(F, B, C, D, A, X(15), 22, 0x49b40821L);
/* Round 1 */
RX(G, A, B, C, D, X( 1), 5, 0xf61e2562L);
RX(G, D, A, B, C, X( 6), 9, 0xc040b340L);
RX(G, C, D, A, B, X(11), 14, 0x265e5a51L);
RX(G, B, C, D, A, X( 0), 20, 0xe9b6c7aaL);
RX(G, A, B, C, D, X( 5), 5, 0xd62f105dL);
RX(G, D, A, B, C, X(10), 9, 0x02441453L);
RX(G, C, D, A, B, X(15), 14, 0xd8a1e681L);
RX(G, B, C, D, A, X( 4), 20, 0xe7d3fbc8L);
RX(G, A, B, C, D, X( 9), 5, 0x21e1cde6L);
RX(G, D, A, B, C, X(14), 9, 0xc33707d6L);
RX(G, C, D, A, B, X( 3), 14, 0xf4d50d87L);
RX(G, B, C, D, A, X( 8), 20, 0x455a14edL);
RX(G, A, B, C, D, X(13), 5, 0xa9e3e905L);
RX(G, D, A, B, C, X( 2), 9, 0xfcefa3f8L);
RX(G, C, D, A, B, X( 7), 14, 0x676f02d9L);
RX(G, B, C, D, A, X(12), 20, 0x8d2a4c8aL);
/* Round 2 */
RX(H, A, B, C, D, X( 5), 4, 0xfffa3942L);
RX(H, D, A, B, C, X( 8), 11, 0x8771f681L);
RX(H, C, D, A, B, X(11), 16, 0x6d9d6122L);
RX(H, B, C, D, A, X(14), 23, 0xfde5380cL);
RX(H, A, B, C, D, X( 1), 4, 0xa4beea44L);
RX(H, D, A, B, C, X( 4), 11, 0x4bdecfa9L);
RX(H, C, D, A, B, X( 7), 16, 0xf6bb4b60L);
RX(H, B, C, D, A, X(10), 23, 0xbebfbc70L);
RX(H, A, B, C, D, X(13), 4, 0x289b7ec6L);
RX(H, D, A, B, C, X( 0), 11, 0xeaa127faL);
RX(H, C, D, A, B, X( 3), 16, 0xd4ef3085L);
RX(H, B, C, D, A, X( 6), 23, 0x04881d05L);
RX(H, A, B, C, D, X( 9), 4, 0xd9d4d039L);
RX(H, D, A, B, C, X(12), 11, 0xe6db99e5L);
RX(H, C, D, A, B, X(15), 16, 0x1fa27cf8L);
RX(H, B, C, D, A, X( 2), 23, 0xc4ac5665L);
/* Round 3 */
RX(I, A, B, C, D, X( 0), 6, 0xf4292244L);
RX(I, D, A, B, C, X( 7), 10, 0x432aff97L);
RX(I, C, D, A, B, X(14), 15, 0xab9423a7L);
RX(I, B, C, D, A, X( 5), 21, 0xfc93a039L);
RX(I, A, B, C, D, X(12), 6, 0x655b59c3L);
RX(I, D, A, B, C, X( 3), 10, 0x8f0ccc92L);
RX(I, C, D, A, B, X(10), 15, 0xffeff47dL);
RX(I, B, C, D, A, X( 1), 21, 0x85845dd1L);
RX(I, A, B, C, D, X( 8), 6, 0x6fa87e4fL);
RX(I, D, A, B, C, X(15), 10, 0xfe2ce6e0L);
RX(I, C, D, A, B, X( 6), 15, 0xa3014314L);
RX(I, B, C, D, A, X(13), 21, 0x4e0811a1L);
RX(I, A, B, C, D, X( 4), 6, 0xf7537e82L);
RX(I, D, A, B, C, X(11), 10, 0xbd3af235L);
RX(I, C, D, A, B, X( 2), 15, 0x2ad7d2bbL);
RX(I, B, C, D, A, X( 9), 21, 0xeb86d391L);
oA = _mm(add_epi32)(oA, A);
oB = _mm(add_epi32)(oB, B);
oC = _mm(add_epi32)(oC, C);
oD = _mm(add_epi32)(oD, D);
}
_mmi(storeu)(vals +0, oA);
_mmi(storeu)(vals +1, oB);
_mmi(storeu)(vals +2, oC);
_mmi(storeu)(vals +3, oD);
MMCLEAR
}