/src/botan/src/lib/hash/sha2_32/sha2_32_x86/sha2_32_x86.cpp
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Support for SHA-256 x86 instrinsic |
3 | | * Based on public domain code by Sean Gulley |
4 | | * (https://github.com/mitls/hacl-star/tree/master/experimental/hash) |
5 | | * |
6 | | * Botan is released under the Simplified BSD License (see license.txt) |
7 | | */ |
8 | | |
9 | | #include <botan/internal/sha2_32.h> |
10 | | #include <immintrin.h> |
11 | | |
12 | | namespace Botan { |
13 | | |
14 | | // called from sha2_32.cpp |
15 | | BOTAN_FUNC_ISA("sha,sse4.1,ssse3") |
16 | | void SHA_256::compress_digest_x86(secure_vector<uint32_t>& digest, const uint8_t input[], size_t blocks) |
17 | 0 | { |
18 | 0 | alignas(64) static const uint32_t K[] = { |
19 | 0 | 0x428A2F98, 0x71374491, 0xB5C0FBCF, 0xE9B5DBA5, |
20 | 0 | 0x3956C25B, 0x59F111F1, 0x923F82A4, 0xAB1C5ED5, |
21 | 0 | 0xD807AA98, 0x12835B01, 0x243185BE, 0x550C7DC3, |
22 | 0 | 0x72BE5D74, 0x80DEB1FE, 0x9BDC06A7, 0xC19BF174, |
23 | 0 | 0xE49B69C1, 0xEFBE4786, 0x0FC19DC6, 0x240CA1CC, |
24 | 0 | 0x2DE92C6F, 0x4A7484AA, 0x5CB0A9DC, 0x76F988DA, |
25 | 0 | 0x983E5152, 0xA831C66D, 0xB00327C8, 0xBF597FC7, |
26 | 0 | 0xC6E00BF3, 0xD5A79147, 0x06CA6351, 0x14292967, |
27 | 0 | 0x27B70A85, 0x2E1B2138, 0x4D2C6DFC, 0x53380D13, |
28 | 0 | 0x650A7354, 0x766A0ABB, 0x81C2C92E, 0x92722C85, |
29 | 0 | 0xA2BFE8A1, 0xA81A664B, 0xC24B8B70, 0xC76C51A3, |
30 | 0 | 0xD192E819, 0xD6990624, 0xF40E3585, 0x106AA070, |
31 | 0 | 0x19A4C116, 0x1E376C08, 0x2748774C, 0x34B0BCB5, |
32 | 0 | 0x391C0CB3, 0x4ED8AA4A, 0x5B9CCA4F, 0x682E6FF3, |
33 | 0 | 0x748F82EE, 0x78A5636F, 0x84C87814, 0x8CC70208, |
34 | 0 | 0x90BEFFFA, 0xA4506CEB, 0xBEF9A3F7, 0xC67178F2, |
35 | 0 | }; |
36 | |
|
37 | 0 | const __m128i* K_mm = reinterpret_cast<const __m128i*>(K); |
38 | |
|
39 | 0 | uint32_t* state = &digest[0]; |
40 | |
|
41 | 0 | const __m128i* input_mm = reinterpret_cast<const __m128i*>(input); |
42 | 0 | const __m128i MASK = _mm_set_epi64x(0x0c0d0e0f08090a0b, 0x0405060700010203); |
43 | | |
44 | | // Load initial values |
45 | 0 | __m128i STATE0 = _mm_loadu_si128(reinterpret_cast<__m128i*>(&state[0])); |
46 | 0 | __m128i STATE1 = _mm_loadu_si128(reinterpret_cast<__m128i*>(&state[4])); |
47 | |
|
48 | 0 | STATE0 = _mm_shuffle_epi32(STATE0, 0xB1); // CDAB |
49 | 0 | STATE1 = _mm_shuffle_epi32(STATE1, 0x1B); // EFGH |
50 | |
|
51 | 0 | __m128i TMP = _mm_alignr_epi8(STATE0, STATE1, 8); // ABEF |
52 | 0 | STATE1 = _mm_blend_epi16(STATE1, STATE0, 0xF0); // CDGH |
53 | 0 | STATE0 = TMP; |
54 | |
|
55 | 0 | while(blocks > 0) |
56 | 0 | { |
57 | | // Save current state |
58 | 0 | const __m128i ABEF_SAVE = STATE0; |
59 | 0 | const __m128i CDGH_SAVE = STATE1; |
60 | |
|
61 | 0 | __m128i MSG; |
62 | |
|
63 | 0 | __m128i TMSG0 = _mm_shuffle_epi8(_mm_loadu_si128(input_mm), MASK); |
64 | 0 | __m128i TMSG1 = _mm_shuffle_epi8(_mm_loadu_si128(input_mm + 1), MASK); |
65 | 0 | __m128i TMSG2 = _mm_shuffle_epi8(_mm_loadu_si128(input_mm + 2), MASK); |
66 | 0 | __m128i TMSG3 = _mm_shuffle_epi8(_mm_loadu_si128(input_mm + 3), MASK); |
67 | | |
68 | | // Rounds 0-3 |
69 | 0 | MSG = _mm_add_epi32(TMSG0, _mm_load_si128(K_mm)); |
70 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
71 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
72 | | |
73 | | // Rounds 4-7 |
74 | 0 | MSG = _mm_add_epi32(TMSG1, _mm_load_si128(K_mm + 1)); |
75 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
76 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
77 | |
|
78 | 0 | TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1); |
79 | | |
80 | | // Rounds 8-11 |
81 | 0 | MSG = _mm_add_epi32(TMSG2, _mm_load_si128(K_mm + 2)); |
82 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
83 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
84 | |
|
85 | 0 | TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2); |
86 | | |
87 | | // Rounds 12-15 |
88 | 0 | MSG = _mm_add_epi32(TMSG3, _mm_load_si128(K_mm + 3)); |
89 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
90 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
91 | |
|
92 | 0 | TMSG0 = _mm_add_epi32(TMSG0, _mm_alignr_epi8(TMSG3, TMSG2, 4)); |
93 | 0 | TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3); |
94 | 0 | TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3); |
95 | | |
96 | | // Rounds 16-19 |
97 | 0 | MSG = _mm_add_epi32(TMSG0, _mm_load_si128(K_mm + 4)); |
98 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
99 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
100 | |
|
101 | 0 | TMSG1 = _mm_add_epi32(TMSG1, _mm_alignr_epi8(TMSG0, TMSG3, 4)); |
102 | 0 | TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0); |
103 | 0 | TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0); |
104 | | |
105 | | // Rounds 20-23 |
106 | 0 | MSG = _mm_add_epi32(TMSG1, _mm_load_si128(K_mm + 5)); |
107 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
108 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
109 | |
|
110 | 0 | TMSG2 = _mm_add_epi32(TMSG2, _mm_alignr_epi8(TMSG1, TMSG0, 4)); |
111 | 0 | TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1); |
112 | 0 | TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1); |
113 | | |
114 | | // Rounds 24-27 |
115 | 0 | MSG = _mm_add_epi32(TMSG2, _mm_load_si128(K_mm + 6)); |
116 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
117 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
118 | |
|
119 | 0 | TMSG3 = _mm_add_epi32(TMSG3, _mm_alignr_epi8(TMSG2, TMSG1, 4)); |
120 | 0 | TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2); |
121 | 0 | TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2); |
122 | | |
123 | | // Rounds 28-31 |
124 | 0 | MSG = _mm_add_epi32(TMSG3, _mm_load_si128(K_mm + 7)); |
125 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
126 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
127 | |
|
128 | 0 | TMSG0 = _mm_add_epi32(TMSG0, _mm_alignr_epi8(TMSG3, TMSG2, 4)); |
129 | 0 | TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3); |
130 | 0 | TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3); |
131 | | |
132 | | // Rounds 32-35 |
133 | 0 | MSG = _mm_add_epi32(TMSG0, _mm_load_si128(K_mm + 8)); |
134 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
135 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
136 | |
|
137 | 0 | TMSG1 = _mm_add_epi32(TMSG1, _mm_alignr_epi8(TMSG0, TMSG3, 4)); |
138 | 0 | TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0); |
139 | 0 | TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0); |
140 | | |
141 | | // Rounds 36-39 |
142 | 0 | MSG = _mm_add_epi32(TMSG1, _mm_load_si128(K_mm + 9)); |
143 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
144 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
145 | |
|
146 | 0 | TMSG2 = _mm_add_epi32(TMSG2, _mm_alignr_epi8(TMSG1, TMSG0, 4)); |
147 | 0 | TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1); |
148 | 0 | TMSG0 = _mm_sha256msg1_epu32(TMSG0, TMSG1); |
149 | | |
150 | | // Rounds 40-43 |
151 | 0 | MSG = _mm_add_epi32(TMSG2, _mm_load_si128(K_mm + 10)); |
152 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
153 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
154 | |
|
155 | 0 | TMSG3 = _mm_add_epi32(TMSG3, _mm_alignr_epi8(TMSG2, TMSG1, 4)); |
156 | 0 | TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2); |
157 | 0 | TMSG1 = _mm_sha256msg1_epu32(TMSG1, TMSG2); |
158 | | |
159 | | // Rounds 44-47 |
160 | 0 | MSG = _mm_add_epi32(TMSG3, _mm_load_si128(K_mm + 11)); |
161 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
162 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
163 | |
|
164 | 0 | TMSG0 = _mm_add_epi32(TMSG0, _mm_alignr_epi8(TMSG3, TMSG2, 4)); |
165 | 0 | TMSG0 = _mm_sha256msg2_epu32(TMSG0, TMSG3); |
166 | 0 | TMSG2 = _mm_sha256msg1_epu32(TMSG2, TMSG3); |
167 | | |
168 | | // Rounds 48-51 |
169 | 0 | MSG = _mm_add_epi32(TMSG0, _mm_load_si128(K_mm + 12)); |
170 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
171 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
172 | |
|
173 | 0 | TMSG1 = _mm_add_epi32(TMSG1, _mm_alignr_epi8(TMSG0, TMSG3, 4)); |
174 | 0 | TMSG1 = _mm_sha256msg2_epu32(TMSG1, TMSG0); |
175 | 0 | TMSG3 = _mm_sha256msg1_epu32(TMSG3, TMSG0); |
176 | | |
177 | | // Rounds 52-55 |
178 | 0 | MSG = _mm_add_epi32(TMSG1, _mm_load_si128(K_mm + 13)); |
179 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
180 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
181 | |
|
182 | 0 | TMSG2 = _mm_add_epi32(TMSG2, _mm_alignr_epi8(TMSG1, TMSG0, 4)); |
183 | 0 | TMSG2 = _mm_sha256msg2_epu32(TMSG2, TMSG1); |
184 | | |
185 | | // Rounds 56-59 |
186 | 0 | MSG = _mm_add_epi32(TMSG2, _mm_load_si128(K_mm + 14)); |
187 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
188 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
189 | |
|
190 | 0 | TMSG3 = _mm_add_epi32(TMSG3, _mm_alignr_epi8(TMSG2, TMSG1, 4)); |
191 | 0 | TMSG3 = _mm_sha256msg2_epu32(TMSG3, TMSG2); |
192 | | |
193 | | // Rounds 60-63 |
194 | 0 | MSG = _mm_add_epi32(TMSG3, _mm_load_si128(K_mm + 15)); |
195 | 0 | STATE1 = _mm_sha256rnds2_epu32(STATE1, STATE0, MSG); |
196 | 0 | STATE0 = _mm_sha256rnds2_epu32(STATE0, STATE1, _mm_shuffle_epi32(MSG, 0x0E)); |
197 | | |
198 | | // Add values back to state |
199 | 0 | STATE0 = _mm_add_epi32(STATE0, ABEF_SAVE); |
200 | 0 | STATE1 = _mm_add_epi32(STATE1, CDGH_SAVE); |
201 | |
|
202 | 0 | input_mm += 4; |
203 | 0 | blocks--; |
204 | 0 | } |
205 | |
|
206 | 0 | STATE0 = _mm_shuffle_epi32(STATE0, 0x1B); // FEBA |
207 | 0 | STATE1 = _mm_shuffle_epi32(STATE1, 0xB1); // DCHG |
208 | | |
209 | | // Save state |
210 | 0 | _mm_storeu_si128(reinterpret_cast<__m128i*>(&state[0]), _mm_blend_epi16(STATE0, STATE1, 0xF0)); // DCBA |
211 | 0 | _mm_storeu_si128(reinterpret_cast<__m128i*>(&state[4]), _mm_alignr_epi8(STATE1, STATE0, 8)); // ABEF |
212 | 0 | } |
213 | | |
214 | | } |