Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456
blake3 0.1.1 - Docs.rs
[go: Go Back, main page]

blake3 0.1.1

the BLAKE3 hash function
Documentation 
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

#include "blake3_impl.h"

#include <immintrin.h>

#define DEGREE 8

INLINE __m256i loadu(const uint8_t src[32]) {
  return _mm256_loadu_si256((const __m256i *)src);
}

INLINE void storeu(__m256i src, uint8_t dest[16]) {
  _mm256_storeu_si256((__m256i *)dest, src);
}

INLINE __m256i addv(__m256i a, __m256i b) { return _mm256_add_epi32(a, b); }

// Note that clang-format doesn't like the name "xor" for some reason.
INLINE __m256i xorv(__m256i a, __m256i b) { return _mm256_xor_si256(a, b); }

INLINE __m256i set1(uint32_t x) { return _mm256_set1_epi32((int32_t)x); }

INLINE __m256i set8(uint32_t a, uint32_t b, uint32_t c, uint32_t d, uint32_t e,
                    uint32_t f, uint32_t g, uint32_t h) {
  return _mm256_setr_epi32((int32_t)a, (int32_t)b, (int32_t)c, (int32_t)d,
                           (int32_t)e, (int32_t)f, (int32_t)g, (int32_t)h);
}

INLINE __m256i rot16(__m256i x) {
  return _mm256_shuffle_epi8(
      x, _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));
}

INLINE __m256i rot12(__m256i x) {
  return xorv(_mm256_srli_epi32(x, 12), _mm256_slli_epi32(x, 32 - 12));
}

INLINE __m256i rot8(__m256i x) {
  return _mm256_shuffle_epi8(
      x, _mm256_set_epi8(12, 15, 14, 13, 8, 11, 10, 9, 4, 7, 6, 5, 0, 3, 2, 1,
                         12, 15, 14, 13, 8, 11, 10, 9, 4, 7, 6, 5, 0, 3, 2, 1));
}

INLINE __m256i rot7(__m256i x) {
  return xorv(_mm256_srli_epi32(x, 7), _mm256_slli_epi32(x, 32 - 7));
}

INLINE void round_fn(__m256i v[16], __m256i m[16], size_t r) {
  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][0]]);
  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][2]]);
  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][4]]);
  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][6]]);
  v[0] = addv(v[0], v[4]);
  v[1] = addv(v[1], v[5]);
  v[2] = addv(v[2], v[6]);
  v[3] = addv(v[3], v[7]);
  v[12] = xorv(v[12], v[0]);
  v[13] = xorv(v[13], v[1]);
  v[14] = xorv(v[14], v[2]);
  v[15] = xorv(v[15], v[3]);
  v[12] = rot16(v[12]);
  v[13] = rot16(v[13]);
  v[14] = rot16(v[14]);
  v[15] = rot16(v[15]);
  v[8] = addv(v[8], v[12]);
  v[9] = addv(v[9], v[13]);
  v[10] = addv(v[10], v[14]);
  v[11] = addv(v[11], v[15]);
  v[4] = xorv(v[4], v[8]);
  v[5] = xorv(v[5], v[9]);
  v[6] = xorv(v[6], v[10]);
  v[7] = xorv(v[7], v[11]);
  v[4] = rot12(v[4]);
  v[5] = rot12(v[5]);
  v[6] = rot12(v[6]);
  v[7] = rot12(v[7]);
  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][1]]);
  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][3]]);
  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][5]]);
  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][7]]);
  v[0] = addv(v[0], v[4]);
  v[1] = addv(v[1], v[5]);
  v[2] = addv(v[2], v[6]);
  v[3] = addv(v[3], v[7]);
  v[12] = xorv(v[12], v[0]);
  v[13] = xorv(v[13], v[1]);
  v[14] = xorv(v[14], v[2]);
  v[15] = xorv(v[15], v[3]);
  v[12] = rot8(v[12]);
  v[13] = rot8(v[13]);
  v[14] = rot8(v[14]);
  v[15] = rot8(v[15]);
  v[8] = addv(v[8], v[12]);
  v[9] = addv(v[9], v[13]);
  v[10] = addv(v[10], v[14]);
  v[11] = addv(v[11], v[15]);
  v[4] = xorv(v[4], v[8]);
  v[5] = xorv(v[5], v[9]);
  v[6] = xorv(v[6], v[10]);
  v[7] = xorv(v[7], v[11]);
  v[4] = rot7(v[4]);
  v[5] = rot7(v[5]);
  v[6] = rot7(v[6]);
  v[7] = rot7(v[7]);

  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][8]]);
  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][10]]);
  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][12]]);
  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][14]]);
  v[0] = addv(v[0], v[5]);
  v[1] = addv(v[1], v[6]);
  v[2] = addv(v[2], v[7]);
  v[3] = addv(v[3], v[4]);
  v[15] = xorv(v[15], v[0]);
  v[12] = xorv(v[12], v[1]);
  v[13] = xorv(v[13], v[2]);
  v[14] = xorv(v[14], v[3]);
  v[15] = rot16(v[15]);
  v[12] = rot16(v[12]);
  v[13] = rot16(v[13]);
  v[14] = rot16(v[14]);
  v[10] = addv(v[10], v[15]);
  v[11] = addv(v[11], v[12]);
  v[8] = addv(v[8], v[13]);
  v[9] = addv(v[9], v[14]);
  v[5] = xorv(v[5], v[10]);
  v[6] = xorv(v[6], v[11]);
  v[7] = xorv(v[7], v[8]);
  v[4] = xorv(v[4], v[9]);
  v[5] = rot12(v[5]);
  v[6] = rot12(v[6]);
  v[7] = rot12(v[7]);
  v[4] = rot12(v[4]);
  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][9]]);
  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][11]]);
  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][13]]);
  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][15]]);
  v[0] = addv(v[0], v[5]);
  v[1] = addv(v[1], v[6]);
  v[2] = addv(v[2], v[7]);
  v[3] = addv(v[3], v[4]);
  v[15] = xorv(v[15], v[0]);
  v[12] = xorv(v[12], v[1]);
  v[13] = xorv(v[13], v[2]);
  v[14] = xorv(v[14], v[3]);
  v[15] = rot8(v[15]);
  v[12] = rot8(v[12]);
  v[13] = rot8(v[13]);
  v[14] = rot8(v[14]);
  v[10] = addv(v[10], v[15]);
  v[11] = addv(v[11], v[12]);
  v[8] = addv(v[8], v[13]);
  v[9] = addv(v[9], v[14]);
  v[5] = xorv(v[5], v[10]);
  v[6] = xorv(v[6], v[11]);
  v[7] = xorv(v[7], v[8]);
  v[4] = xorv(v[4], v[9]);
  v[5] = rot7(v[5]);
  v[6] = rot7(v[6]);
  v[7] = rot7(v[7]);
  v[4] = rot7(v[4]);
}

INLINE void transpose_vecs(__m256i vecs[DEGREE]) {
  // Interleave 32-bit lanes. The low unpack is lanes 00/11/44/55, and the high
  // is 22/33/66/77.
  __m256i ab_0145 = _mm256_unpacklo_epi32(vecs[0], vecs[1]);
  __m256i ab_2367 = _mm256_unpackhi_epi32(vecs[0], vecs[1]);
  __m256i cd_0145 = _mm256_unpacklo_epi32(vecs[2], vecs[3]);
  __m256i cd_2367 = _mm256_unpackhi_epi32(vecs[2], vecs[3]);
  __m256i ef_0145 = _mm256_unpacklo_epi32(vecs[4], vecs[5]);
  __m256i ef_2367 = _mm256_unpackhi_epi32(vecs[4], vecs[5]);
  __m256i gh_0145 = _mm256_unpacklo_epi32(vecs[6], vecs[7]);
  __m256i gh_2367 = _mm256_unpackhi_epi32(vecs[6], vecs[7]);

  // Interleave 64-bit lates. The low unpack is lanes 00/22 and the high is
  // 11/33.
  __m256i abcd_04 = _mm256_unpacklo_epi64(ab_0145, cd_0145);
  __m256i abcd_15 = _mm256_unpackhi_epi64(ab_0145, cd_0145);
  __m256i abcd_26 = _mm256_unpacklo_epi64(ab_2367, cd_2367);
  __m256i abcd_37 = _mm256_unpackhi_epi64(ab_2367, cd_2367);
  __m256i efgh_04 = _mm256_unpacklo_epi64(ef_0145, gh_0145);
  __m256i efgh_15 = _mm256_unpackhi_epi64(ef_0145, gh_0145);
  __m256i efgh_26 = _mm256_unpacklo_epi64(ef_2367, gh_2367);
  __m256i efgh_37 = _mm256_unpackhi_epi64(ef_2367, gh_2367);

  // Interleave 128-bit lanes.
  vecs[0] = _mm256_permute2x128_si256(abcd_04, efgh_04, 0x20);
  vecs[1] = _mm256_permute2x128_si256(abcd_15, efgh_15, 0x20);
  vecs[2] = _mm256_permute2x128_si256(abcd_26, efgh_26, 0x20);
  vecs[3] = _mm256_permute2x128_si256(abcd_37, efgh_37, 0x20);
  vecs[4] = _mm256_permute2x128_si256(abcd_04, efgh_04, 0x31);
  vecs[5] = _mm256_permute2x128_si256(abcd_15, efgh_15, 0x31);
  vecs[6] = _mm256_permute2x128_si256(abcd_26, efgh_26, 0x31);
  vecs[7] = _mm256_permute2x128_si256(abcd_37, efgh_37, 0x31);
}

INLINE void transpose_msg_vecs(const uint8_t *const *inputs,
                               size_t block_offset, __m256i out[16]) {
  out[0] = loadu(&inputs[0][block_offset + 0 * sizeof(__m256i)]);
  out[1] = loadu(&inputs[1][block_offset + 0 * sizeof(__m256i)]);
  out[2] = loadu(&inputs[2][block_offset + 0 * sizeof(__m256i)]);
  out[3] = loadu(&inputs[3][block_offset + 0 * sizeof(__m256i)]);
  out[4] = loadu(&inputs[4][block_offset + 0 * sizeof(__m256i)]);
  out[5] = loadu(&inputs[5][block_offset + 0 * sizeof(__m256i)]);
  out[6] = loadu(&inputs[6][block_offset + 0 * sizeof(__m256i)]);
  out[7] = loadu(&inputs[7][block_offset + 0 * sizeof(__m256i)]);
  out[8] = loadu(&inputs[0][block_offset + 1 * sizeof(__m256i)]);
  out[9] = loadu(&inputs[1][block_offset + 1 * sizeof(__m256i)]);
  out[10] = loadu(&inputs[2][block_offset + 1 * sizeof(__m256i)]);
  out[11] = loadu(&inputs[3][block_offset + 1 * sizeof(__m256i)]);
  out[12] = loadu(&inputs[4][block_offset + 1 * sizeof(__m256i)]);
  out[13] = loadu(&inputs[5][block_offset + 1 * sizeof(__m256i)]);
  out[14] = loadu(&inputs[6][block_offset + 1 * sizeof(__m256i)]);
  out[15] = loadu(&inputs[7][block_offset + 1 * sizeof(__m256i)]);
  transpose_vecs(&out[0]);
  transpose_vecs(&out[8]);
}

INLINE void load_counters(uint64_t counter, bool increment_counter,
                          __m256i *out_low, __m256i *out_high) {
  uint64_t mask = (increment_counter ? ~0 : 0);
  *out_low = set8(
      counter_low(counter + (mask & 0)), counter_low(counter + (mask & 1)),
      counter_low(counter + (mask & 2)), counter_low(counter + (mask & 3)),
      counter_low(counter + (mask & 4)), counter_low(counter + (mask & 5)),
      counter_low(counter + (mask & 6)), counter_low(counter + (mask & 7)));
  *out_high = set8(
      counter_high(counter + (mask & 0)), counter_high(counter + (mask & 1)),
      counter_high(counter + (mask & 2)), counter_high(counter + (mask & 3)),
      counter_high(counter + (mask & 4)), counter_high(counter + (mask & 5)),
      counter_high(counter + (mask & 6)), counter_high(counter + (mask & 7)));
}

void blake3_hash8_avx2(const uint8_t *const *inputs, size_t blocks,
                       const uint32_t key[8], uint64_t counter,
                       bool increment_counter, uint8_t flags,
                       uint8_t flags_start, uint8_t flags_end, uint8_t *out) {
  __m256i h_vecs[8] = {
      set1(key[0]), set1(key[1]), set1(key[2]), set1(key[3]),
      set1(key[4]), set1(key[5]), set1(key[6]), set1(key[7]),
  };
  __m256i counter_low_vec, counter_high_vec;
  load_counters(counter, increment_counter, &counter_low_vec,
                &counter_high_vec);
  uint8_t block_flags = flags | flags_start;

  for (size_t block = 0; block < blocks; block++) {
    if (block + 1 == blocks) {
      block_flags |= flags_end;
    }
    __m256i block_len_vec = set1(BLAKE3_BLOCK_LEN);
    __m256i block_flags_vec = set1(block_flags);
    __m256i msg_vecs[16];
    transpose_msg_vecs(inputs, block * BLAKE3_BLOCK_LEN, msg_vecs);

    __m256i v[16] = {
        h_vecs[0],       h_vecs[1],        h_vecs[2],     h_vecs[3],
        h_vecs[4],       h_vecs[5],        h_vecs[6],     h_vecs[7],
        set1(IV[0]),     set1(IV[1]),      set1(IV[2]),   set1(IV[3]),
        counter_low_vec, counter_high_vec, block_len_vec, block_flags_vec,
    };
    round_fn(v, msg_vecs, 0);
    round_fn(v, msg_vecs, 1);
    round_fn(v, msg_vecs, 2);
    round_fn(v, msg_vecs, 3);
    round_fn(v, msg_vecs, 4);
    round_fn(v, msg_vecs, 5);
    round_fn(v, msg_vecs, 6);
    h_vecs[0] = xorv(v[0], v[8]);
    h_vecs[1] = xorv(v[1], v[9]);
    h_vecs[2] = xorv(v[2], v[10]);
    h_vecs[3] = xorv(v[3], v[11]);
    h_vecs[4] = xorv(v[4], v[12]);
    h_vecs[5] = xorv(v[5], v[13]);
    h_vecs[6] = xorv(v[6], v[14]);
    h_vecs[7] = xorv(v[7], v[15]);

    block_flags = flags;
  }

  transpose_vecs(h_vecs);
  storeu(h_vecs[0], &out[0 * sizeof(__m256i)]);
  storeu(h_vecs[1], &out[1 * sizeof(__m256i)]);
  storeu(h_vecs[2], &out[2 * sizeof(__m256i)]);
  storeu(h_vecs[3], &out[3 * sizeof(__m256i)]);
  storeu(h_vecs[4], &out[4 * sizeof(__m256i)]);
  storeu(h_vecs[5], &out[5 * sizeof(__m256i)]);
  storeu(h_vecs[6], &out[6 * sizeof(__m256i)]);
  storeu(h_vecs[7], &out[7 * sizeof(__m256i)]);
}

void blake3_hash_many_sse41(const uint8_t *const *inputs, size_t num_inputs,
                            size_t blocks, const uint32_t key[8],
                            uint64_t counter, bool increment_counter,
                            uint8_t flags, uint8_t flags_start,
                            uint8_t flags_end, uint8_t *out);

void blake3_hash_many_avx2(const uint8_t *const *inputs, size_t num_inputs,
                           size_t blocks, const uint32_t key[8],
                           uint64_t counter, bool increment_counter,
                           uint8_t flags, uint8_t flags_start,
                           uint8_t flags_end, uint8_t *out) {
  while (num_inputs >= DEGREE) {
    blake3_hash8_avx2(inputs, blocks, key, counter, increment_counter, flags,
                      flags_start, flags_end, out);
    if (increment_counter) {
      counter += DEGREE;
    }
    inputs += DEGREE;
    num_inputs -= DEGREE;
    out = &out[DEGREE * BLAKE3_OUT_LEN];
  }
  blake3_hash_many_sse41(inputs, num_inputs, blocks, key, counter,
                         increment_counter, flags, flags_start, flags_end, out);
}