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
|
#define VEC8_ROT(A, IMM) \
_mm256_or_si256(_mm256_slli_epi32(A, IMM), _mm256_srli_epi32(A, (32 - IMM)))
/* same, but replace 2 of the shift/shift/or "rotation" by byte shuffles (8 &
* 16) (better) */
#define VEC8_QUARTERROUND_SHUFFLE(A, B, C, D) \
x_##A = _mm256_add_epi32(x_##A, x_##B); \
t_##A = _mm256_xor_si256(x_##D, x_##A); \
x_##D = _mm256_shuffle_epi8(t_##A, rot16); \
x_##C = _mm256_add_epi32(x_##C, x_##D); \
t_##C = _mm256_xor_si256(x_##B, x_##C); \
x_##B = VEC8_ROT(t_##C, 12); \
x_##A = _mm256_add_epi32(x_##A, x_##B); \
t_##A = _mm256_xor_si256(x_##D, x_##A); \
x_##D = _mm256_shuffle_epi8(t_##A, rot8); \
x_##C = _mm256_add_epi32(x_##C, x_##D); \
t_##C = _mm256_xor_si256(x_##B, x_##C); \
x_##B = VEC8_ROT(t_##C, 7)
#define VEC8_QUARTERROUND(A, B, C, D) VEC8_QUARTERROUND_SHUFFLE(A, B, C, D)
#define VEC8_LINE1(A, B, C, D) \
x_##A = _mm256_add_epi32(x_##A, x_##B); \
x_##D = _mm256_shuffle_epi8(_mm256_xor_si256(x_##D, x_##A), rot16)
#define VEC8_LINE2(A, B, C, D) \
x_##C = _mm256_add_epi32(x_##C, x_##D); \
x_##B = VEC8_ROT(_mm256_xor_si256(x_##B, x_##C), 12)
#define VEC8_LINE3(A, B, C, D) \
x_##A = _mm256_add_epi32(x_##A, x_##B); \
x_##D = _mm256_shuffle_epi8(_mm256_xor_si256(x_##D, x_##A), rot8)
#define VEC8_LINE4(A, B, C, D) \
x_##C = _mm256_add_epi32(x_##C, x_##D); \
x_##B = VEC8_ROT(_mm256_xor_si256(x_##B, x_##C), 7)
#define VEC8_ROUND_SEQ(A1, B1, C1, D1, A2, B2, C2, D2, A3, B3, C3, D3, A4, B4, \
C4, D4) \
VEC8_LINE1(A1, B1, C1, D1); \
VEC8_LINE1(A2, B2, C2, D2); \
VEC8_LINE1(A3, B3, C3, D3); \
VEC8_LINE1(A4, B4, C4, D4); \
VEC8_LINE2(A1, B1, C1, D1); \
VEC8_LINE2(A2, B2, C2, D2); \
VEC8_LINE2(A3, B3, C3, D3); \
VEC8_LINE2(A4, B4, C4, D4); \
VEC8_LINE3(A1, B1, C1, D1); \
VEC8_LINE3(A2, B2, C2, D2); \
VEC8_LINE3(A3, B3, C3, D3); \
VEC8_LINE3(A4, B4, C4, D4); \
VEC8_LINE4(A1, B1, C1, D1); \
VEC8_LINE4(A2, B2, C2, D2); \
VEC8_LINE4(A3, B3, C3, D3); \
VEC8_LINE4(A4, B4, C4, D4)
#define VEC8_ROUND_HALF(A1, B1, C1, D1, A2, B2, C2, D2, A3, B3, C3, D3, A4, \
B4, C4, D4) \
VEC8_LINE1(A1, B1, C1, D1); \
VEC8_LINE1(A2, B2, C2, D2); \
VEC8_LINE2(A1, B1, C1, D1); \
VEC8_LINE2(A2, B2, C2, D2); \
VEC8_LINE3(A1, B1, C1, D1); \
VEC8_LINE3(A2, B2, C2, D2); \
VEC8_LINE4(A1, B1, C1, D1); \
VEC8_LINE4(A2, B2, C2, D2); \
VEC8_LINE1(A3, B3, C3, D3); \
VEC8_LINE1(A4, B4, C4, D4); \
VEC8_LINE2(A3, B3, C3, D3); \
VEC8_LINE2(A4, B4, C4, D4); \
VEC8_LINE3(A3, B3, C3, D3); \
VEC8_LINE3(A4, B4, C4, D4); \
VEC8_LINE4(A3, B3, C3, D3); \
VEC8_LINE4(A4, B4, C4, D4)
#define VEC8_ROUND_HALFANDHALF(A1, B1, C1, D1, A2, B2, C2, D2, A3, B3, C3, D3, \
A4, B4, C4, D4) \
VEC8_LINE1(A1, B1, C1, D1); \
VEC8_LINE1(A2, B2, C2, D2); \
VEC8_LINE2(A1, B1, C1, D1); \
VEC8_LINE2(A2, B2, C2, D2); \
VEC8_LINE1(A3, B3, C3, D3); \
VEC8_LINE1(A4, B4, C4, D4); \
VEC8_LINE2(A3, B3, C3, D3); \
VEC8_LINE2(A4, B4, C4, D4); \
VEC8_LINE3(A1, B1, C1, D1); \
VEC8_LINE3(A2, B2, C2, D2); \
VEC8_LINE4(A1, B1, C1, D1); \
VEC8_LINE4(A2, B2, C2, D2); \
VEC8_LINE3(A3, B3, C3, D3); \
VEC8_LINE3(A4, B4, C4, D4); \
VEC8_LINE4(A3, B3, C3, D3); \
VEC8_LINE4(A4, B4, C4, D4)
#define VEC8_ROUND(A1, B1, C1, D1, A2, B2, C2, D2, A3, B3, C3, D3, A4, B4, C4, \
D4) \
VEC8_ROUND_SEQ(A1, B1, C1, D1, A2, B2, C2, D2, A3, B3, C3, D3, A4, B4, C4, \
D4)
if (bytes >= 512) {
/* constant for shuffling bytes (replacing multiple-of-8 rotates) */
__m256i rot16 =
_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);
__m256i rot8 =
_mm256_set_epi8(14, 13, 12, 15, 10, 9, 8, 11, 6, 5, 4, 7, 2, 1, 0, 3,
14, 13, 12, 15, 10, 9, 8, 11, 6, 5, 4, 7, 2, 1, 0, 3);
uint32_t in12, in13;
/* the naive way seems as fast (if not a bit faster) than the vector way */
__m256i x_0 = _mm256_set1_epi32(x[0]);
__m256i x_1 = _mm256_set1_epi32(x[1]);
__m256i x_2 = _mm256_set1_epi32(x[2]);
__m256i x_3 = _mm256_set1_epi32(x[3]);
__m256i x_4 = _mm256_set1_epi32(x[4]);
__m256i x_5 = _mm256_set1_epi32(x[5]);
__m256i x_6 = _mm256_set1_epi32(x[6]);
__m256i x_7 = _mm256_set1_epi32(x[7]);
__m256i x_8 = _mm256_set1_epi32(x[8]);
__m256i x_9 = _mm256_set1_epi32(x[9]);
__m256i x_10 = _mm256_set1_epi32(x[10]);
__m256i x_11 = _mm256_set1_epi32(x[11]);
__m256i x_12;
__m256i x_13;
__m256i x_14 = _mm256_set1_epi32(x[14]);
__m256i x_15 = _mm256_set1_epi32(x[15]);
__m256i orig0 = x_0;
__m256i orig1 = x_1;
__m256i orig2 = x_2;
__m256i orig3 = x_3;
__m256i orig4 = x_4;
__m256i orig5 = x_5;
__m256i orig6 = x_6;
__m256i orig7 = x_7;
__m256i orig8 = x_8;
__m256i orig9 = x_9;
__m256i orig10 = x_10;
__m256i orig11 = x_11;
__m256i orig12;
__m256i orig13;
__m256i orig14 = x_14;
__m256i orig15 = x_15;
__m256i t_0, t_1, t_2, t_3, t_4, t_5, t_6, t_7, t_8, t_9, t_10, t_11, t_12,
t_13, t_14, t_15;
while (bytes >= 512) {
const __m256i addv12 = _mm256_set_epi64x(3, 2, 1, 0);
const __m256i addv13 = _mm256_set_epi64x(7, 6, 5, 4);
const __m256i permute = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
__m256i t12, t13;
uint64_t in1213;
int i;
x_0 = orig0;
x_1 = orig1;
x_2 = orig2;
x_3 = orig3;
x_4 = orig4;
x_5 = orig5;
x_6 = orig6;
x_7 = orig7;
x_8 = orig8;
x_9 = orig9;
x_10 = orig10;
x_11 = orig11;
x_14 = orig14;
x_15 = orig15;
in12 = x[12];
in13 = x[13];
in1213 = ((uint64_t) in12) | (((uint64_t) in13) << 32);
x_12 = x_13 = _mm256_broadcastq_epi64(_mm_cvtsi64_si128(in1213));
t12 = _mm256_add_epi64(addv12, x_12);
t13 = _mm256_add_epi64(addv13, x_13);
x_12 = _mm256_unpacklo_epi32(t12, t13);
x_13 = _mm256_unpackhi_epi32(t12, t13);
t12 = _mm256_unpacklo_epi32(x_12, x_13);
t13 = _mm256_unpackhi_epi32(x_12, x_13);
/* required because unpack* are intra-lane */
x_12 = _mm256_permutevar8x32_epi32(t12, permute);
x_13 = _mm256_permutevar8x32_epi32(t13, permute);
orig12 = x_12;
orig13 = x_13;
in1213 += 8;
x[12] = in1213 & 0xFFFFFFFF;
x[13] = (in1213 >> 32) & 0xFFFFFFFF;
for (i = 0; i < ROUNDS; i += 2) {
VEC8_ROUND(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15);
VEC8_ROUND(0, 5, 10, 15, 1, 6, 11, 12, 2, 7, 8, 13, 3, 4, 9, 14);
}
#define ONEQUAD_TRANSPOSE(A, B, C, D) \
{ \
__m128i t0, t1, t2, t3; \
x_##A = _mm256_add_epi32(x_##A, orig##A); \
x_##B = _mm256_add_epi32(x_##B, orig##B); \
x_##C = _mm256_add_epi32(x_##C, orig##C); \
x_##D = _mm256_add_epi32(x_##D, orig##D); \
t_##A = _mm256_unpacklo_epi32(x_##A, x_##B); \
t_##B = _mm256_unpacklo_epi32(x_##C, x_##D); \
t_##C = _mm256_unpackhi_epi32(x_##A, x_##B); \
t_##D = _mm256_unpackhi_epi32(x_##C, x_##D); \
x_##A = _mm256_unpacklo_epi64(t_##A, t_##B); \
x_##B = _mm256_unpackhi_epi64(t_##A, t_##B); \
x_##C = _mm256_unpacklo_epi64(t_##C, t_##D); \
x_##D = _mm256_unpackhi_epi64(t_##C, t_##D); \
t0 = _mm_xor_si128(_mm256_extracti128_si256(x_##A, 0), \
_mm_loadu_si128((const __m128i*) (m + 0))); \
_mm_storeu_si128((__m128i*) (c + 0), t0); \
t1 = _mm_xor_si128(_mm256_extracti128_si256(x_##B, 0), \
_mm_loadu_si128((const __m128i*) (m + 64))); \
_mm_storeu_si128((__m128i*) (c + 64), t1); \
t2 = _mm_xor_si128(_mm256_extracti128_si256(x_##C, 0), \
_mm_loadu_si128((const __m128i*) (m + 128))); \
_mm_storeu_si128((__m128i*) (c + 128), t2); \
t3 = _mm_xor_si128(_mm256_extracti128_si256(x_##D, 0), \
_mm_loadu_si128((const __m128i*) (m + 192))); \
_mm_storeu_si128((__m128i*) (c + 192), t3); \
t0 = _mm_xor_si128(_mm256_extracti128_si256(x_##A, 1), \
_mm_loadu_si128((const __m128i*) (m + 256))); \
_mm_storeu_si128((__m128i*) (c + 256), t0); \
t1 = _mm_xor_si128(_mm256_extracti128_si256(x_##B, 1), \
_mm_loadu_si128((const __m128i*) (m + 320))); \
_mm_storeu_si128((__m128i*) (c + 320), t1); \
t2 = _mm_xor_si128(_mm256_extracti128_si256(x_##C, 1), \
_mm_loadu_si128((const __m128i*) (m + 384))); \
_mm_storeu_si128((__m128i*) (c + 384), t2); \
t3 = _mm_xor_si128(_mm256_extracti128_si256(x_##D, 1), \
_mm_loadu_si128((const __m128i*) (m + 448))); \
_mm_storeu_si128((__m128i*) (c + 448), t3); \
}
#define ONEQUAD(A, B, C, D) ONEQUAD_TRANSPOSE(A, B, C, D)
#define ONEQUAD_UNPCK(A, B, C, D) \
{ \
x_##A = _mm256_add_epi32(x_##A, orig##A); \
x_##B = _mm256_add_epi32(x_##B, orig##B); \
x_##C = _mm256_add_epi32(x_##C, orig##C); \
x_##D = _mm256_add_epi32(x_##D, orig##D); \
t_##A = _mm256_unpacklo_epi32(x_##A, x_##B); \
t_##B = _mm256_unpacklo_epi32(x_##C, x_##D); \
t_##C = _mm256_unpackhi_epi32(x_##A, x_##B); \
t_##D = _mm256_unpackhi_epi32(x_##C, x_##D); \
x_##A = _mm256_unpacklo_epi64(t_##A, t_##B); \
x_##B = _mm256_unpackhi_epi64(t_##A, t_##B); \
x_##C = _mm256_unpacklo_epi64(t_##C, t_##D); \
x_##D = _mm256_unpackhi_epi64(t_##C, t_##D); \
}
#define ONEOCTO(A, B, C, D, A2, B2, C2, D2) \
{ \
ONEQUAD_UNPCK(A, B, C, D); \
ONEQUAD_UNPCK(A2, B2, C2, D2); \
t_##A = _mm256_permute2x128_si256(x_##A, x_##A2, 0x20); \
t_##A2 = _mm256_permute2x128_si256(x_##A, x_##A2, 0x31); \
t_##B = _mm256_permute2x128_si256(x_##B, x_##B2, 0x20); \
t_##B2 = _mm256_permute2x128_si256(x_##B, x_##B2, 0x31); \
t_##C = _mm256_permute2x128_si256(x_##C, x_##C2, 0x20); \
t_##C2 = _mm256_permute2x128_si256(x_##C, x_##C2, 0x31); \
t_##D = _mm256_permute2x128_si256(x_##D, x_##D2, 0x20); \
t_##D2 = _mm256_permute2x128_si256(x_##D, x_##D2, 0x31); \
t_##A = _mm256_xor_si256( \
t_##A, _mm256_loadu_si256((const __m256i*) (m + 0))); \
t_##B = _mm256_xor_si256( \
t_##B, _mm256_loadu_si256((const __m256i*) (m + 64))); \
t_##C = _mm256_xor_si256( \
t_##C, _mm256_loadu_si256((const __m256i*) (m + 128))); \
t_##D = _mm256_xor_si256( \
t_##D, _mm256_loadu_si256((const __m256i*) (m + 192))); \
t_##A2 = _mm256_xor_si256( \
t_##A2, _mm256_loadu_si256((const __m256i*) (m + 256))); \
t_##B2 = _mm256_xor_si256( \
t_##B2, _mm256_loadu_si256((const __m256i*) (m + 320))); \
t_##C2 = _mm256_xor_si256( \
t_##C2, _mm256_loadu_si256((const __m256i*) (m + 384))); \
t_##D2 = _mm256_xor_si256( \
t_##D2, _mm256_loadu_si256((const __m256i*) (m + 448))); \
_mm256_storeu_si256((__m256i*) (c + 0), t_##A); \
_mm256_storeu_si256((__m256i*) (c + 64), t_##B); \
_mm256_storeu_si256((__m256i*) (c + 128), t_##C); \
_mm256_storeu_si256((__m256i*) (c + 192), t_##D); \
_mm256_storeu_si256((__m256i*) (c + 256), t_##A2); \
_mm256_storeu_si256((__m256i*) (c + 320), t_##B2); \
_mm256_storeu_si256((__m256i*) (c + 384), t_##C2); \
_mm256_storeu_si256((__m256i*) (c + 448), t_##D2); \
}
ONEOCTO(0, 1, 2, 3, 4, 5, 6, 7);
m += 32;
c += 32;
ONEOCTO(8, 9, 10, 11, 12, 13, 14, 15);
m -= 32;
c -= 32;
#undef ONEQUAD
#undef ONEQUAD_TRANSPOSE
#undef ONEQUAD_UNPCK
#undef ONEOCTO
bytes -= 512;
c += 512;
m += 512;
}
}
#undef VEC8_ROT
#undef VEC8_QUARTERROUND
#undef VEC8_QUARTERROUND_NAIVE
#undef VEC8_QUARTERROUND_SHUFFLE
#undef VEC8_QUARTERROUND_SHUFFLE2
#undef VEC8_LINE1
#undef VEC8_LINE2
#undef VEC8_LINE3
#undef VEC8_LINE4
#undef VEC8_ROUND
#undef VEC8_ROUND_SEQ
#undef VEC8_ROUND_HALF
#undef VEC8_ROUND_HALFANDHALF
|
