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
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
|
use crate::{portable, CVWords, IncrementCounter, BLOCK_LEN};
use arrayref::{array_mut_ref, array_ref};
#[cfg(feature = "c_avx512")]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
use crate::c_avx512;
#[cfg(feature = "c_neon")]
use crate::c_neon;
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
use crate::{avx2, sse41};
cfg_if::cfg_if! {
if #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] {
cfg_if::cfg_if! {
if #[cfg(feature = "c_avx512")] {
pub const MAX_SIMD_DEGREE: usize = 16;
} else {
pub const MAX_SIMD_DEGREE: usize = 8;
}
}
} else if #[cfg(feature = "c_neon")] {
pub const MAX_SIMD_DEGREE: usize = 4;
} else {
pub const MAX_SIMD_DEGREE: usize = 1;
}
}
// There are some places where we want a static size that's equal to the
// MAX_SIMD_DEGREE, but also at least 2. Constant contexts aren't currently
// allowed to use cmp::max, so we have to hardcode this additional constant
// value. Get rid of this once cmp::max is a const fn.
cfg_if::cfg_if! {
if #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] {
cfg_if::cfg_if! {
if #[cfg(feature = "c_avx512")] {
pub const MAX_SIMD_DEGREE_OR_2: usize = 16;
} else {
pub const MAX_SIMD_DEGREE_OR_2: usize = 8;
}
}
} else if #[cfg(feature = "c_neon")] {
pub const MAX_SIMD_DEGREE_OR_2: usize = 4;
} else {
pub const MAX_SIMD_DEGREE_OR_2: usize = 2;
}
}
#[derive(Clone, Copy, Debug)]
pub enum Platform {
Portable,
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
SSE41,
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
AVX2,
#[cfg(feature = "c_avx512")]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
AVX512,
#[cfg(feature = "c_neon")]
NEON,
}
impl Platform {
#[allow(unreachable_code)]
pub fn detect() -> Self {
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
{
#[cfg(feature = "c_avx512")]
{
if avx512_detected() {
return Platform::AVX512;
}
}
if avx2_detected() {
return Platform::AVX2;
}
if sse41_detected() {
return Platform::SSE41;
}
}
// We don't use dynamic feature detection for NEON. If the "c_neon"
// feature is on, NEON is assumed to be supported.
#[cfg(feature = "c_neon")]
{
return Platform::NEON;
}
Platform::Portable
}
pub fn simd_degree(&self) -> usize {
let degree = match self {
Platform::Portable => 1,
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::SSE41 => 4,
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::AVX2 => 8,
#[cfg(feature = "c_avx512")]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::AVX512 => 16,
#[cfg(feature = "c_neon")]
Platform::NEON => 4,
};
debug_assert!(degree <= MAX_SIMD_DEGREE);
degree
}
pub(crate) fn compress_in_place(
&self,
cv: &mut CVWords,
block: &[u8; BLOCK_LEN],
block_len: u8,
counter: u64,
flags: u8,
) {
match self {
Platform::Portable => portable::compress_in_place(cv, block, block_len, counter, flags),
// Safe because detect() checked for platform support.
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::SSE41 | Platform::AVX2 => unsafe {
sse41::compress_in_place(cv, block, block_len, counter, flags)
},
// Safe because detect() checked for platform support.
#[cfg(feature = "c_avx512")]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::AVX512 => unsafe {
c_avx512::compress_in_place(cv, block, block_len, counter, flags)
},
// No NEON compress_in_place() implementation yet.
#[cfg(feature = "c_neon")]
Platform::NEON => portable::compress_in_place(cv, block, block_len, counter, flags),
}
}
pub(crate) fn compress_xof(
&self,
cv: &CVWords,
block: &[u8; BLOCK_LEN],
block_len: u8,
counter: u64,
flags: u8,
) -> [u8; 64] {
match self {
Platform::Portable => portable::compress_xof(cv, block, block_len, counter, flags),
// Safe because detect() checked for platform support.
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::SSE41 | Platform::AVX2 => unsafe {
sse41::compress_xof(cv, block, block_len, counter, flags)
},
// Safe because detect() checked for platform support.
#[cfg(feature = "c_avx512")]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::AVX512 => unsafe {
c_avx512::compress_xof(cv, block, block_len, counter, flags)
},
// No NEON compress_xof() implementation yet.
#[cfg(feature = "c_neon")]
Platform::NEON => portable::compress_xof(cv, block, block_len, counter, flags),
}
}
// IMPLEMENTATION NOTE
// ===================
// hash_many() applies two optimizations. The critically important
// optimization is the high-performance parallel SIMD hashing mode,
// described in detail in the spec. This more than doubles throughput per
// thread. Another optimization is keeping the state vectors transposed
// from block to block within a chunk. When state vectors are transposed
// after every block, there's a small but measurable performance loss.
// Compressing chunks with a dedicated loop avoids this.
pub(crate) fn hash_many<A: arrayvec::Array<Item = u8>>(
&self,
inputs: &[&A],
key: &CVWords,
counter: u64,
increment_counter: IncrementCounter,
flags: u8,
flags_start: u8,
flags_end: u8,
out: &mut [u8],
) {
match self {
Platform::Portable => portable::hash_many(
inputs,
key,
counter,
increment_counter,
flags,
flags_start,
flags_end,
out,
),
// Safe because detect() checked for platform support.
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::SSE41 => unsafe {
sse41::hash_many(
inputs,
key,
counter,
increment_counter,
flags,
flags_start,
flags_end,
out,
)
},
// Safe because detect() checked for platform support.
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::AVX2 => unsafe {
avx2::hash_many(
inputs,
key,
counter,
increment_counter,
flags,
flags_start,
flags_end,
out,
)
},
// Safe because detect() checked for platform support.
#[cfg(feature = "c_avx512")]
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
Platform::AVX512 => unsafe {
c_avx512::hash_many(
inputs,
key,
counter,
increment_counter,
flags,
flags_start,
flags_end,
out,
)
},
// Assumed to be safe if the "c_neon" feature is on.
#[cfg(feature = "c_neon")]
Platform::NEON => unsafe {
c_neon::hash_many(
inputs,
key,
counter,
increment_counter,
flags,
flags_start,
flags_end,
out,
)
},
}
}
}
// Note that AVX-512 is divided into multiple featuresets, and we use two of
// them, F and VL.
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
#[inline(always)]
pub fn avx512_detected() -> bool {
// Static check, e.g. for building with target-cpu=native.
#[cfg(all(target_feature = "avx512f", target_feature = "avx512vl"))]
{
return true;
}
// Dyanmic check, if std is enabled.
#[cfg(feature = "std")]
{
if is_x86_feature_detected!("avx512f") && is_x86_feature_detected!("avx512vl") {
return true;
}
}
false
}
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
#[inline(always)]
pub fn avx2_detected() -> bool {
// Static check, e.g. for building with target-cpu=native.
#[cfg(target_feature = "avx2")]
{
return true;
}
// Dyanmic check, if std is enabled.
#[cfg(feature = "std")]
{
if is_x86_feature_detected!("avx2") {
return true;
}
}
false
}
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
#[inline(always)]
pub fn sse41_detected() -> bool {
// Static check, e.g. for building with target-cpu=native.
#[cfg(target_feature = "sse4.1")]
{
return true;
}
// Dyanmic check, if std is enabled.
#[cfg(feature = "std")]
{
if is_x86_feature_detected!("sse4.1") {
return true;
}
}
false
}
#[inline(always)]
pub fn words_from_le_bytes_32(bytes: &[u8; 32]) -> [u32; 8] {
let mut out = [0; 8];
out[0] = u32::from_le_bytes(*array_ref!(bytes, 0 * 4, 4));
out[1] = u32::from_le_bytes(*array_ref!(bytes, 1 * 4, 4));
out[2] = u32::from_le_bytes(*array_ref!(bytes, 2 * 4, 4));
out[3] = u32::from_le_bytes(*array_ref!(bytes, 3 * 4, 4));
out[4] = u32::from_le_bytes(*array_ref!(bytes, 4 * 4, 4));
out[5] = u32::from_le_bytes(*array_ref!(bytes, 5 * 4, 4));
out[6] = u32::from_le_bytes(*array_ref!(bytes, 6 * 4, 4));
out[7] = u32::from_le_bytes(*array_ref!(bytes, 7 * 4, 4));
out
}
#[inline(always)]
pub fn words_from_le_bytes_64(bytes: &[u8; 64]) -> [u32; 16] {
let mut out = [0; 16];
out[0] = u32::from_le_bytes(*array_ref!(bytes, 0 * 4, 4));
out[1] = u32::from_le_bytes(*array_ref!(bytes, 1 * 4, 4));
out[2] = u32::from_le_bytes(*array_ref!(bytes, 2 * 4, 4));
out[3] = u32::from_le_bytes(*array_ref!(bytes, 3 * 4, 4));
out[4] = u32::from_le_bytes(*array_ref!(bytes, 4 * 4, 4));
out[5] = u32::from_le_bytes(*array_ref!(bytes, 5 * 4, 4));
out[6] = u32::from_le_bytes(*array_ref!(bytes, 6 * 4, 4));
out[7] = u32::from_le_bytes(*array_ref!(bytes, 7 * 4, 4));
out[8] = u32::from_le_bytes(*array_ref!(bytes, 8 * 4, 4));
out[9] = u32::from_le_bytes(*array_ref!(bytes, 9 * 4, 4));
out[10] = u32::from_le_bytes(*array_ref!(bytes, 10 * 4, 4));
out[11] = u32::from_le_bytes(*array_ref!(bytes, 11 * 4, 4));
out[12] = u32::from_le_bytes(*array_ref!(bytes, 12 * 4, 4));
out[13] = u32::from_le_bytes(*array_ref!(bytes, 13 * 4, 4));
out[14] = u32::from_le_bytes(*array_ref!(bytes, 14 * 4, 4));
out[15] = u32::from_le_bytes(*array_ref!(bytes, 15 * 4, 4));
out
}
#[inline(always)]
pub fn le_bytes_from_words_32(words: &[u32; 8]) -> [u8; 32] {
let mut out = [0; 32];
*array_mut_ref!(out, 0 * 4, 4) = words[0].to_le_bytes();
*array_mut_ref!(out, 1 * 4, 4) = words[1].to_le_bytes();
*array_mut_ref!(out, 2 * 4, 4) = words[2].to_le_bytes();
*array_mut_ref!(out, 3 * 4, 4) = words[3].to_le_bytes();
*array_mut_ref!(out, 4 * 4, 4) = words[4].to_le_bytes();
*array_mut_ref!(out, 5 * 4, 4) = words[5].to_le_bytes();
*array_mut_ref!(out, 6 * 4, 4) = words[6].to_le_bytes();
*array_mut_ref!(out, 7 * 4, 4) = words[7].to_le_bytes();
out
}
#[inline(always)]
pub fn le_bytes_from_words_64(words: &[u32; 16]) -> [u8; 64] {
let mut out = [0; 64];
*array_mut_ref!(out, 0 * 4, 4) = words[0].to_le_bytes();
*array_mut_ref!(out, 1 * 4, 4) = words[1].to_le_bytes();
*array_mut_ref!(out, 2 * 4, 4) = words[2].to_le_bytes();
*array_mut_ref!(out, 3 * 4, 4) = words[3].to_le_bytes();
*array_mut_ref!(out, 4 * 4, 4) = words[4].to_le_bytes();
*array_mut_ref!(out, 5 * 4, 4) = words[5].to_le_bytes();
*array_mut_ref!(out, 6 * 4, 4) = words[6].to_le_bytes();
*array_mut_ref!(out, 7 * 4, 4) = words[7].to_le_bytes();
*array_mut_ref!(out, 8 * 4, 4) = words[8].to_le_bytes();
*array_mut_ref!(out, 9 * 4, 4) = words[9].to_le_bytes();
*array_mut_ref!(out, 10 * 4, 4) = words[10].to_le_bytes();
*array_mut_ref!(out, 11 * 4, 4) = words[11].to_le_bytes();
*array_mut_ref!(out, 12 * 4, 4) = words[12].to_le_bytes();
*array_mut_ref!(out, 13 * 4, 4) = words[13].to_le_bytes();
*array_mut_ref!(out, 14 * 4, 4) = words[14].to_le_bytes();
*array_mut_ref!(out, 15 * 4, 4) = words[15].to_le_bytes();
out
}
|
