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
|
// Copyright 2020 The Abseil Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "y_absl/hash/internal/low_level_hash.h"
#include <cstddef>
#include <cstdint>
#include "y_absl/base/internal/unaligned_access.h"
#include "y_absl/base/prefetch.h"
#include "y_absl/numeric/int128.h"
namespace y_absl {
Y_ABSL_NAMESPACE_BEGIN
namespace hash_internal {
static uint64_t Mix(uint64_t v0, uint64_t v1) {
y_absl::uint128 p = v0;
p *= v1;
return y_absl::Uint128Low64(p) ^ y_absl::Uint128High64(p);
}
uint64_t LowLevelHashLenGt16(const void* data, size_t len, uint64_t seed,
const uint64_t salt[5]) {
// Prefetch the cacheline that data resides in.
PrefetchToLocalCache(data);
const uint8_t* ptr = static_cast<const uint8_t*>(data);
uint64_t starting_length = static_cast<uint64_t>(len);
const uint8_t* last_16_ptr = ptr + starting_length - 16;
uint64_t current_state = seed ^ salt[0];
if (len > 64) {
// If we have more than 64 bytes, we're going to handle chunks of 64
// bytes at a time. We're going to build up two separate hash states
// which we will then hash together.
uint64_t duplicated_state0 = current_state;
uint64_t duplicated_state1 = current_state;
uint64_t duplicated_state2 = current_state;
do {
// Always prefetch the next cacheline.
PrefetchToLocalCache(ptr + Y_ABSL_CACHELINE_SIZE);
uint64_t a = y_absl::base_internal::UnalignedLoad64(ptr);
uint64_t b = y_absl::base_internal::UnalignedLoad64(ptr + 8);
uint64_t c = y_absl::base_internal::UnalignedLoad64(ptr + 16);
uint64_t d = y_absl::base_internal::UnalignedLoad64(ptr + 24);
uint64_t e = y_absl::base_internal::UnalignedLoad64(ptr + 32);
uint64_t f = y_absl::base_internal::UnalignedLoad64(ptr + 40);
uint64_t g = y_absl::base_internal::UnalignedLoad64(ptr + 48);
uint64_t h = y_absl::base_internal::UnalignedLoad64(ptr + 56);
current_state = Mix(a ^ salt[1], b ^ current_state);
duplicated_state0 = Mix(c ^ salt[2], d ^ duplicated_state0);
duplicated_state1 = Mix(e ^ salt[3], f ^ duplicated_state1);
duplicated_state2 = Mix(g ^ salt[4], h ^ duplicated_state2);
ptr += 64;
len -= 64;
} while (len > 64);
current_state = (current_state ^ duplicated_state0) ^
(duplicated_state1 + duplicated_state2);
}
// We now have a data `ptr` with at most 64 bytes and the current state
// of the hashing state machine stored in current_state.
if (len > 32) {
uint64_t a = y_absl::base_internal::UnalignedLoad64(ptr);
uint64_t b = y_absl::base_internal::UnalignedLoad64(ptr + 8);
uint64_t c = y_absl::base_internal::UnalignedLoad64(ptr + 16);
uint64_t d = y_absl::base_internal::UnalignedLoad64(ptr + 24);
uint64_t cs0 = Mix(a ^ salt[1], b ^ current_state);
uint64_t cs1 = Mix(c ^ salt[2], d ^ current_state);
current_state = cs0 ^ cs1;
ptr += 32;
len -= 32;
}
// We now have a data `ptr` with at most 32 bytes and the current state
// of the hashing state machine stored in current_state.
if (len > 16) {
uint64_t a = y_absl::base_internal::UnalignedLoad64(ptr);
uint64_t b = y_absl::base_internal::UnalignedLoad64(ptr + 8);
current_state = Mix(a ^ salt[1], b ^ current_state);
}
// We now have a data `ptr` with at least 1 and at most 16 bytes. But we can
// safely read from `ptr + len - 16`.
uint64_t a = y_absl::base_internal::UnalignedLoad64(last_16_ptr);
uint64_t b = y_absl::base_internal::UnalignedLoad64(last_16_ptr + 8);
return Mix(a ^ salt[1] ^ starting_length, b ^ current_state);
}
uint64_t LowLevelHash(const void* data, size_t len, uint64_t seed,
const uint64_t salt[5]) {
if (len > 16) return LowLevelHashLenGt16(data, len, seed, salt);
// Prefetch the cacheline that data resides in.
PrefetchToLocalCache(data);
const uint8_t* ptr = static_cast<const uint8_t*>(data);
uint64_t starting_length = static_cast<uint64_t>(len);
uint64_t current_state = seed ^ salt[0];
if (len == 0) return current_state;
uint64_t a = 0;
uint64_t b = 0;
// We now have a data `ptr` with at least 1 and at most 16 bytes.
if (len > 8) {
// When we have at least 9 and at most 16 bytes, set A to the first 64
// bits of the input and B to the last 64 bits of the input. Yes, they
// will overlap in the middle if we are working with less than the full 16
// bytes.
a = y_absl::base_internal::UnalignedLoad64(ptr);
b = y_absl::base_internal::UnalignedLoad64(ptr + len - 8);
} else if (len > 3) {
// If we have at least 4 and at most 8 bytes, set A to the first 32
// bits and B to the last 32 bits.
a = y_absl::base_internal::UnalignedLoad32(ptr);
b = y_absl::base_internal::UnalignedLoad32(ptr + len - 4);
} else {
// If we have at least 1 and at most 3 bytes, read 2 bytes into A and the
// other byte into B, with some adjustments.
a = static_cast<uint64_t>((ptr[0] << 8) | ptr[len - 1]);
b = static_cast<uint64_t>(ptr[len >> 1]);
}
return Mix(a ^ salt[1] ^ starting_length, b ^ current_state);
}
} // namespace hash_internal
Y_ABSL_NAMESPACE_END
} // namespace y_absl
|
