File: murmur.go

package info (click to toggle)
golang-github-retailnext-hllpp 1.0.0+git20170901.6e8b6d3-3
  • links: PTS, VCS
  • area: main
  • in suites: buster
  • size: 196 kB
  • sloc: makefile: 2
file content (155 lines) | stat: -rw-r--r-- 2,816 bytes parent folder | download | duplicates (2)
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
// Copyright (c) 2015, RetailNext, Inc.
// All rights reserved.

package hllpp

import (
	"encoding/binary"
	"reflect"
	"unsafe"
)

// Non-streaming implementation of murmur3. Not only is it faster to begin with
// vs cryptographic hashing functions, but I can avoid memory allocations by
// not using the go streaming hash.Hash interface.

const (
	murmurC1 = 0x87c37b91114253d5
	murmurC2 = 0x4cf5ad432745937f
)

var bigEndian bool

func init() {
	var t uint16 = 1
	bigEndian = (*[2]byte)(unsafe.Pointer(&t))[0] == 0
}

// This is a port of MurmurHash3_x64_128 from MurmurHash3.cpp
func murmurSum64(data []byte) uint64 {
	var h1, h2, k1, k2 uint64

	len := len(data)

	nBlocks := len / 16

	var data64 []uint64

	if bigEndian {
		data64 = make([]uint64, nBlocks*2)
		for i := 0; i < nBlocks*2; i++ {
			data64[i] = binary.LittleEndian.Uint64(data[8*i:])
		}
	} else {
		dataHeader := (*reflect.SliceHeader)(unsafe.Pointer(&data))
		data64Header := (*reflect.SliceHeader)(unsafe.Pointer(&data64))
		data64Header.Data = dataHeader.Data
		data64Header.Len = 2 * nBlocks
		data64Header.Cap = 2 * nBlocks
	}

	for i := 0; i < nBlocks; i++ {
		k1 = data64[2*i]
		k2 = data64[2*i+1]

		k1 *= murmurC1
		k1 = (k1 << 31) | (k1 >> (64 - 31))
		k1 *= murmurC2
		h1 ^= k1

		h1 = (h1 << 27) | (h1 >> (64 - 27))
		h1 += h2
		h1 = h1*5 + 0x52dce729

		k2 *= murmurC2
		k2 = (k2 << 33) | (k2 >> (64 - 33))
		k2 *= murmurC1
		h2 ^= k2

		h2 = (h2 << 31) | (h2 >> (64 - 31))
		h2 += h1
		h2 = h2*5 + 0x38495ab5
	}

	tail := data[nBlocks*16:]

	k1, k2 = 0, 0

	switch len & 15 {
	case 15:
		k2 ^= uint64(tail[14]) << 48
		fallthrough
	case 14:
		k2 ^= uint64(tail[13]) << 40
		fallthrough
	case 13:
		k2 ^= uint64(tail[12]) << 32
		fallthrough
	case 12:
		k2 ^= uint64(tail[11]) << 24
		fallthrough
	case 11:
		k2 ^= uint64(tail[10]) << 16
		fallthrough
	case 10:
		k2 ^= uint64(tail[9]) << 8
		fallthrough
	case 9:
		k2 ^= uint64(tail[8]) << 0
		k2 *= murmurC2
		k2 = (k2 << 33) | (k2 >> (64 - 33))
		k2 *= murmurC1
		h2 ^= k2
		fallthrough
	case 8:
		k1 ^= uint64(tail[7]) << 56
		fallthrough
	case 7:
		k1 ^= uint64(tail[6]) << 48
		fallthrough
	case 6:
		k1 ^= uint64(tail[5]) << 40
		fallthrough
	case 5:
		k1 ^= uint64(tail[4]) << 32
		fallthrough
	case 4:
		k1 ^= uint64(tail[3]) << 24
		fallthrough
	case 3:
		k1 ^= uint64(tail[2]) << 16
		fallthrough
	case 2:
		k1 ^= uint64(tail[1]) << 8
		fallthrough
	case 1:
		k1 ^= uint64(tail[0]) << 0
		k1 *= murmurC1
		k1 = (k1 << 31) | (k1 >> (64 - 31))
		k1 *= murmurC2
		h1 ^= k1
	}

	h1 ^= uint64(len)
	h2 ^= uint64(len)

	h1 += h2
	h2 += h1

	h1 ^= h1 >> 33
	h1 *= 0xff51afd7ed558ccd
	h1 ^= h1 >> 33
	h1 *= 0xc4ceb9fe1a85ec53
	h1 ^= h1 >> 33

	h2 ^= h2 >> 33
	h2 *= 0xff51afd7ed558ccd
	h2 ^= h2 >> 33
	h2 *= 0xc4ceb9fe1a85ec53
	h2 ^= h2 >> 33

	h1 += h2
	h2 += h1

	return h1
}