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
|
//+build !noasm,!appengine,gc
// Copyright (c) 2020 MinIO Inc. All rights reserved.
// Use of this source code is governed by a license that can be
// found in the LICENSE file.
package md5simd
import (
"bytes"
"hash"
"runtime"
"sync"
"testing"
"github.com/klauspost/cpuid/v2"
)
const benchmarkWithSum = true
func BenchmarkAvx512(b *testing.B) {
if !hasAVX512 {
b.SkipNow()
}
b.Run("32KB", func(b *testing.B) {
benchmarkSingle(b, 32*1024)
})
b.Run("64KB", func(b *testing.B) {
benchmarkSingle(b, 64*1024)
})
b.Run("128KB", func(b *testing.B) {
benchmarkSingle(b, 128*1024)
})
b.Run("256KB", func(b *testing.B) {
benchmarkSingle(b, 256*1024)
})
b.Run("512KB", func(b *testing.B) {
benchmarkSingle(b, 512*1024)
})
b.Run("1MB", func(b *testing.B) {
benchmarkSingle(b, 1024*1024)
})
b.Run("2MB", func(b *testing.B) {
benchmarkSingle(b, 2*1024*1024)
})
b.Run("4MB", func(b *testing.B) {
benchmarkSingle(b, 4*1024*1024)
})
b.Run("8MB", func(b *testing.B) {
benchmarkSingle(b, 8*1024*1024)
})
}
func BenchmarkAvx512Parallel(b *testing.B) {
if !hasAVX512 {
b.SkipNow()
}
b.Run("32KB", func(b *testing.B) {
benchmarkParallel(b, 32*1024)
})
b.Run("64KB", func(b *testing.B) {
benchmarkParallel(b, 64*1024)
})
b.Run("128KB", func(b *testing.B) {
benchmarkParallel(b, 128*1024)
})
b.Run("256KB", func(b *testing.B) {
benchmarkParallel(b, 256*1024)
})
b.Run("512KB", func(b *testing.B) {
benchmarkParallel(b, 512*1024)
})
b.Run("1MB", func(b *testing.B) {
benchmarkParallel(b, 1024*1024)
})
b.Run("2MB", func(b *testing.B) {
benchmarkParallel(b, 2*1024*1024)
})
b.Run("4MB", func(b *testing.B) {
benchmarkParallel(b, 4*1024*1024)
})
b.Run("8MB", func(b *testing.B) {
benchmarkParallel(b, 8*1024*1024)
})
}
func benchmarkSingle(b *testing.B, blockSize int) {
server := NewServer()
defer server.Close()
h16 := [16]hash.Hash{}
input := [16][]byte{}
for i := range h16 {
h16[i] = server.NewHash()
input[i] = bytes.Repeat([]byte{0x61 + byte(i)}, blockSize)
}
// Technically this uses up to 2 cores, but it is the throughput of a single server.
b.SetBytes(int64(blockSize * 16))
b.ReportAllocs()
b.ResetTimer()
var tmp [Size]byte
for j := 0; j < b.N; j++ {
var wg sync.WaitGroup
wg.Add(16)
for i := range h16 {
go func(i int) {
// write to all concurrently
defer wg.Done()
h16[i].Reset()
h16[i].Write(input[i])
if benchmarkWithSum {
_ = h16[i].Sum(tmp[:0])
}
}(i)
}
wg.Wait()
}
}
func benchmarkSingleWriter(b *testing.B, blockSize int) {
server := NewServer()
defer server.Close()
h := server.NewHash()
input := bytes.Repeat([]byte{0x61}, blockSize)
b.SetBytes(int64(blockSize))
b.ReportAllocs()
b.ResetTimer()
var tmp [Size]byte
for j := 0; j < b.N; j++ {
h.Write(input)
if benchmarkWithSum {
_ = h.Sum(tmp[:0])
}
}
}
func benchmarkParallel(b *testing.B, blockSize int) {
// We write input 16x per loop.
// We have to alloc per parallel
b.SetBytes(int64(blockSize * 16))
b.ReportAllocs()
b.ResetTimer()
b.RunParallel(func(pb *testing.PB) {
input := bytes.Repeat([]byte{0x61}, blockSize)
server := NewServer()
defer server.Close()
var h16 [16]Hasher
for i := range h16 {
h16[i] = server.NewHash()
defer h16[i].Close()
}
var tmp [Size]byte
for pb.Next() {
var wg sync.WaitGroup
wg.Add(16)
for i := range h16 {
// Write to all concurrently
go func(i int) {
defer wg.Done()
h16[i].Reset()
h16[i].Write(input)
if benchmarkWithSum {
_ = h16[i].Sum(tmp[:0])
}
}(i)
}
wg.Wait()
}
})
}
func BenchmarkAvx2(b *testing.B) {
// Make sure AVX512 is disabled
restore := hasAVX512
hasAVX512 = false
b.Run("32KB", func(b *testing.B) {
benchmarkSingle(b, 32*1024)
})
b.Run("64KB", func(b *testing.B) {
benchmarkSingle(b, 64*1024)
})
b.Run("128KB", func(b *testing.B) {
benchmarkSingle(b, 128*1024)
})
b.Run("256KB", func(b *testing.B) {
benchmarkSingle(b, 256*1024)
})
b.Run("512KB", func(b *testing.B) {
benchmarkSingle(b, 512*1024)
})
b.Run("1MB", func(b *testing.B) {
benchmarkSingle(b, 1024*1024)
})
b.Run("2MB", func(b *testing.B) {
benchmarkSingle(b, 2*1024*1024)
})
b.Run("4MB", func(b *testing.B) {
benchmarkSingle(b, 4*1024*1024)
})
b.Run("8MB", func(b *testing.B) {
benchmarkSingle(b, 8*1024*1024)
})
hasAVX512 = restore
}
func BenchmarkAvx2Parallel(b *testing.B) {
if !cpuid.CPU.Supports(cpuid.AVX2) {
b.SkipNow()
}
restore := hasAVX512
// Make sure AVX512 is disabled
hasAVX512 = false
b.SetParallelism((runtime.GOMAXPROCS(0) + 1) / 2)
b.Run("32KB", func(b *testing.B) {
benchmarkParallel(b, 32*1024)
})
b.Run("64KB", func(b *testing.B) {
benchmarkParallel(b, 64*1024)
})
b.Run("128KB", func(b *testing.B) {
benchmarkParallel(b, 128*1024)
})
b.Run("256KB", func(b *testing.B) {
benchmarkParallel(b, 256*1024)
})
b.Run("512KB", func(b *testing.B) {
benchmarkParallel(b, 512*1024)
})
b.Run("1MB", func(b *testing.B) {
benchmarkParallel(b, 1024*1024)
})
b.Run("2MB", func(b *testing.B) {
benchmarkParallel(b, 2*1024*1024)
})
b.Run("4MB", func(b *testing.B) {
benchmarkParallel(b, 4*1024*1024)
})
b.Run("8MB", func(b *testing.B) {
benchmarkParallel(b, 8*1024*1024)
})
hasAVX512 = restore
}
// BenchmarkAvx2SingleWriter will benchmark the speed having only a single writer
// writing blocks with the specified size.
// This is pretty much the worst case scenario.
func BenchmarkAvx2SingleWriter(b *testing.B) {
// Make sure AVX512 is disabled
restore := hasAVX512
hasAVX512 = false
b.Run("32KB", func(b *testing.B) {
benchmarkSingleWriter(b, 32*1024)
})
b.Run("64KB", func(b *testing.B) {
benchmarkSingleWriter(b, 64*1024)
})
b.Run("128KB", func(b *testing.B) {
benchmarkSingleWriter(b, 128*1024)
})
b.Run("256KB", func(b *testing.B) {
benchmarkSingleWriter(b, 256*1024)
})
b.Run("512KB", func(b *testing.B) {
benchmarkSingleWriter(b, 512*1024)
})
b.Run("1MB", func(b *testing.B) {
benchmarkSingleWriter(b, 1024*1024)
})
b.Run("2MB", func(b *testing.B) {
benchmarkSingleWriter(b, 2*1024*1024)
})
b.Run("4MB", func(b *testing.B) {
benchmarkSingleWriter(b, 4*1024*1024)
})
b.Run("8MB", func(b *testing.B) {
benchmarkSingleWriter(b, 8*1024*1024)
})
hasAVX512 = restore
}
|
