benchmark-inputs 1.0.1 → 1.1.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +5 -5
- data/CHANGELOG.md +13 -0
- data/README.md +46 -41
- data/benchmark-inputs.gemspec +2 -2
- data/lib/benchmark/inputs.rb +124 -48
- data/lib/benchmark/inputs/version.rb +1 -1
- metadata +8 -9
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
|
-
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
2
|
+
SHA256:
|
3
|
+
metadata.gz: a6e909d515239a8262e68c3f13b3cdb6d3840be4999b647aa1267dbd3184f69b
|
4
|
+
data.tar.gz: 37c78630c9af1f254438db2cb427e073a2bc97e2b25f7cedf4a0f73deba34d00
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 184afc9a7110ad536dce0881db8c092958986e245e0a90af9040ac70372afac583811b29765c0098b5233e7e1c637c3c10188cce44e4e4c0d48930b96e5ccf8d
|
7
|
+
data.tar.gz: 9859889f725c6849d524d6fd08cb1a579c6598f1135b7000adc4cc77cb3910899a8700ee931f3679d36b36cbd11b7c19e08350b4c03b708dc8896b4f7af0c22e
|
data/CHANGELOG.md
ADDED
data/README.md
CHANGED
@@ -1,47 +1,51 @@
|
|
1
1
|
# benchmark-inputs
|
2
2
|
|
3
|
-
Input-focused benchmarking for Ruby. Given one or more blocks and
|
4
|
-
|
5
|
-
(in invocations per second) of each block. Blocks which
|
6
|
-
quickly, as in microbenchmarks, are automatically invoked
|
7
|
-
provide accurate measurements.
|
3
|
+
Input-focused benchmarking for Ruby. Given one or more blocks and an
|
4
|
+
array of inputs to yield to each of them, benchmark-inputs will measure
|
5
|
+
the speed (in invocations per second) of each block. Blocks which
|
6
|
+
execute very quickly, as in microbenchmarks, are automatically invoked
|
7
|
+
repeatedly to provide accurate measurements.
|
8
8
|
|
9
9
|
|
10
10
|
## Motivation
|
11
11
|
|
12
12
|
I <3 [Fast Ruby][fast-ruby]. By extension, I <3 [benchmark-ips]. But,
|
13
|
-
for
|
14
|
-
way I'd like.
|
13
|
+
for some use cases, benchmark-ips doesn't let me write benchmarks the
|
14
|
+
way I'd like. Consider the following example, *using benchmark-ips*:
|
15
15
|
|
16
16
|
```ruby
|
17
|
-
require
|
17
|
+
require "benchmark/ips" ### USING benchmark-ips (NOT benchmark-inputs)
|
18
18
|
|
19
|
-
STRINGS = [
|
19
|
+
STRINGS = ["abc", "aaa", "xyz", ""]
|
20
20
|
Benchmark.ips do |job|
|
21
|
-
job.report(
|
22
|
-
job.report(
|
21
|
+
job.report("String#tr"){ STRINGS.each{|s| s.tr("a", "A") } }
|
22
|
+
job.report("String#gsub"){ STRINGS.each{|s| s.gsub(/a/, "A") } }
|
23
23
|
job.compare!
|
24
24
|
end
|
25
25
|
```
|
26
26
|
|
27
|
-
The calls to `STRINGS.each` introduce performance overhead
|
27
|
+
The calls to `STRINGS.each` introduce performance overhead that skews
|
28
28
|
the time measurements. The less time the target function takes, the
|
29
|
-
more relative overhead, and thus
|
30
|
-
|
31
|
-
|
29
|
+
more relative overhead, and thus more skew. For a microbenchmark this
|
30
|
+
can be a problem. A possible workaround is to invoke the function on
|
31
|
+
each value individually, but that is more verbose and more error-prone:
|
32
32
|
|
33
33
|
```ruby
|
34
|
-
require
|
34
|
+
require "benchmark/ips" ### USING benchmark-ips (NOT benchmark-inputs)
|
35
35
|
|
36
|
-
s1
|
36
|
+
s1, s2, s3, s4 = ["abc", "aaa", "xyz", ""]
|
37
37
|
Benchmark.ips do |job|
|
38
|
-
job.report(
|
39
|
-
s1.tr(
|
40
|
-
|
38
|
+
job.report("String#tr") do
|
39
|
+
s1.tr("a", "A")
|
40
|
+
s2.tr("a", "A")
|
41
|
+
s3.tr("a", "A")
|
42
|
+
s4.tr("a", "A")
|
41
43
|
end
|
42
|
-
job.report(
|
43
|
-
s1.gsub(/a/,
|
44
|
-
|
44
|
+
job.report("String#gsub") do
|
45
|
+
s1.gsub(/a/, "A")
|
46
|
+
s2.gsub(/a/, "A")
|
47
|
+
s3.gsub(/a/, "A")
|
48
|
+
s4.gsub(/a/, "A")
|
45
49
|
end
|
46
50
|
job.compare!
|
47
51
|
end
|
@@ -51,11 +55,11 @@ end
|
|
51
55
|
this gem: <a name="example1"></a>
|
52
56
|
|
53
57
|
```ruby
|
54
|
-
require
|
58
|
+
require "benchmark/inputs" ### USING benchmark-inputs
|
55
59
|
|
56
|
-
Benchmark.inputs([
|
57
|
-
job.report(
|
58
|
-
job.report(
|
60
|
+
Benchmark.inputs(["abc", "aaa", "xyz", ""]) do |job|
|
61
|
+
job.report("String#tr"){|s| s.tr("a", "A") }
|
62
|
+
job.report("String#gsub"){|s| s.gsub(/a/, "A") }
|
59
63
|
job.compare!
|
60
64
|
end
|
61
65
|
```
|
@@ -80,18 +84,17 @@ Destructive operations also pose a challenge for microbenchmarks. Each
|
|
80
84
|
invocation needs to operate on the same data, but `dup`ing the data
|
81
85
|
introduces too much overhead and skew.
|
82
86
|
|
83
|
-
benchmark-inputs' solution is to estimate the overhead incurred by
|
84
|
-
`dup`, and exclude that from time measurements. Because the
|
85
|
-
job already controls the input data,
|
86
|
-
|
87
|
+
benchmark-inputs' solution is to estimate the overhead incurred by each
|
88
|
+
`dup`, and exclude that from the time measurements. Because the
|
89
|
+
benchmark job already controls the input data, everything can be handled
|
90
|
+
behind the scenes. To enable this, use the `dup_inputs` option:
|
87
91
|
|
88
92
|
```ruby
|
89
|
-
require
|
93
|
+
require "benchmark/inputs"
|
90
94
|
|
91
|
-
Benchmark.inputs([
|
92
|
-
job.
|
93
|
-
job.report(
|
94
|
-
job.report('String#gsub!'){|s| s.gsub!(/a/, 'A') }
|
95
|
+
Benchmark.inputs(["abc", "aaa", "xyz", ""], dup_inputs: true) do |job|
|
96
|
+
job.report("String#tr!"){|s| s.tr!("a", "A") }
|
97
|
+
job.report("String#gsub!"){|s| s.gsub!(/a/, "A") }
|
95
98
|
job.compare!
|
96
99
|
end
|
97
100
|
```
|
@@ -111,7 +114,7 @@ Comparison:
|
|
111
114
|
|
112
115
|
That shows a slightly larger performance gap than the previous
|
113
116
|
benchmark. This makes sense because the overhead of allocating new
|
114
|
-
strings--previously via a non-bang method, but now via `dup
|
117
|
+
strings -- previously via a non-bang method, but now via `dup` -- is now
|
115
118
|
excluded from the timings. Thus, the speed of `tr!` relative to `gsub!`
|
116
119
|
is further emphasized.
|
117
120
|
|
@@ -129,13 +132,15 @@ problem.
|
|
129
132
|
|
130
133
|
## Installation
|
131
134
|
|
132
|
-
|
135
|
+
```bash
|
136
|
+
$ gem install benchmark-inputs
|
137
|
+
```
|
133
138
|
|
134
139
|
|
135
140
|
## Usage
|
136
141
|
|
137
|
-
See the [example above](#example1), or check the
|
138
|
-
[documentation](http://www.rubydoc.info/gems/benchmark-inputs).
|
142
|
+
See the [example above](#example1), or check out the
|
143
|
+
[API documentation](http://www.rubydoc.info/gems/benchmark-inputs).
|
139
144
|
|
140
145
|
|
141
146
|
## License
|
@@ -146,4 +151,4 @@ See the [example above](#example1), or check the
|
|
146
151
|
|
147
152
|
|
148
153
|
[fast-ruby]: https://github.com/JuanitoFatas/fast-ruby
|
149
|
-
[benchmark-ips]: https://
|
154
|
+
[benchmark-ips]: https://rubygems.org/gems/benchmark-ips
|
data/benchmark-inputs.gemspec
CHANGED
@@ -19,7 +19,7 @@ Gem::Specification.new do |spec|
|
|
19
19
|
spec.require_paths = ["lib"]
|
20
20
|
|
21
21
|
spec.add_development_dependency "bundler", "~> 1.12"
|
22
|
-
spec.add_development_dependency "rake", "
|
22
|
+
spec.add_development_dependency "rake", ">= 10.0"
|
23
23
|
spec.add_development_dependency "minitest", "~> 5.0"
|
24
|
-
spec.add_development_dependency "yard", "~> 0.
|
24
|
+
spec.add_development_dependency "yard", "~> 0.9"
|
25
25
|
end
|
data/lib/benchmark/inputs.rb
CHANGED
@@ -1,23 +1,45 @@
|
|
1
|
-
require
|
1
|
+
require "benchmark/inputs/version"
|
2
2
|
|
3
3
|
module Benchmark
|
4
4
|
|
5
5
|
# Initializes a benchmark job with the given inputs and yields that
|
6
6
|
# job to the given block.
|
7
7
|
#
|
8
|
-
#
|
9
|
-
# Benchmark.inputs([
|
10
|
-
# job.report(
|
11
|
-
# job.report(
|
8
|
+
# @example Benchmarking non-destructive operations
|
9
|
+
# Benchmark.inputs(["abc", "aaa", "xyz", ""]) do |job|
|
10
|
+
# job.report("String#tr"){|s| s.tr("a", "A") }
|
11
|
+
# job.report("String#gsub"){|s| s.gsub(/a/, "A") }
|
12
12
|
# job.compare!
|
13
13
|
# end
|
14
14
|
#
|
15
|
-
# @
|
16
|
-
#
|
17
|
-
#
|
18
|
-
#
|
19
|
-
|
20
|
-
|
15
|
+
# @example Benchmarking destructive operations
|
16
|
+
# Benchmark.inputs(["abc", "aaa", "xyz", ""], dup_inputs: true) do |job|
|
17
|
+
# job.report("String#tr!"){|s| s.tr!("a", "A") }
|
18
|
+
# job.report("String#gsub!"){|s| s.gsub!(/a/, "A") }
|
19
|
+
# job.compare!
|
20
|
+
# end
|
21
|
+
#
|
22
|
+
# @param values [Array]
|
23
|
+
# input values to yield to each benchmark action
|
24
|
+
# @param options [Hash]
|
25
|
+
# @option options :dup_inputs [Boolean]
|
26
|
+
# whether input values will be +dup+-ed before they are passed to a
|
27
|
+
# {Inputs::Job#report} block
|
28
|
+
# @option options :sample_n [Integer]
|
29
|
+
# number of samples to take when benchmarking
|
30
|
+
# @option options :sample_dt [Integer]
|
31
|
+
# approximate duration of time (in nanoseconds) each sample should
|
32
|
+
# take when benchmarking
|
33
|
+
# @yield [job]
|
34
|
+
# configures job and runs benchmarks
|
35
|
+
# @yieldparam job [Benchmark::Inputs::Job]
|
36
|
+
# benchmark runner
|
37
|
+
# @return [Benchmark::Inputs::Job]
|
38
|
+
# benchmark runner
|
39
|
+
# @raise [ArgumentError]
|
40
|
+
# if +values+ is empty
|
41
|
+
def self.inputs(values, **options)
|
42
|
+
job = Inputs::Job.new(values, options)
|
21
43
|
yield job
|
22
44
|
job
|
23
45
|
end
|
@@ -29,46 +51,78 @@ module Benchmark
|
|
29
51
|
NS_PER_MS = NS_PER_S / 1_000
|
30
52
|
|
31
53
|
class Job
|
32
|
-
attr_accessor :sample_n, :sample_dt
|
33
|
-
attr_reader :dup_inputs, :reports
|
34
54
|
|
35
|
-
|
55
|
+
# @param inputs [Array]
|
56
|
+
# input values to yield to each benchmark action
|
57
|
+
# @param dup_inputs [Boolean]
|
58
|
+
# whether input values will be +dup+-ed before they are passed
|
59
|
+
# to a {report} block
|
60
|
+
# @param sample_n [Integer]
|
61
|
+
# number of samples to take when benchmarking
|
62
|
+
# @param sample_dt [Integer]
|
63
|
+
# approximate duration of time (in nanoseconds) each sample
|
64
|
+
# should take when benchmarking
|
65
|
+
# @raise [ArgumentError]
|
66
|
+
# if +inputs+ is empty
|
67
|
+
def initialize(inputs, dup_inputs: false, sample_n: 10, sample_dt: NS_PER_MS * 200)
|
68
|
+
raise ArgumentError, "No inputs specified" if inputs.empty?
|
69
|
+
|
36
70
|
@inputs = inputs
|
37
|
-
@dup_inputs =
|
38
|
-
@sample_n =
|
39
|
-
@sample_dt =
|
71
|
+
@dup_inputs = dup_inputs
|
72
|
+
@sample_n = sample_n
|
73
|
+
@sample_dt = sample_dt
|
40
74
|
@reports = []
|
41
75
|
def_bench!
|
42
76
|
end
|
43
77
|
|
44
|
-
#
|
45
|
-
#
|
46
|
-
#
|
78
|
+
# Indicates whether input values will be +dup+-ed before they are
|
79
|
+
# passed to a {report} block. Defaults to +false+. This should
|
80
|
+
# be set to +true+ if {report} blocks destructively modify their
|
47
81
|
# arguments.
|
48
82
|
#
|
49
|
-
#
|
50
|
-
|
51
|
-
|
52
|
-
#
|
53
|
-
# job.report('String#gsub!'){|s| s.gsub!(/a/, 'A') }
|
54
|
-
# job.compare!
|
55
|
-
# end
|
83
|
+
# @return [Boolean]
|
84
|
+
attr_reader :dup_inputs
|
85
|
+
|
86
|
+
# See {dup_inputs}.
|
56
87
|
#
|
57
|
-
# @param [Boolean]
|
58
|
-
|
59
|
-
|
88
|
+
# @param flag [Boolean]
|
89
|
+
# @return [Boolean]
|
90
|
+
def dup_inputs=(flag)
|
91
|
+
@dup_inputs = flag
|
60
92
|
def_bench!
|
61
93
|
@dup_inputs
|
62
94
|
end
|
63
95
|
|
64
|
-
#
|
65
|
-
#
|
66
|
-
#
|
67
|
-
|
96
|
+
# The number of samples to take when benchmarking. Defaults to 10.
|
97
|
+
#
|
98
|
+
# @return [Integer]
|
99
|
+
attr_accessor :sample_n
|
100
|
+
|
101
|
+
# The approximate duration of time (in nanoseconds) each sample
|
102
|
+
# should take when benchmarking. Defaults to 200,000 nanoseconds.
|
103
|
+
#
|
104
|
+
# @return [Integer]
|
105
|
+
attr_accessor :sample_dt
|
106
|
+
|
107
|
+
# Array of benchmark reports. Each call to {report} adds an
|
108
|
+
# element to this array.
|
68
109
|
#
|
69
|
-
# @
|
70
|
-
|
71
|
-
|
110
|
+
# @return [Array<Benchmark::Inputs::Report>]
|
111
|
+
attr_reader :reports
|
112
|
+
|
113
|
+
# Benchmarks the given block using the previously provided input
|
114
|
+
# values. If {dup_inputs} is set to true, each input value is
|
115
|
+
# +dup+'d before being passed to the block. The block's
|
116
|
+
# iterations per second (i/s) is printed to +$stdout+, and a
|
117
|
+
# {Report} is added to {reports}.
|
118
|
+
#
|
119
|
+
# @param label [String]
|
120
|
+
# label for the benchmark
|
121
|
+
# @yield [input]
|
122
|
+
# action to benchmark
|
123
|
+
# @yieldparam input [Object]
|
124
|
+
# one of the initially provided input values
|
125
|
+
# @return [void]
|
72
126
|
def report(label)
|
73
127
|
# estimate repititions
|
74
128
|
reps = 1
|
@@ -91,13 +145,14 @@ module Benchmark
|
|
91
145
|
$stdout.puts(r.label)
|
92
146
|
$stdout.printf(" %.1f i/s (\u00B1%.2f%%)\n", r.ips, r.stddev / r.ips * 100)
|
93
147
|
@reports << r
|
94
|
-
r
|
95
148
|
end
|
96
149
|
|
97
150
|
# Prints the relative speeds (from fastest to slowest) of all
|
98
|
-
#
|
151
|
+
# {report}-ed benchmarks to +$stdout+.
|
152
|
+
#
|
153
|
+
# @return [void]
|
99
154
|
def compare!
|
100
|
-
return $stdout.puts(
|
155
|
+
return $stdout.puts("Nothing to compare!") if @reports.empty?
|
101
156
|
|
102
157
|
@reports.sort_by!{|r| -r.ips }
|
103
158
|
@reports.each{|r| r.slower_than!(@reports.first) }
|
@@ -109,9 +164,9 @@ module Benchmark
|
|
109
164
|
@reports.each_with_index do |r, i|
|
110
165
|
$stdout.printf(format, r.label, r.ips)
|
111
166
|
if r.ratio
|
112
|
-
$stdout.printf(
|
167
|
+
$stdout.printf(" - %.2fx slower", r.ratio)
|
113
168
|
elsif i > 0
|
114
|
-
$stdout.printf(
|
169
|
+
$stdout.printf(" - same-ish: difference falls within error")
|
115
170
|
end
|
116
171
|
$stdout.puts
|
117
172
|
end
|
@@ -123,12 +178,11 @@ module Benchmark
|
|
123
178
|
def def_bench!
|
124
179
|
assigns = @inputs.each_index.map do |i|
|
125
180
|
"x#{i} = @inputs[#{i}]"
|
126
|
-
end.join(
|
181
|
+
end.join(";")
|
127
182
|
|
128
|
-
yields = @inputs.
|
129
|
-
|
130
|
-
|
131
|
-
end.join(';')
|
183
|
+
yields = @inputs.each_index.map do |i|
|
184
|
+
dup_inputs ? "yield(x#{i}.dup)" : "yield(x#{i})"
|
185
|
+
end.join(";")
|
132
186
|
|
133
187
|
code = <<-CODE
|
134
188
|
def bench(reps)
|
@@ -151,8 +205,21 @@ module Benchmark
|
|
151
205
|
|
152
206
|
|
153
207
|
class Report
|
154
|
-
|
208
|
+
# The label for the report.
|
209
|
+
#
|
210
|
+
# @return [String]
|
211
|
+
attr_reader :label
|
212
|
+
|
213
|
+
# The ratio of iterations per second for this report compared to
|
214
|
+
# the fastest report. Will be +nil+ if the difference between the
|
215
|
+
# two falls within the combined measurement error.
|
216
|
+
#
|
217
|
+
# This value is set by {Benchmark::Inputs::Job#compare!}.
|
218
|
+
#
|
219
|
+
# @return [Float, nil]
|
220
|
+
attr_reader :ratio
|
155
221
|
|
222
|
+
# @!visibility private
|
156
223
|
def initialize(label, invocs_per_sample)
|
157
224
|
@label = label.to_s
|
158
225
|
@invocs_per_sample = invocs_per_sample.to_f
|
@@ -163,6 +230,7 @@ module Benchmark
|
|
163
230
|
@m2 = 0.0
|
164
231
|
end
|
165
232
|
|
233
|
+
# @!visibility private
|
166
234
|
def add_sample(time_ns)
|
167
235
|
sample_ips = @invocs_per_sample * NS_PER_S / time_ns
|
168
236
|
|
@@ -175,18 +243,26 @@ module Benchmark
|
|
175
243
|
@stddev = nil
|
176
244
|
end
|
177
245
|
|
246
|
+
# The estimated iterations per second for the report.
|
247
|
+
#
|
248
|
+
# @return [Float]
|
178
249
|
def ips
|
179
250
|
@mean
|
180
251
|
end
|
181
252
|
|
253
|
+
# The {ips} standard deviation.
|
254
|
+
#
|
255
|
+
# @return [Float]
|
182
256
|
def stddev
|
183
257
|
@stddev ||= @n < 2 ? 0.0 : Math.sqrt(@m2 / (@n - 1))
|
184
258
|
end
|
185
259
|
|
260
|
+
# @!visibility private
|
186
261
|
def slower_than!(faster)
|
187
262
|
@ratio = overlap?(faster) ? nil : (faster.ips / self.ips)
|
188
263
|
end
|
189
264
|
|
265
|
+
# @!visibility private
|
190
266
|
def overlap?(faster)
|
191
267
|
(faster.ips - faster.stddev) <= (self.ips + self.stddev)
|
192
268
|
end
|
metadata
CHANGED
@@ -1,14 +1,14 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: benchmark-inputs
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 1.0
|
4
|
+
version: 1.1.0
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Jonathan Hefner
|
8
8
|
autorequire:
|
9
9
|
bindir: exe
|
10
10
|
cert_chain: []
|
11
|
-
date:
|
11
|
+
date: 2019-03-25 00:00:00.000000000 Z
|
12
12
|
dependencies:
|
13
13
|
- !ruby/object:Gem::Dependency
|
14
14
|
name: bundler
|
@@ -28,14 +28,14 @@ dependencies:
|
|
28
28
|
name: rake
|
29
29
|
requirement: !ruby/object:Gem::Requirement
|
30
30
|
requirements:
|
31
|
-
- - "
|
31
|
+
- - ">="
|
32
32
|
- !ruby/object:Gem::Version
|
33
33
|
version: '10.0'
|
34
34
|
type: :development
|
35
35
|
prerelease: false
|
36
36
|
version_requirements: !ruby/object:Gem::Requirement
|
37
37
|
requirements:
|
38
|
-
- - "
|
38
|
+
- - ">="
|
39
39
|
- !ruby/object:Gem::Version
|
40
40
|
version: '10.0'
|
41
41
|
- !ruby/object:Gem::Dependency
|
@@ -58,14 +58,14 @@ dependencies:
|
|
58
58
|
requirements:
|
59
59
|
- - "~>"
|
60
60
|
- !ruby/object:Gem::Version
|
61
|
-
version: '0.
|
61
|
+
version: '0.9'
|
62
62
|
type: :development
|
63
63
|
prerelease: false
|
64
64
|
version_requirements: !ruby/object:Gem::Requirement
|
65
65
|
requirements:
|
66
66
|
- - "~>"
|
67
67
|
- !ruby/object:Gem::Version
|
68
|
-
version: '0.
|
68
|
+
version: '0.9'
|
69
69
|
description:
|
70
70
|
email:
|
71
71
|
- jonathan.hefner@gmail.com
|
@@ -75,6 +75,7 @@ extra_rdoc_files: []
|
|
75
75
|
files:
|
76
76
|
- ".gitignore"
|
77
77
|
- ".travis.yml"
|
78
|
+
- CHANGELOG.md
|
78
79
|
- Gemfile
|
79
80
|
- LICENSE.txt
|
80
81
|
- README.md
|
@@ -101,10 +102,8 @@ required_rubygems_version: !ruby/object:Gem::Requirement
|
|
101
102
|
- !ruby/object:Gem::Version
|
102
103
|
version: '0'
|
103
104
|
requirements: []
|
104
|
-
|
105
|
-
rubygems_version: 2.4.8
|
105
|
+
rubygems_version: 3.0.1
|
106
106
|
signing_key:
|
107
107
|
specification_version: 4
|
108
108
|
summary: Input-focused benchmarking
|
109
109
|
test_files: []
|
110
|
-
has_rdoc:
|