d_heap 0.6.1 → 0.7.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/.clang-format +21 -0
- data/.github/workflows/main.yml +16 -1
- data/.rubocop.yml +1 -0
- data/CHANGELOG.md +17 -0
- data/{N → D} +1 -1
- data/README.md +313 -261
- data/d_heap.gemspec +16 -5
- data/docs/benchmarks-2.txt +79 -61
- data/docs/benchmarks.txt +587 -416
- data/docs/profile.txt +99 -133
- data/ext/d_heap/.rubocop.yml +7 -0
- data/ext/d_heap/d_heap.c +575 -424
- data/ext/d_heap/extconf.rb +34 -3
- data/images/push_n.png +0 -0
- data/images/push_n_pop_n.png +0 -0
- data/images/push_pop.png +0 -0
- data/lib/d_heap.rb +25 -1
- data/lib/d_heap/version.rb +1 -1
- metadata +6 -30
- data/.rspec +0 -3
- data/.travis.yml +0 -6
- data/Gemfile +0 -20
- data/Gemfile.lock +0 -83
- data/Rakefile +0 -20
- data/benchmarks/perf.rb +0 -29
- data/benchmarks/push_n.yml +0 -35
- data/benchmarks/push_n_pop_n.yml +0 -52
- data/benchmarks/push_pop.yml +0 -32
- data/benchmarks/stackprof.rb +0 -31
- data/bin/bench_charts +0 -13
- data/bin/bench_n +0 -7
- data/bin/benchmark-driver +0 -29
- data/bin/benchmarks +0 -10
- data/bin/console +0 -15
- data/bin/profile +0 -10
- data/bin/rake +0 -29
- data/bin/rspec +0 -29
- data/bin/rubocop +0 -29
- data/bin/setup +0 -8
- data/lib/benchmark_driver/runner/ips_zero_fail.rb +0 -158
- data/lib/d_heap/benchmarks.rb +0 -112
- data/lib/d_heap/benchmarks/benchmarker.rb +0 -116
- data/lib/d_heap/benchmarks/implementations.rb +0 -224
- data/lib/d_heap/benchmarks/profiler.rb +0 -71
- data/lib/d_heap/benchmarks/rspec_matchers.rb +0 -352
@@ -1,116 +0,0 @@
|
|
1
|
-
# frozen_string_literal: true
|
2
|
-
|
3
|
-
require "d_heap/benchmarks"
|
4
|
-
|
5
|
-
require "benchmark_driver"
|
6
|
-
require "shellwords"
|
7
|
-
require "English"
|
8
|
-
|
9
|
-
module DHeap::Benchmarks
|
10
|
-
# Benchmarks different implementations with different sizes
|
11
|
-
class Benchmarker
|
12
|
-
include Randomness
|
13
|
-
include Scenarios
|
14
|
-
|
15
|
-
N_COUNTS = [
|
16
|
-
5, # 1 + 4
|
17
|
-
21, # 1 + 4 + 16
|
18
|
-
85, # 1 + 4 + 16 + 64
|
19
|
-
341, # 1 + 4 + 16 + 64 + 256
|
20
|
-
1365, # 1 + 4 + 16 + 64 + 256 + 1024
|
21
|
-
5461, # 1 + 4 + 16 + 64 + 256 + 1024 + 4096
|
22
|
-
21_845, # 1 + 4 + 16 + 64 + 256 + 1024 + 4096 + 16384
|
23
|
-
87_381, # 1 + 4 + 16 + 64 + 256 + 1024 + 4096 + 16384 + 65536
|
24
|
-
].freeze
|
25
|
-
|
26
|
-
attr_reader :time
|
27
|
-
attr_reader :iterations_for_push_pop
|
28
|
-
attr_reader :io
|
29
|
-
|
30
|
-
def initialize(
|
31
|
-
time: Integer(ENV.fetch("BENCHMARK_TIME", 10)),
|
32
|
-
iterations_for_push_pop: 10_000,
|
33
|
-
io: $stdout
|
34
|
-
)
|
35
|
-
@time = time
|
36
|
-
@iterations_for_push_pop = Integer(iterations_for_push_pop)
|
37
|
-
@io = io
|
38
|
-
end
|
39
|
-
|
40
|
-
def call(queue_size: ENV.fetch("BENCHMARK_QUEUE_SIZE", :unset))
|
41
|
-
DHeap::Benchmarks.puts_version_info("Benchmarking")
|
42
|
-
sizes = (queue_size == :unset) ? N_COUNTS : [Integer(queue_size)]
|
43
|
-
sizes.each do |size|
|
44
|
-
benchmark_size(size)
|
45
|
-
end
|
46
|
-
end
|
47
|
-
|
48
|
-
def benchmark_size(size)
|
49
|
-
sep "#", "Benchmarks with N=#{size} (t=#{time}sec/benchmark)", big: true
|
50
|
-
io.puts
|
51
|
-
benchmark_push_n size
|
52
|
-
benchmark_push_n_then_pop_n size
|
53
|
-
benchmark_repeated_push_pop size
|
54
|
-
end
|
55
|
-
|
56
|
-
def benchmark_push_n(queue_size)
|
57
|
-
benchmarking("push N", "push_n", queue_size)
|
58
|
-
end
|
59
|
-
|
60
|
-
def benchmark_push_n_then_pop_n(queue_size)
|
61
|
-
benchmarking("push N then pop N", "push_n_pop_n", queue_size)
|
62
|
-
end
|
63
|
-
|
64
|
-
def benchmark_repeated_push_pop(queue_size)
|
65
|
-
benchmarking(
|
66
|
-
"Push/pop with pre-filled queue (size=N)", "push_pop", queue_size
|
67
|
-
)
|
68
|
-
end
|
69
|
-
|
70
|
-
private
|
71
|
-
|
72
|
-
# TODO: move somewhere else...
|
73
|
-
def skip_profiling?(queue_size, impl)
|
74
|
-
impl.klass == DHeap::Benchmarks::PushAndResort && 10_000 < queue_size
|
75
|
-
end
|
76
|
-
|
77
|
-
# rubocop:disable Metrics/MethodLength, Metrics/AbcSize
|
78
|
-
|
79
|
-
def benchmarking(name, file, size)
|
80
|
-
Bundler.with_unbundled_env do
|
81
|
-
sep "==", "#{name} (N=#{size})"
|
82
|
-
cmd = %W[
|
83
|
-
bin/benchmark-driver
|
84
|
-
--bundler
|
85
|
-
--run-duration 6
|
86
|
-
--timeout 15
|
87
|
-
--runner ips_zero_fail
|
88
|
-
benchmarks/#{file}.yml
|
89
|
-
]
|
90
|
-
if file == "push_n"
|
91
|
-
cmd << "--filter" << /dheap|\bstl\b|\bbsearch\b|\brb_heap\b/.to_s
|
92
|
-
end
|
93
|
-
env = ENV.to_h.merge(
|
94
|
-
"BENCH_N" => size.to_s,
|
95
|
-
"RUBYLIB" => File.expand_path("../..", __dir__),
|
96
|
-
)
|
97
|
-
system(env, *cmd)
|
98
|
-
end
|
99
|
-
end
|
100
|
-
|
101
|
-
def sep(sep, msg = "", width: 80, big: false)
|
102
|
-
txt = String.new
|
103
|
-
txt += "#{sep * (width / sep.length)}\n" if big
|
104
|
-
txt += sep
|
105
|
-
txt += " #{msg}" if msg && !msg.empty?
|
106
|
-
txt += " " unless big
|
107
|
-
txt += sep * ((width - txt.length) / sep.length) unless big
|
108
|
-
txt += "\n"
|
109
|
-
txt += "#{sep * (width / sep.length)}\n" if big
|
110
|
-
io.print txt
|
111
|
-
end
|
112
|
-
|
113
|
-
# rubocop:enable Metrics/MethodLength, Metrics/AbcSize
|
114
|
-
|
115
|
-
end
|
116
|
-
end
|
@@ -1,224 +0,0 @@
|
|
1
|
-
# frozen_string_literal: true
|
2
|
-
|
3
|
-
require "fc"
|
4
|
-
|
5
|
-
module DHeap::Benchmarks
|
6
|
-
|
7
|
-
# base class for example priority queues
|
8
|
-
class ExamplePriorityQueue
|
9
|
-
attr_reader :a
|
10
|
-
|
11
|
-
# quick initialization by simply sorting the array once.
|
12
|
-
def initialize(count = nil, &block)
|
13
|
-
@a = []
|
14
|
-
return unless count
|
15
|
-
count.times {|i| @a << block.call(i) }
|
16
|
-
@a.sort!
|
17
|
-
end
|
18
|
-
|
19
|
-
def clear
|
20
|
-
@a.clear
|
21
|
-
end
|
22
|
-
|
23
|
-
def empty?
|
24
|
-
@a.empty?
|
25
|
-
end
|
26
|
-
|
27
|
-
if ENV["LOG_LEVEL"] == "debug"
|
28
|
-
def dbg(msg)
|
29
|
-
puts "%20s: %p, %p" % [msg, @a.first, (@a[1..-1] || []).each_slice(2).to_a]
|
30
|
-
end
|
31
|
-
else
|
32
|
-
def dbg(msg) nil end
|
33
|
-
end
|
34
|
-
|
35
|
-
end
|
36
|
-
|
37
|
-
# The most naive approach--completely unsorted!--is ironically not the worst.
|
38
|
-
class FindMin < ExamplePriorityQueue
|
39
|
-
|
40
|
-
# O(1)
|
41
|
-
def <<(score)
|
42
|
-
raise ArgumentError unless score
|
43
|
-
@a.push score
|
44
|
-
end
|
45
|
-
|
46
|
-
# O(n)
|
47
|
-
def pop
|
48
|
-
return unless (score = @a.min)
|
49
|
-
index = @a.rindex(score)
|
50
|
-
@a.delete_at(index)
|
51
|
-
score
|
52
|
-
end
|
53
|
-
|
54
|
-
end
|
55
|
-
|
56
|
-
# Re-sorting after each insert: this both naive and performs the worst.
|
57
|
-
class Sorting < ExamplePriorityQueue
|
58
|
-
|
59
|
-
# O(n log n)
|
60
|
-
def <<(score)
|
61
|
-
raise ArgumentError unless score
|
62
|
-
@a.push score
|
63
|
-
@a.sort!
|
64
|
-
end
|
65
|
-
|
66
|
-
# O(1)
|
67
|
-
def pop
|
68
|
-
@a.shift
|
69
|
-
end
|
70
|
-
|
71
|
-
end
|
72
|
-
|
73
|
-
# A very simple example priority queue that is implemented with a sorted array.
|
74
|
-
#
|
75
|
-
# It uses Array#bsearch + Array#insert to push new values, and Array#pop to pop
|
76
|
-
# the min value.
|
77
|
-
class BSearch < ExamplePriorityQueue
|
78
|
-
|
79
|
-
# Array#bsearch_index is O(log n)
|
80
|
-
# Array#insert is O(n)
|
81
|
-
#
|
82
|
-
# So this should be O(n).
|
83
|
-
#
|
84
|
-
# In practice though, memcpy has a *very* small constant factor.
|
85
|
-
# And bsearch_index uses *exactly* (log n / log 2) comparisons.
|
86
|
-
def <<(score)
|
87
|
-
raise ArgumentError unless score
|
88
|
-
index = @a.bsearch_index {|other| score > other } || @a.length
|
89
|
-
@a.insert(index, score)
|
90
|
-
end
|
91
|
-
|
92
|
-
# Array#pop is O(1). It updates length without changing capacity or contents.
|
93
|
-
#
|
94
|
-
# No comparisons are necessary.
|
95
|
-
#
|
96
|
-
# shift is usually also O(1) and could be used if it were sorted normally.
|
97
|
-
def pop
|
98
|
-
@a.pop
|
99
|
-
end
|
100
|
-
|
101
|
-
end
|
102
|
-
|
103
|
-
# a very simple pure ruby binary heap
|
104
|
-
class RbHeap < ExamplePriorityQueue
|
105
|
-
|
106
|
-
def <<(value)
|
107
|
-
raise ArgumentError unless value
|
108
|
-
@a.push(value)
|
109
|
-
sift_up(@a.size - 1, value)
|
110
|
-
end
|
111
|
-
|
112
|
-
def pop
|
113
|
-
return if @a.empty?
|
114
|
-
popped = @a.first
|
115
|
-
value = @a.pop
|
116
|
-
last_index = @a.size - 1
|
117
|
-
return popped unless 0 <= last_index
|
118
|
-
|
119
|
-
sift_down(0, last_index, value)
|
120
|
-
popped
|
121
|
-
end
|
122
|
-
|
123
|
-
private
|
124
|
-
|
125
|
-
def sift_up(index, value = @a[index])
|
126
|
-
while 0 < index # rubocop:disable Style/NumericPredicate
|
127
|
-
parent_index = (index - 1) / 2
|
128
|
-
parent_value = @a[parent_index]
|
129
|
-
break if parent_value <= value
|
130
|
-
@a[index] = parent_value
|
131
|
-
index = parent_index
|
132
|
-
end
|
133
|
-
@a[index] = value
|
134
|
-
# check_heap!(index)
|
135
|
-
end
|
136
|
-
|
137
|
-
def sift_down(index, last_index = @a.size - 1, value = @a[index])
|
138
|
-
last_parent = (last_index - 1) / 2
|
139
|
-
while index <= last_parent
|
140
|
-
child_index, child_value = select_min_child(index, last_index)
|
141
|
-
break if value <= child_value
|
142
|
-
@a[index] = child_value
|
143
|
-
index = child_index
|
144
|
-
child_index = index * 2 + 1
|
145
|
-
end
|
146
|
-
@a[index] = value
|
147
|
-
end
|
148
|
-
|
149
|
-
def select_min_child(index, last_index = @a.size - 1)
|
150
|
-
child_index = index * 2 + 1
|
151
|
-
if child_index < last_index && a[child_index + 1] < @a[child_index]
|
152
|
-
child_index += 1
|
153
|
-
end
|
154
|
-
[child_index, @a[child_index]]
|
155
|
-
end
|
156
|
-
|
157
|
-
def check_heap!(idx)
|
158
|
-
check_heap_up!(idx)
|
159
|
-
check_heap_dn!(idx)
|
160
|
-
end
|
161
|
-
|
162
|
-
# compares index to its parent
|
163
|
-
def check_heap_at!(idx)
|
164
|
-
value = @a[idx]
|
165
|
-
unless idx <= 0
|
166
|
-
pidx = (idx - 1) / 2
|
167
|
-
pval = @a[pidx]
|
168
|
-
raise "@a[#{idx}] == #{value}, #{pval} > #{value}" if pval > value
|
169
|
-
end
|
170
|
-
value
|
171
|
-
end
|
172
|
-
|
173
|
-
def check_heap_up!(idx)
|
174
|
-
return if idx <= 0
|
175
|
-
pidx = (idx - 1) / 2
|
176
|
-
check_heap_at!(pidx)
|
177
|
-
check_heap_up!(pidx)
|
178
|
-
end
|
179
|
-
|
180
|
-
def check_heap_dn!(idx)
|
181
|
-
return unless @a.size <= idx
|
182
|
-
check_heap_at!(idx)
|
183
|
-
check_heap_down!(idx * 2 + 1)
|
184
|
-
check_heap_down!(idx * 2 + 2)
|
185
|
-
end
|
186
|
-
|
187
|
-
end
|
188
|
-
|
189
|
-
# minor adjustments to the "priority_queue_cxx" gem, to match the API
|
190
|
-
class CppSTL
|
191
|
-
|
192
|
-
def initialize
|
193
|
-
clear
|
194
|
-
end
|
195
|
-
|
196
|
-
def <<(value); @q.push(value, value) end
|
197
|
-
|
198
|
-
def clear
|
199
|
-
@q = FastContainers::PriorityQueue.new(:min)
|
200
|
-
end
|
201
|
-
|
202
|
-
def empty?
|
203
|
-
@q.empty?
|
204
|
-
end
|
205
|
-
|
206
|
-
def pop
|
207
|
-
@q.pop
|
208
|
-
rescue RuntimeError
|
209
|
-
nil
|
210
|
-
end
|
211
|
-
|
212
|
-
end
|
213
|
-
|
214
|
-
# Different duck-typed priority queue implemenations
|
215
|
-
IMPLEMENTATIONS = [
|
216
|
-
OpenStruct.new(name: " push and resort", klass: Sorting).freeze,
|
217
|
-
OpenStruct.new(name: " find min + del", klass: FindMin).freeze,
|
218
|
-
OpenStruct.new(name: "bsearch + insert", klass: BSearch).freeze,
|
219
|
-
OpenStruct.new(name: "ruby binary heap", klass: RbHeap).freeze,
|
220
|
-
OpenStruct.new(name: "C++STL PriorityQ", klass: CppSTL).freeze,
|
221
|
-
OpenStruct.new(name: "quaternary DHeap", klass: DHeap).freeze,
|
222
|
-
].freeze
|
223
|
-
|
224
|
-
end
|
@@ -1,71 +0,0 @@
|
|
1
|
-
# frozen_string_literal: true
|
2
|
-
|
3
|
-
require "d_heap/benchmarks"
|
4
|
-
|
5
|
-
require "ruby-prof"
|
6
|
-
|
7
|
-
module DHeap::Benchmarks
|
8
|
-
# Profiles different implementations with different sizes
|
9
|
-
class Profiler
|
10
|
-
include Randomness
|
11
|
-
include Scenarios
|
12
|
-
|
13
|
-
N_COUNTS = [
|
14
|
-
5, # 1 + 4
|
15
|
-
1365, # 1 + 4 + 16 + 64 + 256 + 1024
|
16
|
-
87_381, # 1 + 4 + 16 + 64 + 256 + 1024 + 4096 + 16384 + 65536
|
17
|
-
].freeze
|
18
|
-
|
19
|
-
def call(
|
20
|
-
queue_size: ENV.fetch("PROFILE_QUEUE_SIZE", :unset),
|
21
|
-
iterations: ENV.fetch("PROFILE_ITERATIONS", 1_000_000)
|
22
|
-
)
|
23
|
-
DHeap::Benchmarks.puts_version_info("Profiling")
|
24
|
-
fill_random_vals
|
25
|
-
sizes = queue_size == :unset ? N_COUNTS : [Integer(queue_size)]
|
26
|
-
sizes.each do |size|
|
27
|
-
profile_all(size, iterations)
|
28
|
-
end
|
29
|
-
end
|
30
|
-
|
31
|
-
def profile_all(queue_size, iterations, io: $stdout)
|
32
|
-
io.puts <<~TEXT
|
33
|
-
########################################################################
|
34
|
-
# Profile w/ N=#{queue_size} (i=#{iterations})
|
35
|
-
# (n.b. RubyProf & tracepoint can change relative performance.
|
36
|
-
# A sampling profiler can provide more accurate relative metrics.
|
37
|
-
########################################################################
|
38
|
-
|
39
|
-
TEXT
|
40
|
-
DHeap::Benchmarks::IMPLEMENTATIONS.each do |impl|
|
41
|
-
profile_one(impl, queue_size, iterations, io: io)
|
42
|
-
end
|
43
|
-
end
|
44
|
-
|
45
|
-
# TODO: move somewhere else...
|
46
|
-
def skip_profiling?(queue_size, impl)
|
47
|
-
impl.klass == DHeap::Benchmarks::Sorting && 10_000 < queue_size
|
48
|
-
end
|
49
|
-
|
50
|
-
def profile_one(impl, queue_size, iterations, io: $stdout)
|
51
|
-
return if skip_profiling?(queue_size, impl)
|
52
|
-
io.puts "Filling #{impl.name} ---------------------------"
|
53
|
-
queue = impl.klass.new
|
54
|
-
push_n(queue, queue_size)
|
55
|
-
io.puts "Profiling #{impl.name} ---------------------------"
|
56
|
-
profiling do
|
57
|
-
repeated_push_pop(queue, iterations)
|
58
|
-
end
|
59
|
-
end
|
60
|
-
|
61
|
-
def profiling(io: $stdout, &block)
|
62
|
-
# do the thing
|
63
|
-
result = RubyProf.profile(&block)
|
64
|
-
# report_the_thing
|
65
|
-
printer = RubyProf::FlatPrinter.new(result)
|
66
|
-
printer.print($stdout, min_percent: 1.0)
|
67
|
-
io.puts
|
68
|
-
end
|
69
|
-
|
70
|
-
end
|
71
|
-
end
|
@@ -1,352 +0,0 @@
|
|
1
|
-
# frozen_string_literal: true
|
2
|
-
|
3
|
-
require "d_heap/benchmarks"
|
4
|
-
|
5
|
-
module DHeap::Benchmarks
|
6
|
-
|
7
|
-
# Profiles different implementations with different sizes
|
8
|
-
module RSpecMatchers # rubocop:disable Metrics/ModuleLength
|
9
|
-
extend RSpec::Matchers::DSL
|
10
|
-
|
11
|
-
# Assert ips (iterations per second):
|
12
|
-
#
|
13
|
-
# expect { ... }.to perform_at_least(1_000_000).ips
|
14
|
-
# .running_at_least(10).times # optional, defaults to 1
|
15
|
-
# .running_at_least(10).seconds # optional, defaults to 1s
|
16
|
-
# .running_at_most(10_000_000).times # optional, defaults to nil
|
17
|
-
# .running_at_most(2).seconds # optional, defaults to 2s
|
18
|
-
# .warmup_at_most(1000).times # optional, defaults to 1k
|
19
|
-
# .warmup_at_most(0.100).seconds # optional, defaults to 0.1s
|
20
|
-
# .iterations_per_round # optional, defaults to 1
|
21
|
-
# .and_at_least(1.1).times.faster_than { ... } # can also compare
|
22
|
-
#
|
23
|
-
# Assert comparison (and optionally runtime or ips):
|
24
|
-
#
|
25
|
-
# expect { ... }.to perform_at_least(2.5).times_faster_than { ... }
|
26
|
-
# .running_at_least(10).times # optional, defaults to 1
|
27
|
-
# .running_at_least(10).seconds # optional, defaults to 1s
|
28
|
-
# .running_at_most(10_000_000).times # optional, defaults to nil
|
29
|
-
# .running_at_most(2).seconds # optional, defaults to 2s
|
30
|
-
# .warmup_at_most(1000).times # optional, defaults to 1k
|
31
|
-
# .warmup_at_most(0.100).seconds # optional, defaults to 0.1s
|
32
|
-
# .iterations_per_call # optional, defaults to 1
|
33
|
-
# .and_at_least(100).ips { ... } # can also assert ips
|
34
|
-
#
|
35
|
-
# n.b: Given a known constant number of iterations, run time and ips are both
|
36
|
-
# measuring the same underlying metric.
|
37
|
-
#
|
38
|
-
# rubocop:disable Metrics/BlockLength, Layout/SpaceAroundOperators
|
39
|
-
matcher :perform_at_least do |expected|
|
40
|
-
supports_block_expectations
|
41
|
-
|
42
|
-
%i[
|
43
|
-
is_at_least
|
44
|
-
running_at_most
|
45
|
-
running_at_least
|
46
|
-
warmup_at_most
|
47
|
-
].each do |type|
|
48
|
-
chain type do |number|
|
49
|
-
reason, value = ___number_reason_and_value___
|
50
|
-
if reason || value
|
51
|
-
raise "Need to handle unit-less number first: %s(%p)" % [reason, value]
|
52
|
-
end
|
53
|
-
@number_for = type
|
54
|
-
@number_val = number
|
55
|
-
end
|
56
|
-
end
|
57
|
-
|
58
|
-
alias_method :and_at_least, :is_at_least
|
59
|
-
|
60
|
-
%i[
|
61
|
-
times
|
62
|
-
seconds
|
63
|
-
milliseconds
|
64
|
-
].each do |unit|
|
65
|
-
chain unit do
|
66
|
-
reason, value = ___number_reason_and_value___
|
67
|
-
raise "No number was specified" unless reason && value
|
68
|
-
apply_number_to_reason(reason, value, unit)
|
69
|
-
@number_for = @number_val = nil
|
70
|
-
end
|
71
|
-
end
|
72
|
-
|
73
|
-
# TODO: let IPS set time to run instead of iterations to run
|
74
|
-
chain :ips do
|
75
|
-
reason, value = ___number_reason_and_value___
|
76
|
-
raise "'ips' unit is only for assertions" unless reason == :is_at_least
|
77
|
-
raise "Already asserting %s ips" % [@expect_ips] if @expect_ips
|
78
|
-
raise "'ips' assertion has already been made" if @expect_ips
|
79
|
-
raise "Unknown assertion count" unless value
|
80
|
-
@expect_ips = Integer(value)
|
81
|
-
@number_for = @number_val = nil
|
82
|
-
end
|
83
|
-
|
84
|
-
# need to use method because "chain" can't take a block
|
85
|
-
def times_faster_than(&other)
|
86
|
-
reason, value = ___number_reason_and_value___
|
87
|
-
raise "'times_faster_than' is only for assertions" unless reason == :is_at_least
|
88
|
-
raise "Already asserting %sx comparison" % [@expect_cmp] if @expect_cmp
|
89
|
-
raise ArgumentError, "must provide a proc" unless other
|
90
|
-
@expect_cmp = Float(value)
|
91
|
-
@cmp_proc = other
|
92
|
-
@number_for = @number_val = nil
|
93
|
-
self
|
94
|
-
end
|
95
|
-
|
96
|
-
chain :loudly do @volume = :loud end
|
97
|
-
chain :quietly do @volume = :quiet end
|
98
|
-
chain :volume do |volume|
|
99
|
-
raise "Invalid volume" unless %i[loud quiet].include?(volume)
|
100
|
-
@volume = volume
|
101
|
-
end
|
102
|
-
|
103
|
-
chain :iterations_per_round do |iterations|
|
104
|
-
if @iterations_per_round
|
105
|
-
raise "Already set iterations per round (%p)" [@iterations_per_round]
|
106
|
-
end
|
107
|
-
@iterations_per_round = Integer(iterations)
|
108
|
-
end
|
109
|
-
|
110
|
-
match do |actual|
|
111
|
-
require "benchmark"
|
112
|
-
raise "Need to expect a proc or block" unless actual.respond_to?(:to_proc)
|
113
|
-
raise "Need a performance assertion" unless assertion?
|
114
|
-
@actual_proc = actual
|
115
|
-
prepare_for_measurement
|
116
|
-
if @max_iter && (@max_iter % @iterations_per_round) != 0
|
117
|
-
raise "Iterations per round (%p) must divide evenly by max iterations (%p)" % [
|
118
|
-
@iterations_per_round, @max_iter,
|
119
|
-
]
|
120
|
-
end
|
121
|
-
run_measurements
|
122
|
-
cmp_okay? && ips_okay?
|
123
|
-
end
|
124
|
-
|
125
|
-
description do
|
126
|
-
[
|
127
|
-
@expect_cmp && cmp_okay_msg,
|
128
|
-
@expect_ips && ips_okay_msg,
|
129
|
-
].join(", and ")
|
130
|
-
end
|
131
|
-
|
132
|
-
failure_message do
|
133
|
-
[
|
134
|
-
cmp_okay? ? nil : "expected to #{cmp_okay_msg} but #{cmp_fail_msg}", # =>
|
135
|
-
ips_okay? ? nil : "expected to #{ips_okay_msg} but #{ips_fail_msg}",
|
136
|
-
].compact.join(", and ")
|
137
|
-
end
|
138
|
-
|
139
|
-
private
|
140
|
-
|
141
|
-
chain :__convert_expected_to_ivars__ do
|
142
|
-
@number_val ||= expected
|
143
|
-
@number_for ||= :is_at_least if @number_val
|
144
|
-
expected = nil
|
145
|
-
end
|
146
|
-
private :__convert_expected_to_ivars__
|
147
|
-
|
148
|
-
def ___number_reason_and_value___
|
149
|
-
__convert_expected_to_ivars__
|
150
|
-
[@number_for, @number_val]
|
151
|
-
end
|
152
|
-
|
153
|
-
def apply_number_to_reason(reason, value, unit)
|
154
|
-
normalized_value, normalized_unit = normalize_unit(unit)
|
155
|
-
case reason
|
156
|
-
when :running_at_most; apply_max_run normalized_value, normalized_unit
|
157
|
-
when :running_at_least; apply_min_run normalized_value, normalized_unit
|
158
|
-
when :warmup_at_most; apply_warmup normalized_value, normalized_unit
|
159
|
-
else raise "%s is incompatible with %s(%p)" % [unit, reason, value]
|
160
|
-
end
|
161
|
-
end
|
162
|
-
|
163
|
-
def normalize_unit(unit)
|
164
|
-
case unit
|
165
|
-
when :seconds; [Float(@number_val), :seconds]
|
166
|
-
when :milliseconds; [Float(@number_val) / 1000.0, :seconds]
|
167
|
-
when :times; [Integer(@number_val), :times]
|
168
|
-
else raise "Invalid unit %s for %s(%p)" % [unit, reason, value]
|
169
|
-
end
|
170
|
-
end
|
171
|
-
|
172
|
-
def apply_min_run(value, unit)
|
173
|
-
case unit
|
174
|
-
when :seconds; @min_time = value
|
175
|
-
when :times; @min_iter = value
|
176
|
-
end
|
177
|
-
end
|
178
|
-
|
179
|
-
def apply_max_run(value, unit)
|
180
|
-
case unit
|
181
|
-
when :seconds; @max_time = value
|
182
|
-
when :times; @max_iter = value
|
183
|
-
end
|
184
|
-
end
|
185
|
-
|
186
|
-
def apply_warmup(value, unit)
|
187
|
-
case unit
|
188
|
-
when :seconds; @warmup_time = value
|
189
|
-
when :times; @warmup_iter = value
|
190
|
-
end
|
191
|
-
end
|
192
|
-
|
193
|
-
def prepare_for_measurement
|
194
|
-
@volume ||= ENV.fetch("RSPEC_BENCHMARK_VOLUME", :quiet).to_sym
|
195
|
-
@max_time ||= 2
|
196
|
-
@min_time ||= 1
|
197
|
-
@min_iter ||= 1
|
198
|
-
@warmup_time ||= 0.100
|
199
|
-
@warmup_iter ||= 1000
|
200
|
-
@iterations_per_round ||= 1
|
201
|
-
nil
|
202
|
-
end
|
203
|
-
|
204
|
-
def run_measurements
|
205
|
-
puts header if loud?
|
206
|
-
warmup
|
207
|
-
take_measurements
|
208
|
-
end
|
209
|
-
|
210
|
-
def header
|
211
|
-
max_rounds = @max_iter && @max_iter / @iterations_per_round
|
212
|
-
[
|
213
|
-
"Warmup time %s, or iterations: %s" % [@min_iter, @max_iter],
|
214
|
-
"Benchmark time (%s..%s) or iterations (%s..%s), max rounds: %p" % [
|
215
|
-
@min_time, @max_time, @min_iter, @max_iter, max_rounds,
|
216
|
-
],
|
217
|
-
"%-10s %s" % ["", Benchmark::CAPTION],
|
218
|
-
].join("\n")
|
219
|
-
end
|
220
|
-
|
221
|
-
def warmup
|
222
|
-
return unless 0 < @warmup_time && 0 < @warmup_iter # rubocop:disable Style/NumericPredicate
|
223
|
-
args = [@warmup_iter, 0, @warmup_time, 1, @warmup_iter]
|
224
|
-
measure("warmup", *args, &@actual_proc)
|
225
|
-
measure("warmup cmp", *args, &@cmp_proc) if @cmp_proc
|
226
|
-
end
|
227
|
-
|
228
|
-
def take_measurements
|
229
|
-
args = [@iterations_per_round, @min_time, @max_time, @min_iter, @max_iter]
|
230
|
-
@actual_tms = measure("actual", *args, &@actual_proc)
|
231
|
-
@cmp_tms = measure("cmp", *args, &@cmp_proc) if @cmp_proc
|
232
|
-
return unless @cmp_proc
|
233
|
-
# how many times faster?
|
234
|
-
@actual_cmp = @actual_tms.ips_real / @cmp_tms.ips_real
|
235
|
-
puts "Ran %0.3fx as fast as comparison" % [@actual_cmp] if loud?
|
236
|
-
end
|
237
|
-
|
238
|
-
def loud?; @volume == :loud end
|
239
|
-
|
240
|
-
def assertion?; !!(@expect_cmp || @expect_ips) end
|
241
|
-
|
242
|
-
def cmp_okay?; !@expect_cmp || @expect_cmp < @actual_cmp end
|
243
|
-
def ips_okay?; !@expect_tms || @expect_tms.ips < @actual_tms.ips end
|
244
|
-
|
245
|
-
def measure(name, ipr, *args)
|
246
|
-
measurements = TmsMeasurements.new(name, ipr, *args)
|
247
|
-
measurements.max_rounds.times do
|
248
|
-
# GC.start(full_mark: true, immediate_sweep: true)
|
249
|
-
# GC.compact
|
250
|
-
measurements << Benchmark.measure do
|
251
|
-
yield ipr
|
252
|
-
end
|
253
|
-
# p measurements.real
|
254
|
-
break if measurements.max_time < measurements.real
|
255
|
-
end
|
256
|
-
log_measurement(name, measurements)
|
257
|
-
measurements
|
258
|
-
end
|
259
|
-
|
260
|
-
# rubocop:disable Metrics/AbcSize
|
261
|
-
def units_str(num)
|
262
|
-
if num >= 10**12; "%7.3fT" % [num.to_f / 10**12]
|
263
|
-
elsif num >= 10** 9; "%7.3fB" % [num.to_f / 10** 9]
|
264
|
-
elsif num >= 10** 6; "%7.3fM" % [num.to_f / 10** 6]
|
265
|
-
elsif num >= 10** 3; "%7.3fk" % [num.to_f / 10** 3]
|
266
|
-
else "%7.3f" % [num.to_f]
|
267
|
-
end
|
268
|
-
end
|
269
|
-
# rubocop:enable Metrics/AbcSize
|
270
|
-
|
271
|
-
def log_measurement(name, measurements)
|
272
|
-
return unless loud?
|
273
|
-
puts "%-10s %s => %s ips (%d rounds)" % [
|
274
|
-
name,
|
275
|
-
measurements.tms.to_s.rstrip,
|
276
|
-
units_str(measurements.ips_real),
|
277
|
-
measurements.size,
|
278
|
-
]
|
279
|
-
end
|
280
|
-
|
281
|
-
def cmp_okay_msg; "run %0.2fx faster" % [@expect_cmp] end
|
282
|
-
def cmp_fail_msg; "was only %0.2fx as fast" % [@actual_cmp] end
|
283
|
-
def ips_okay_msg; "run with %s ips" % [units_str(@expect_ips)] end
|
284
|
-
def ips_fail_msg; "was only %s ips" % [units_str(@actual_ips)] end
|
285
|
-
|
286
|
-
end
|
287
|
-
# rubocop:enable Metrics/BlockLength, Layout/SpaceAroundOperators
|
288
|
-
|
289
|
-
alias_matcher :perform_with, :perform
|
290
|
-
|
291
|
-
end
|
292
|
-
|
293
|
-
# Replicates a subset of the functionality in benchmark-ips
|
294
|
-
#
|
295
|
-
# TODO: merge this with benchmark-ips
|
296
|
-
# TODO: implement (or remove) min_time, min_iter
|
297
|
-
class TmsMeasurements
|
298
|
-
attr_reader :iterations_per_entry
|
299
|
-
attr_reader :iterations
|
300
|
-
|
301
|
-
attr_reader :min_time
|
302
|
-
attr_reader :max_time
|
303
|
-
|
304
|
-
attr_reader :min_iter
|
305
|
-
attr_reader :max_iter
|
306
|
-
|
307
|
-
def initialize(name, ipe, min_time, max_time, min_iter, max_iter) # rubocop:disable Metrics/ParameterLists
|
308
|
-
@name = name
|
309
|
-
@iterations_per_entry = Integer(ipe)
|
310
|
-
@min_time = Float(min_time)
|
311
|
-
@max_time = Float(max_time)
|
312
|
-
@min_iter = Integer(min_iter)
|
313
|
-
@max_iter = Integer(max_iter)
|
314
|
-
@entries = []
|
315
|
-
@sum = Benchmark::Tms.new
|
316
|
-
@iterations = 0
|
317
|
-
end
|
318
|
-
|
319
|
-
def size; entries.size end
|
320
|
-
|
321
|
-
def <<(tms)
|
322
|
-
raise TypeError, "not a #{Benchmark::Tms}" unless tms.is_a?(Benchmark::Tms)
|
323
|
-
raise IndexError, "full" if @max_iter <= size
|
324
|
-
@sum += tms
|
325
|
-
@iterations += @iterations_per_entry
|
326
|
-
@entries << tms
|
327
|
-
self
|
328
|
-
end
|
329
|
-
|
330
|
-
def sum; @sum.dup end
|
331
|
-
alias tms sum
|
332
|
-
|
333
|
-
def entries; @entries.dup end
|
334
|
-
|
335
|
-
def cstime; @sum.cstime end
|
336
|
-
def cutime; @sum.cutime end
|
337
|
-
def real; @sum.real end
|
338
|
-
def stime; @sum.stime end
|
339
|
-
def total; @sum.total end
|
340
|
-
def utime; @sum.utime end
|
341
|
-
|
342
|
-
def ips_real; @iterations / real end
|
343
|
-
def ips_total; @iterations / total end
|
344
|
-
def ips_utime; @iterations / utime end
|
345
|
-
|
346
|
-
def max_rounds
|
347
|
-
@max_iter && @max_iter / @iterations_per_entry
|
348
|
-
end
|
349
|
-
|
350
|
-
end
|
351
|
-
|
352
|
-
end
|