* spec/default.mspec: removed -I options for useless or non-
existent paths from flags. there is no library scripts and .ext
directory in the source directory.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58949 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* misc/lldb_cruby.py (lldb_init): get constants from
ruby_dummy_gdb_enums in the target.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58943 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby.c (load_file): move opened file to an argument, to reduce
open/close calls in the near future.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58938 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby.c (ruby_cmdline_options_t): reordered members and turned
simple flags into bit fields to reduce the size (136->104 on
LP64).
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58937 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
I miscalculated for r58934, since we sleep 100ms, and
the worst possible case is 100 Hz in the kernel meaning
we only have 10ms resolution. So, we need to increase
our CPU percentage to >= 10% for this.
This should be more than enough for our CI machines which
have 300 Hz kernels [ruby-core:81429]:
http://ci.rvm.jp/results/trunk-test@sasada-8core/1495942555
* test/ruby/test_io.rb (test_copy_stream_no_busy_wait):
override default percentage
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58936 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
if getrusage(2) is available, to improve precision of Process.times and
its user like lib/benchmark.rb.
On macOS, since getrusage(2) has better precision than times(3),
they are much improved like:
* Before
Process.times
=> #<struct Process::Tms utime=0.56, stime=0.35, cutime=0.04, cstime=0.03>
puts Benchmark.measure { "a" * 1_000_000_000 }
0.340000 0.310000 0.650000 ( 0.674025)
* After
Process.times
=> #<struct Process::Tms utime=0.561899, stime=0.35076, cutime=0.046483, cstime=0.038929>
puts Benchmark.measure { "a" * 1_000_000_000 }
0.343223 0.310037 0.653260 ( 0.674025)
On Linux, since struct rusage from getrusage(2) is used instead of struct tms
from times(2), they are slightly improved like:
* Before
Process.times
=> #<struct Process::Tms utime=0.43, stime=0.11, cutime=0.0, cstime=0.0>
puts Benchmark.measure { "a" * 1_000_000_000 }
0.120000 0.040000 0.170000 ( 0.171621)
* After
Process.times
=> #<struct Process::Tms utime=0.432, stime=0.116, cutime=0.0, cstime=0.0>
puts Benchmark.measure { "a" * 1_000_000_000 }
0.124000 0.048000 0.172000 ( 0.171621)
[ruby-dev:49471] [Feature #11952]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58935 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
When sleeping for the tick rate of 100ms (defined in
thread_pthread.c) as we do in test/ruby/test_io.rb
(test_copy_stream_no_busy_wait), it may not be possible to
measure with <= 10ms resolution on 100HZ systems (CONFIG_HZ in
the Linux kernel). So increase the threshold to 15ms (10ms +
5ms slack for slow systems).
* test/lib/test/unit/assertions.rb (assert_cpu_usage_low):
increase pct default value [ruby-core:81427]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58934 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Instead, match the poll() implementation used on Linux for now;
as the Linux poll(2) manpage describes using negative FD to
easily ignore an FD in a larger FD set while (sleeping the given
timeout). I'm not entirely sure if matching poll() behavior
is a good idea for a single FD, but it's better than segfaulting
or NoMemoryError.
* thread.c (init_set_fd): ignore negative FD
* test/-ext-/wait_for_single_fd/test_wait_for_single_fd.rb
(test_wait_for_invalid_fd): check values which may trigger
segfaults or OOM
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58925 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
I'm likely to make similar mistakes in the future when working
on Fiber auto-scheduling. Start adding assertions for existing
code, first.
* test/ruby/test_io.rb (test_copy_stream_no_busy_wait): added
* test/lib/test/unit/assertions.rb (assert_cpu_usage_low): added
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58924 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This reverts r58919.
Apparently skipping exts is intentional since r58759
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58923 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
rational.c (i_gcd): replace GCD algorithm from Euclidean algorithm to Stein
algorithm (https://en.wikipedia.org/wiki/Binary_GCD_algorithm).
Some Time methods will call internal quov() function and it calls
Rational#quo -> f_muldiv() -> i_gcd() in rational.c
And some Rational methods also call i_gcd().
The implementation of Euclidean algorithm spent a long time at modulo
operation (ie "x = y % x;").
The Stein algorithm will replace with shift operation which is faster
than modulo.
Time#subsec -> 36 % up
Time#to_r -> 26 % up
Rational#+ -> 14 % up
Rational#- -> 15 % up
Rational#* -> 13 % up
[ruby-core:80843] [Bug #13503] [Fix GH-1596]
### Before
Time#subsec 2.142M (± 9.8%) i/s - 10.659M in 5.022659s
Time#to_r 2.003M (± 9.1%) i/s - 9.959M in 5.012445s
Rational#+ 3.843M (± 0.9%) i/s - 19.274M in 5.016254s
Rational#- 3.820M (± 1.3%) i/s - 19.149M in 5.014137s
Rational#* 5.198M (± 1.4%) i/s - 26.016M in 5.005664s
* After
Time#subsec 2.902M (± 2.9%) i/s - 14.505M in 5.001815s
Time#to_r 2.503M (± 4.8%) i/s - 12.512M in 5.011454s
Rational#+ 4.390M (± 1.2%) i/s - 22.001M in 5.012413s
Rational#- 4.391M (± 1.2%) i/s - 22.013M in 5.014584s
Rational#* 5.872M (± 2.2%) i/s - 29.369M in 5.003666s
* Test code
require 'benchmark/ips'
Benchmark.ips do |x|
x.report "Time#subsec" do |t|
time = Time.now
t.times { time.subsec }
end
x.report "Time#to_r" do |t|
time = Time.now
t.times { time.to_r }
end
x.report "Rational#+" do |t|
rat1 = 1/2r
rat2 = 1/3r
t.times { rat1 + rat2 }
end
x.report "Rational#-" do |t|
rat1 = 1/3r
rat2 = 1/2r
t.times { rat1 - rat2 }
end
x.report "Rational#*" do |t|
rat1 = 1/3r
rat2 = 1/2r
t.times { rat1 * rat2 }
end
end
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58922 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This allows me to test changes to ext/ more easily by only typing:
make test-all TESTS=/path/to/ext/test_foo.rb
I spent a few minutes wondering what was wrong before I realized
changes to exts were not taking effect.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58919 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* spec/mspec/lib/mspec/commands/mspec.rb (MSpecMain#multi_exec):
as multi_exec children must run with yaml formatter, append the
option for it after other options to override another formatter
option with a warning if it is given.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58918 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
to skip unnecessary string allocation on frozen_string_literal: false.
str_uplus can bypass calling rb_str_dup when OBJ_FROZEN is true.
* Before
erb_render 1.064
* Afete
erb_render 0.909
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58916 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
from bm_app_erb_render.rb.
I'm told from ko1 that bm_app_* is namespace for Ruby applications,
not for ERB and we should use bm_erb_* for ERB benchmark instead.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58915 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* numeric.c (flo_floor, flo_ceil): should not return zero for small
number. [ruby-core:81394] [Bug #13599]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58913 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* array.c (rb_ary_concat_multi): concatenate the array without generating
temporary Array object if only one argument is given.
This is very similar with r58886.
Array#concat will be faster around 19%.
[Fix GH-1634]
### Before
Array#concat 2.187M (± 3.5%) i/s - 10.926M in 5.002829s
### After
Array#concat 2.598M (± 1.8%) i/s - 13.008M in 5.008201s
### Test code
require 'benchmark/ips'
Benchmark.ips do |x|
x.report "Array#concat" do |i|
other = [4]
i.times { [1, 2, 3].concat(other) }
end
end
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58909 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
by reducing string allocation.
* Before
app_erb 0.687
* After
app_erb 0.679
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58908 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
to skip object allocation for static string.
We can't always enable frozen_string_literal pragma because we can't
freeze string literals embedded by user for backward compatibility.
So we need to use fstring for each static string.
Since adding ".freeze" to string literals in #content_dump is slow
on compiling, I used unary "-" operator instead.
benchmark/bm_app_erb_render.rb: Added rendering-only benchmark to
test rendering performance on production environment.
This benchmark is created to reproduce the behavior on Sinatra (Tilt).
Thus it doesn't use ERB#result to skip parsing compiled code.
It doesn't use ERB#def_method too to regard `title` and `content` as
local variables. If we use #def_method, `title` and `content` needs
to be method call. I wanted to avoid it.
This patch's benchmark results is:
* Before
app_erb_render 1.250
app_erb 0.704
* After
app_erb_render 1.066
app_erb 0.686
This patch optimizes rendering performance (app_erb_render) without
spoiling (total of rendering +) compiling performance (app_erb).
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58905 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
The original intention of introducing `_erbout.force_encoding`
in r21170 was:
- "returns a string in the same character encoding as the input string."
- "When the input string has a magic comment, however, it returns a string
in the encoding specified by the magic comment."
And they are tested by test/erb/test_erb_m17n.rb well and this patch
passes the test.
Since magic comment is always added in ERB compiled code, using ''.dup
instead of String.new will set correct encoding without calling
force_encoding method.
The benchmark results are:
* Before
$ ./ruby benchmark/run.rb --matzruby=./ruby -m bm_app_erb
MatzRuby:
ruby 2.5.0dev (2017-05-26 skip-force-enc.. 58903) [x86_64-linux]
last_commit=Skip force_encoding in compiled code of erb
Ruby:
app_erb:
matz 0.715
* After
$ ./ruby benchmark/run.rb --matzruby=./ruby -m bm_app_erb
MatzRuby:
ruby 2.5.0dev (2017-05-26 skip-force-enc.. 58903) [x86_64-linux]
last_commit=Skip force_encoding in compiled code of erb
Ruby:
app_erb:
matz 0.672
And perf(1) results are:
* Before
$ sudo perf stat ./ruby benchmark/bm_app_erb.rb
Performance counter stats for './ruby benchmark/bm_app_erb.rb':
709.571746 task-clock (msec) # 1.000 CPUs utilized
5 context-switches # 0.007 K/sec
1 cpu-migrations # 0.001 K/sec
1,337 page-faults # 0.002 M/sec
3,088,936,521 cycles # 4.353 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
4,849,564,282 instructions # 1.57 insns per cycle
1,027,042,087 branches # 1447.411 M/sec
19,983,456 branch-misses # 1.95% of all branches
0.709747823 seconds time elapsed
* After
$ sudo perf stat ./ruby benchmark/bm_app_erb.rb
Performance counter stats for './ruby benchmark/bm_app_erb.rb':
693.494673 task-clock (msec) # 1.000 CPUs utilized
7 context-switches # 0.010 K/sec
1 cpu-migrations # 0.001 K/sec
1,316 page-faults # 0.002 M/sec
3,025,639,349 cycles # 4.363 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
4,694,848,271 instructions # 1.55 insns per cycle
994,496,704 branches # 1434.037 M/sec
19,693,239 branch-misses # 1.98% of all branches
0.693724345 seconds time elapsed
[fix GH-1147]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58904 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
