Performance regression in 3.16.14 series? #150

ewoodh2o · 2014-11-05T19:15:42Z

We upgraded libv8 to the 3.16.14 series last week and started seeing significantly slower asset compilation times (through Uglifier) compared to v3.11.8.17. I came at this through Rails' Asset Pipeline, so I'm way in the weeds here and don't know if it's libv8 or V8 itself, but thought I'd ask here to see if anyone had similar experiences.

This doesn't seem to affect times on my dev machine (OS X 10.10, Intel Core i7), but it's between 2x and 5x slower on our production servers (Scientific Linux 6.5, AMD). Seems to be worse for larger files (obviously in strict wall-clock time, but also closer to 5x+ slower on the big files).

Benchmarks for Uglifier.compile(), which is just the easiest way I know how to test:

_libv8-3.11.8.17-x86_64-linux_

Compiling bootstrap.js (62kb, not minified)
  0.700000   0.030000   0.730000 (  0.721680)
  0.670000   0.030000   0.700000 (  0.688308)

Compiling highcharts.js (496kb, not minified)
  4.510000   0.040000   4.550000 (  4.499848)
  4.250000   0.030000   4.280000 (  4.238179)

Compiling jquery-1.8.1.min.js (93kb, already minified)
  2.630000   0.040000   2.670000 (  2.639083)
  2.540000   0.030000   2.570000 (  2.548431)

Compiling jquery-ui-1.8.23.min.js (205kb, already minified)
  3.320000   0.070000   3.390000 (  3.345508)
  3.360000   0.070000   3.430000 (  3.382746)

Compiling application.js (900kb, mixed)
 15.020000   0.300000  15.320000 ( 15.129480)
 16.170000   0.180000  16.350000 ( 16.162925)

_libv8-3.16.14.3-x86_64-linux_

Compiling bootstrap.js (62kb, not minified)
  0.980000   0.010000   0.990000 (  0.976608)
  0.860000   0.020000   0.880000 (  0.865921)

Compiling highcharts.js (496kb, not minified)
  8.720000   0.080000   8.800000 (  8.736296)
  7.230000   0.060000   7.290000 (  7.230722)

Compiling jquery-1.8.1.min.js (93kb, already minified)
  4.110000   0.080000   4.190000 (  4.153838)
  3.830000   0.030000   3.860000 (  3.830293)

Compiling jquery-ui-1.8.23.min.js (205kb, already minified)
  7.260000   0.060000   7.320000 (  7.273115)
  6.580000   0.030000   6.610000 (  6.579161)

Compiling application.js (900kb, mixed)
 82.280000   0.190000  82.470000 ( 82.314145)
 87.490000   0.220000  87.710000 ( 87.526244)

Anyone experiencing anything similar, or have any ideas on what might cause this?

The text was updated successfully, but these errors were encountered:

ledbettj · 2014-11-12T20:26:37Z

Here are some ruby-prof traces of the two versions mentioned above, simply using Uglifier to compress a 1.4mb javascript file. 3.16.14.7 took approximately 70 seconds, while 3.11.8.17 completed in about 18.

V8::C::String#Utf8Value seems to be the culprit at first glance.

$ uname -a
Linux throttle 3.14.1-1-ARCH #1 SMP PREEMPT Mon Apr 14 20:40:47 CEST 2014 x86_64 GNU/Linux

john@throttle:~/Projects/test$ be ./test.rb
Uglifier 2.5.3 libv8 3.16.14.7
Thread ID: 5649160
Fiber ID: 5500900
Total: 70.067854
Sort by: self_time

 %self      total      self      wait     child     calls  name
 73.85     51.749    51.749     0.000     0.000        1   V8::C::String#Utf8Value
 26.08     18.272    18.272     0.000     0.000        1   V8::C::Script#Run
  0.03      0.026     0.018     0.000     0.008        1   JSON::Ext::Generator::State#generate
  0.02      0.017     0.017     0.000     0.000        1   <Class::V8::C::Script>#New
  0.01      0.007     0.007     0.000     0.000       55   Symbol#to_s
  0.00     70.041     0.001     0.000    70.040        1   ExecJS::RubyRacerRuntime::Context#exec
  0.00     70.039     0.001     0.000    70.038        5  *Proc#call
  0.00      0.001     0.001     0.000     0.000        1   String#%
  0.00      0.000     0.000     0.000     0.000       61   String#encode
  0.00     70.068     0.000     0.000    70.068        1   Global#[No method]
  0.00     70.040     0.000     0.000    70.039        2  *<Module::V8::C>#Locker
  0.00     18.289     0.000     0.000    18.289        2   <Module::V8::C>#TryCatch
  0.00      0.000     0.000     0.000     0.000        3   Hash#initialize_copy
  0.00      0.000     0.000     0.000     0.000        2   <Class::V8::Context>#current=
  0.00      0.000     0.000     0.000     0.000        1   ExecJS::RubyRacerRuntime::Context#unbox
  0.00      0.000     0.000     0.000     0.000        1   Hash#reject
  0.00     70.068     0.000     0.000    70.068        1   Uglifier#run_uglifyjs
  0.00     18.290     0.000     0.000    18.289        2   V8::Error::Try#try
  0.00     70.040     0.000     0.000    70.040        1   ExecJS::RubyRacerRuntime::Context#eval
  0.00      0.000     0.000     0.000     0.000        3   Hash#merge
  0.00      0.026     0.000     0.000     0.026        1   JSON#generate
  0.00     51.749     0.000     0.000    51.749        1   V8::Conversion::Identity#to_ruby
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#output_options
  0.00      0.000     0.000     0.000     0.000        1   V8::C::Context#Exit
  0.00     70.038     0.000     0.000    70.038        1   V8::Context#eval
  0.00      0.000     0.000     0.000     0.000        7   Module#===
  0.00      0.000     0.000     0.000     0.000        7   Hash#keys
  0.00      0.000     0.000     0.000     0.000        3   Uglifier#conditional_option
  0.00      0.026     0.000     0.000     0.026        1   JSON#dump
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#comment_options
  0.00      0.000     0.000     0.000     0.000        2   V8::C::TryCatch#HasCaught
  0.00      0.000     0.000     0.000     0.000        4   <Class::Thread>#current
  0.00     51.749     0.000     0.000    51.749        1   V8::Context#to_ruby
  0.00     51.749     0.000     0.000    51.749        1   V8::Conversion::NativeString#to_ruby
  0.00      0.000     0.000     0.000     0.000        4   Kernel#initialize_dup
  0.00      0.000     0.000     0.000     0.000        2   Kernel#respond_to?
  0.00     70.038     0.000     0.000    70.038        1   V8::Context#lock_scope_and_enter
  0.00     51.749     0.000     0.000    51.749        1   V8::Conversion#to_ruby
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#encode_regexp
  0.00      0.000     0.000     0.000     0.000        2   ExecJS::Encoding#encode
  0.00     70.038     0.000     0.000    70.038        1   <Module::V8::C>#HandleScope
  0.00      0.000     0.000     0.000     0.000       15   Symbol#==
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#comment_setting
  0.00     70.038     0.000     0.000    70.038        1   V8::Context#enter
  0.00     70.040     0.000     0.000    70.040        1   ExecJS::RubyRacerRuntime::Context#lock
  0.00      0.000     0.000     0.000     0.000        1   String#==
  0.00      0.000     0.000     0.000     0.000        1   String#force_encoding
  0.00      0.000     0.000     0.000     0.000        2   Thread#[]=
  0.00     70.068     0.000     0.000    70.068        1   Uglifier#compile
  0.00      0.026     0.000     0.000     0.026        1   Uglifier#json_encode
  0.00      0.000     0.000     0.000     0.000        1   V8::C::Context#Enter
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#compressor_options
  0.00      0.000     0.000     0.000     0.000        2   <Class::V8::Context>#current
  0.00      0.000     0.000     0.000     0.000        3   String#to_s
  0.00     51.749     0.000     0.000    51.749        1   V8::Conversion::Fundamental#to_ruby
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#read_source
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#screw_ie8?
  0.00      0.000     0.000     0.000     0.000        1   JSON::Ext::Generator::State#configure
  0.00      0.000     0.000     0.000     0.000        1   V8::Context#entered?
  0.00      0.000     0.000     0.000     0.000        3   Kernel#respond_to_missing?
  0.00      0.000     0.000     0.000     0.000        2   Thread#[]
  0.00      0.000     0.000     0.000     0.000        1   Hash#==
  0.00      0.000     0.000     0.000     0.000        1   Kernel#dup
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#enclose_options
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#mangle_options
  0.00      0.000     0.000     0.000     0.000        1   JSON::Ext::Generator::State#initialize_copy
  0.00      0.000     0.000     0.000     0.000        3   Hash#has_key?
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#source_map_options
  0.00      0.000     0.000     0.000     0.000        3   Hash#key?
  0.00      0.000     0.000     0.000     0.000        3   Symbol#===
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#parse_options
  0.00      0.000     0.000     0.000     0.000        1   Kernel#===
  0.00      0.000     0.000     0.000     0.000        1   Module#<=
  0.00      0.000     0.000     0.000     0.000        2   NilClass#nil?
  0.00      0.000     0.000     0.000     0.000        1   Regexp#source
  0.00      0.000     0.000     0.000     0.000        1   Kernel#class
  0.00      0.000     0.000     0.000     0.000        1   Regexp#casefold?
  0.00      0.000     0.000     0.000     0.000        1   Hash#to_hash
  0.00      0.000     0.000     0.000     0.000        1   Kernel#block_given?
  0.00      0.000     0.000     0.000     0.000        1   Kernel#nil?
  0.00      0.000     0.000     0.000     0.000        1   BasicObject#==

* indicates recursively called methods

john@throttle:~/Projects/test$ be ./test.rb
Uglifier 2.5.3 libv8 3.11.8.17
Thread ID: 18510600
Fiber ID: 19495280
Total: 17.847674
Sort by: self_time

 %self      total      self      wait     child     calls  name
 99.73     17.799    17.799     0.000     0.000        1   V8::C::Script#Run
  0.10      0.019     0.018     0.000     0.000        1   JSON::Ext::Generator::State#generate
  0.10      0.018     0.018     0.000     0.000        1   <Class::V8::C::Script>#New
  0.05      0.009     0.009     0.000     0.000        1   V8::C::String#Utf8Value
  0.01     17.828     0.001     0.000    17.827        1   ExecJS::RubyRacerRuntime::Context#exec
  0.01     17.826     0.001     0.000    17.826        5  *Proc#call
  0.00      0.001     0.001     0.000     0.000        1   String#%
  0.00      0.000     0.000     0.000     0.000       55   Symbol#to_s
  0.00      0.000     0.000     0.000     0.000       61   String#encode
  0.00     17.848     0.000     0.000    17.848        1   Global#[No method]
  0.00     17.816     0.000     0.000    17.816        2   <Module::V8::C>#TryCatch
  0.00      0.000     0.000     0.000     0.000        1   ExecJS::RubyRacerRuntime::Context#unbox
  0.00      0.009     0.000     0.000     0.009        1   V8::Conversion::Identity#to_ruby
  0.00     17.826     0.000     0.000    17.826        2  *<Module::V8::C>#Locker
  0.00      0.000     0.000     0.000     0.000        1   Hash#reject
  0.00     17.848     0.000     0.000    17.848        1   Uglifier#run_uglifyjs
  0.00     17.827     0.000     0.000    17.827        1   ExecJS::RubyRacerRuntime::Context#eval
  0.00      0.000     0.000     0.000     0.000        7   Module#===
  0.00      0.000     0.000     0.000     0.000        3   Hash#initialize_copy
  0.00      0.000     0.000     0.000     0.000        3   Hash#merge
  0.00      0.019     0.000     0.000     0.019        1   JSON#generate
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#output_options
  0.00      0.000     0.000     0.000     0.000        7   Hash#keys
  0.00      0.019     0.000     0.000     0.019        1   JSON#dump
  0.00      0.000     0.000     0.000     0.000        3   Uglifier#conditional_option
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#comment_options
  0.00     17.825     0.000     0.000    17.825        1   V8::Context#eval
  0.00      0.009     0.000     0.000     0.009        1   V8::Conversion::NativeString#to_ruby
  0.00      0.000     0.000     0.000     0.000        2   Kernel#respond_to?
  0.00      0.000     0.000     0.000     0.000        2   V8::C::TryCatch#HasCaught
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#encode_regexp
  0.00      0.000     0.000     0.000     0.000        4   Kernel#initialize_dup
  0.00      0.000     0.000     0.000     0.000        1   V8::C::Context#Exit
  0.00      0.009     0.000     0.000     0.009        1   V8::Context#to_ruby
  0.00      0.000     0.000     0.000     0.000        2   <Class::V8::Context>#current
  0.00     17.817     0.000     0.000    17.816        2   V8::Error::Try#try
  0.00      0.000     0.000     0.000     0.000        2   Thread#[]=
  0.00      0.000     0.000     0.000     0.000       15   Symbol#==
  0.00     17.825     0.000     0.000    17.825        1   V8::Context#lock_scope_and_enter
  0.00      0.009     0.000     0.000     0.009        1   V8::Conversion#to_ruby
  0.00     17.848     0.000     0.000    17.848        1   Uglifier#compile
  0.00      0.000     0.000     0.000     0.000        2   ExecJS::Encoding#encode
  0.00      0.000     0.000     0.000     0.000        2   <Class::V8::Context>#current=
  0.00      0.019     0.000     0.000     0.019        1   Uglifier#json_encode
  0.00      0.000     0.000     0.000     0.000        4   <Class::Thread>#current
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#comment_setting
  0.00     17.826     0.000     0.000    17.826        1   ExecJS::RubyRacerRuntime::Context#lock
  0.00      0.000     0.000     0.000     0.000        3   Kernel#respond_to_missing?
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#compressor_options
  0.00     17.825     0.000     0.000    17.825        1   V8::Context#enter
  0.00      0.000     0.000     0.000     0.000        3   String#to_s
  0.00      0.000     0.000     0.000     0.000        1   String#force_encoding
  0.00     17.825     0.000     0.000    17.825        1   <Module::V8::C>#HandleScope
  0.00      0.000     0.000     0.000     0.000        1   JSON::Ext::Generator::State#configure
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#read_source
  0.00      0.000     0.000     0.000     0.000        1   V8::C::Context#Enter
  0.00      0.009     0.000     0.000     0.009        1   V8::Conversion::Fundamental#to_ruby
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#screw_ie8?
  0.00      0.000     0.000     0.000     0.000        1   String#==
  0.00      0.000     0.000     0.000     0.000        1   Kernel#dup
  0.00      0.000     0.000     0.000     0.000        1   Hash#==
  0.00      0.000     0.000     0.000     0.000        1   V8::Context#entered?
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#enclose_options
  0.00      0.000     0.000     0.000     0.000        1   JSON::Ext::Generator::State#initialize_copy
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#mangle_options
  0.00      0.000     0.000     0.000     0.000        3   Hash#has_key?
  0.00      0.000     0.000     0.000     0.000        1   Kernel#class
  0.00      0.000     0.000     0.000     0.000        2   Thread#[]
  0.00      0.000     0.000     0.000     0.000        3   Symbol#===
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#source_map_options
  0.00      0.000     0.000     0.000     0.000        1   Kernel#===
  0.00      0.000     0.000     0.000     0.000        1   Uglifier#parse_options
  0.00      0.000     0.000     0.000     0.000        1   Regexp#source
  0.00      0.000     0.000     0.000     0.000        3   Hash#key?
  0.00      0.000     0.000     0.000     0.000        1   Module#<=
  0.00      0.000     0.000     0.000     0.000        2   NilClass#nil?
  0.00      0.000     0.000     0.000     0.000        1   Hash#to_hash
  0.00      0.000     0.000     0.000     0.000        1   Regexp#casefold?
  0.00      0.000     0.000     0.000     0.000        1   Kernel#nil?
  0.00      0.000     0.000     0.000     0.000        1   Kernel#block_given?
  0.00      0.000     0.000     0.000     0.000        1   BasicObject#==

* indicates recursively called methods

ignisf · 2014-11-12T20:40:14Z

Very interesting. Thank you, @ledbettj.

ignisf · 2014-11-12T20:48:37Z

Hmm... Looking at https://github.com/cowboyd/therubyracer/blob/master/ext/v8/string.cc#L22 I'm not sure the regression is caused by v8. @cowboyd needs to weigh in here.

ledbettj · 2014-11-12T20:59:37Z

Just to make sure I include all the variables, the version of therubyracer also changed. Sorry for not including this in the initial post.

Faster: Uglifier 2.5.3 / libv8 3.11.8.17 / therubyracer 0.11.4
Slower: Uglifier 2.5.3 / libv8 3.16.14.7 / therubyracer 0.12.1

beekermememe · 2015-02-04T02:57:48Z

+1 on this, We are going from 5 minute deploys up to 18 minutes when deploying to C1 large instances
Rails 3.2.21. We are running a SPA so a good few JS files.

pseidemann · 2015-07-06T19:23:44Z

+1

stefankaiser · 2015-09-09T16:00:48Z

+1

SamSaffron · 2016-05-02T08:08:17Z

I am going to have to debug this, as it stands it takes us 12 seconds to simply run our JS snippets through babel to build application.js, this feels way too long.

I completely eliminated the ruby racer from our assets minification pipeline due to this huge performance hole. Going to work on some simpler repros this week, but this issue does need more attention.

ignisf · 2016-05-02T11:03:51Z

A reproduction would help. Let me know if I can help in some way. Unfortunately profiling c++ is a bit out of my league or I would have taken a crack at this.

bf4 · 2016-05-02T13:40:08Z

@ewoodh2o what's in your test.rb?

ignisf · 2016-05-02T14:24:27Z

I fear that this is a regression in the upstream code itself. If that proves to be the case, we might have to do one of the following things to get around the issue:

dig up what exactly is causing the slowdown and create a patch or backport one from upstream
wait up until TRR gains support for new v8 APIs and roll out a new version
create a new gem that only provides the subset of TRR's functionality necessary for asset generation through execjs to alleviate the dependency on v8's APIs and consequently ease updating to new versions of v8.

In any case, pinpointing the culprit is essential.

SamSaffron · 2016-05-04T07:32:08Z

I started working on this ...

I am making opening moves on building the world's most minimal ruby racer that links against latest v8:

https://github.com/SamSaffron/mini_racer

Performance is significantly better for uglification:

% time ruby uglify_mini_racer.rb app.js 
12.26s user 0.55s system 103% cpu 12.386 total

% time uglifyjs app.js 
4.44s user 0.18s system 103% cpu 4.454 total

% time ruby uglify.rb app.js
144.76s user 0.37s system 100% cpu 2:24.83 total

So there you are, exact same args as https://github.com/lautis/uglifier uses except that instead of taking 144 seconds for our app.js it takes 12.

I am still very uncomfortable with node being this much faster and think a lot of this is due to the wrapping and es5 shims etc that uglifier gem adds will continue to investigate

uglify mini racer code
https://gist.github.com/SamSaffron/2ecc595349d96b572adf2b7c0b69448f

uglify ruby racer code
https://gist.github.com/SamSaffron/336c4bb73c6ea6ca071dba7455af8cbf

our gigantic already minified application js file I was testing with
https://gist.github.com/SamSaffron/148dfd50ecb63296040c4baad2ff34a7

A lot of the performance pain is due to therubyracer running a 2.5 year old version of v8

Note ... this https://gist.github.com/SamSaffron/fe9f938f43a63b2f709b9bd546ba9087 takes 149 seconds (against therubyracer) its EXACTLY the same code that mini racer runs, only diff is that there is V8::Context.new vs MiniRacer::Context.new.

ignisf · 2016-05-04T08:14:04Z

Are you running this on Linux or on OS X? Keep in mind that I specifically disabled link-time optimisations on Linux as the build system was trying to run an ld.gold binary it had pulled and was failing.

SamSaffron · 2016-05-04T08:19:55Z

Running this on Linux x64, perhaps it is those missing optimisations, will try with a "bring my own" libv8 tomorrow

SamSaffron · 2016-05-05T09:15:31Z

no luck finding my missing 20% or so of perf, still @ignisf a new libv8 gem would be mighty welcome even with this performance weirdness

ignisf · 2016-05-05T11:30:28Z

I have a few nits to pick and will push a 5.0 version

SamSaffron · 2016-05-05T23:25:59Z

@ignisf awesome!

Be sure to pick latest stable from https://omahaproxy.appspot.com/ might as well package the latest stable version of v8

ignisf · 2016-05-05T23:53:49Z

k, sure

SamSaffron · 2016-05-08T13:52:09Z

@ignisf I just noticed I had to add:

$LDFLAGS.insert 0, " -stdlib=libstdc++ " if RUBY_PLATFORM =~ /darwin/

To work around a linker issue on mac turns out it links to libc++ by default or something, be sure to do a --pre release prior to doing a stable one so I can confirm mini racer compiles right

ignisf · 2016-05-13T01:08:35Z

@SamSaffron are you by any chance hitting this issue: #202?

SamSaffron · 2016-05-13T01:10:51Z

I was able to install fine on my osx but had to upgrade xcode to latest
first to get latest clang

On Fri, May 13, 2016 at 11:08 AM, Petko Bordjukov [email protected]
wrote:

@SamSaffron https://github.com/SamSaffron are you by any chance hitting
this issue: #202 #202?

—
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub
#150 (comment)

ignisf · 2016-05-15T15:27:47Z

For the record, here are results from running mini_racer's benchmarks/bench_uglify.rb:

Runtime	discourse_app.js	discourse_app_minified.js	discourse_app.js twice on 2 threads
mini_racer (V8 5.0.71.48)	14496.41 ms	18417.71 ms	16932.05 ms
node (nodejs v6.1.0)	18518.89 ms	25905.57 ms	21642.31 ms
therubyracer (V8 3.16.14)	187749.27 ms	203036.86 ms	507423.35 ms

Will keep this issue open for now and hopefully close it after TRR is updated as we'll have to keep around 3.16.14 support at least until then.

coding-horror · 2016-05-15T23:12:48Z

Wow an exponential scale eh? :)

ignisf · 2016-05-15T23:20:55Z

😞

SamSaffron · 2016-05-16T03:53:09Z

This does illustrate a huge issue with exponential scale graphs :)

I would really like to rerun these numbers on Linux x64 comparing GCC vs clang compiled libv8 and mini_racer, I am very curious to see the performance difference

ignisf · 2016-05-16T04:08:08Z

You should be able to do this by locking the version to the source release and configuring a compiler. One thing to note is that at least from the build guide it seems like clang is the preferred compiler for v8.

I don't expect there to be any major differences in performance just solely based on compiler choice, though as the recent versions of gcc and clang don't differ too much in their benchmark scores. What could lead to a performance boost would be enabling link-time optimisations. I have not looked into that yet, though I think there's support for this in v8's build system (for linux).

ignisf · 2016-05-16T04:29:21Z

I am really not sure we should just flick on LTO :D

Alrighty, let's retry this without MAKEFLAGS=-j4 and see how much memory is required for linking...

ignisf · 2016-05-16T05:19:19Z

Peak memory used for linking with lto on one make job is around 3,5512 GB. The produced binary gem archive is 275 MB.

It seems that LTO is enabled automatically when linking the mini_racer_extension.so against a LT optimised version of the libv8 static library (Linux x86_64, GCC 6.1.1).

Could not benchmark because invoking the benchmark script causes a segfault. LTO-enabled branch can be found here: https://github.com/cowboyd/libv8/tree/lto.

I don't see us turning on the lto flag by default for the source distribution (due to the memory footprint), but it could be used for the binaries if anybody can get anything linked to not segfault.

SamSaffron · 2016-05-16T05:48:40Z

phoronix had some pretty mixed results with lto ..
http://www.phoronix.com/scan.php?page=article&item=gcc_471_lto&num=3

I mentioned the clang tests cause another bench I have has node clocking
20% faster so I am curious if this is the compilation of something
internal.

On Mon, May 16, 2016 at 3:19 PM, Petko Bordjukov [email protected]
wrote:

Peak memory used for linking with lto on one make job is around 3,5512 GB.
The produced binary gem archive is 275 MB.

It seems that the LTO is enabled automatically when linking the
mini_racer_extension.so.

Could not benchmark because invoking the benchmark script causes a
segfault. LTO-enabled branch can be found here:
https://github.com/cowboyd/libv8/tree/lto

—
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub
#150 (comment)

SamSaffron · 2016-06-09T22:40:48Z

I think it is fine to close this for now, nothing more for libv8 to do in this regard

ignisf · 2016-06-15T15:20:18Z

I had to add -flto flag for in mini_racer's CXXFLAGS to get it to compile against LTO libv8. Binaries produced with GCC and -flto consistently segfault. My verdict for now is that LTO at this point is more trouble than it's worth.

ignisf added the needs: investigation label Nov 6, 2014

ignisf added type: upstream issue type: bug and removed needs: investigation type: upstream issue labels Nov 12, 2014

ignisf added the needs: investigation label Nov 17, 2014

ignisf mentioned this issue May 3, 2016

libv8 is quite a few versions behind stable #213

Closed

ignisf added the status: blocked label May 15, 2016

ignisf removed the needs: investigation label May 15, 2016

ignisf closed this as completed Jun 10, 2016

ignisf removed the status: blocked label Jun 10, 2016

ignisf mentioned this issue Aug 24, 2016

Runtime Benchmarks rails/execjs#53

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Performance regression in 3.16.14 series? #150

Performance regression in 3.16.14 series? #150

ewoodh2o commented Nov 5, 2014

ledbettj commented Nov 12, 2014

ignisf commented Nov 12, 2014

ignisf commented Nov 12, 2014

ledbettj commented Nov 12, 2014

beekermememe commented Feb 4, 2015

pseidemann commented Jul 6, 2015

stefankaiser commented Sep 9, 2015

SamSaffron commented May 2, 2016

ignisf commented May 2, 2016

bf4 commented May 2, 2016

ignisf commented May 2, 2016

SamSaffron commented May 4, 2016 •

edited

Loading

ignisf commented May 4, 2016

SamSaffron commented May 4, 2016

SamSaffron commented May 5, 2016

ignisf commented May 5, 2016

SamSaffron commented May 5, 2016 •

edited

Loading

ignisf commented May 5, 2016

SamSaffron commented May 8, 2016

ignisf commented May 13, 2016

SamSaffron commented May 13, 2016

ignisf commented May 15, 2016 •

edited

Loading

coding-horror commented May 15, 2016

ignisf commented May 15, 2016

SamSaffron commented May 16, 2016

ignisf commented May 16, 2016 •

edited

Loading

ignisf commented May 16, 2016 •

edited

Loading

ignisf commented May 16, 2016 •

edited

Loading

SamSaffron commented May 16, 2016

SamSaffron commented Jun 9, 2016

ignisf commented Jun 15, 2016 •

edited

Loading

Performance regression in 3.16.14 series? #150

Performance regression in 3.16.14 series? #150

Comments

ewoodh2o commented Nov 5, 2014

ledbettj commented Nov 12, 2014

ignisf commented Nov 12, 2014

ignisf commented Nov 12, 2014

ledbettj commented Nov 12, 2014

beekermememe commented Feb 4, 2015

pseidemann commented Jul 6, 2015

stefankaiser commented Sep 9, 2015

SamSaffron commented May 2, 2016

ignisf commented May 2, 2016

bf4 commented May 2, 2016

ignisf commented May 2, 2016

SamSaffron commented May 4, 2016 • edited Loading

ignisf commented May 4, 2016

SamSaffron commented May 4, 2016

SamSaffron commented May 5, 2016

ignisf commented May 5, 2016

SamSaffron commented May 5, 2016 • edited Loading

ignisf commented May 5, 2016

SamSaffron commented May 8, 2016

ignisf commented May 13, 2016

SamSaffron commented May 13, 2016

ignisf commented May 15, 2016 • edited Loading

coding-horror commented May 15, 2016

ignisf commented May 15, 2016

SamSaffron commented May 16, 2016

ignisf commented May 16, 2016 • edited Loading

ignisf commented May 16, 2016 • edited Loading

ignisf commented May 16, 2016 • edited Loading

SamSaffron commented May 16, 2016

SamSaffron commented Jun 9, 2016

ignisf commented Jun 15, 2016 • edited Loading

SamSaffron commented May 4, 2016 •

edited

Loading

SamSaffron commented May 5, 2016 •

edited

Loading

ignisf commented May 15, 2016 •

edited

Loading

ignisf commented May 16, 2016 •

edited

Loading

ignisf commented May 16, 2016 •

edited

Loading

ignisf commented May 16, 2016 •

edited

Loading

ignisf commented Jun 15, 2016 •

edited

Loading