Optimizing Ruby with TruffleRuby

Copyright © 2016, Oracle and/or its aﬃliates. All rights reserved. |
Op?mizing Ruby with
TruﬄeRuby
The Ruby Logo is Copyright (c) 2006, Yukihiro Matsumoto. It is licensed under the terms of the Crea?ve Commons APribu?on-‐ShareAlike 2.5 agreement.
Slides from Chris Seaton
Oracle Labs @ChrisGSeaton

Benoit Daloze
@eregontp

15 March 2017
Linz Ruby User Group

View Slide

Safe Harbor Statement
The following is intended to provide some insight into a line of research in Oracle Labs. It
is intended for informa?on purposes only, and may not be incorporated into any
contract. It is not a commitment to deliver any material, code, or func?onality, and
should not be relied upon in making purchasing decisions. Oracle reserves the right to
alter its development plans and prac?ces at any ?me, and the development, release, and
?ming of any features or func?onality described in connec?on with any Oracle product or
service remains at the sole discre?on of Oracle. Any views expressed in this presenta?on
are my own and do not necessarily reﬂect the views of Oracle.
2

View Slide

Who am I ?
•  Benoit Daloze
•  PhD student at Johannes Kepler University
•  Research with TruﬄeRuby on concurrency
•  MRI and JRuby commiPer
•  Maintainer of the Ruby Spec Suite (ruby/spec)
3
TwiPer: @eregontp
GitHub: @eregon

View Slide

Agenda
•  Introduc?on
•  Truﬄe and Ruby implementa?ons
•  Why is Ruby apparently so hard to run fast?
•  Par?al Evalua?on
•  Performance & Compa?bility
4

View Slide

Computer Language Benchmarks Game

View Slide

Goal:
Computer Language Benchmarks Game

View Slide

What is Truﬄe?

View Slide

How do Ruby interpreters work?

View Slide

x + y * z

View Slide

x + y * z
+
x *
y z

View Slide

x + y * z
+
x *
y z
load_local x
load_local y
load_local z
call :*
call :+

View Slide

Just-‐In-‐Time Compilers

View Slide

x + y * z
+
x *
y z
load_local x
load_local y
load_local z
call :*
call :+

View Slide

x + y * z
+
x *
y z
load_local x
load_local y
load_local z
call :*
call :+
pushq %rbp
movq %rsp, %rbp
movq %rdi, -‐8(%rbp)
movq %rsi, -‐16(%rbp)
movq %rdx, -‐24(%rbp)
movq -‐16(%rbp), %rax
movl %eax, %edx
movq -‐24(%rbp), %rax
imull %edx, %eax
movq -‐8(%rbp), %rdx
addl %edx, %eax
popq %rbp
ret

View Slide

The JRuby logo is copyright (c) Tony Price 2011, licensed under the terms of Crea?ve Commons APribu?on-‐NoDerivs 3.0 Unported (CC BY-‐ND 3.0)
JRuby
JITs by emihng JVM bytecode
VM in Java
Core library mostly in Java

View Slide

The Rubinius logo is copyright 2011 Shane Becker, licensed under the terms of Crea?ve Commons APribu?on-‐NoDeriva?ves 4.0 Interna?onal — CC BY-‐ND 4.0
Rubinius
JITs by emihng LLVM bitcode
VM in C++
Core library mostly in Ruby

View Slide

JVM
18

View Slide

JVM
19

View Slide

JVM
JIT
20

View Slide

JVM
JIT
21

View Slide

JVM
Graal
22

View Slide

JVM
Graal
Truﬄe
23

View Slide

Truﬄe
Graal VM
JVM
24

View Slide

U
U U
U
U I
I I
G
G I
I I
G
G
Node Rewriting
for Profiling Feedback
AST Interpreter
Rewritten Nodes
AST Interpreter
Uninitialized Nodes
Compilation using
Partial Evaluation
Compiled Code
Node Transitions
S
U
I
D
G
Uninitialized Integer
Generic
Double
String
T. Würthinger, C. Wimmer, A. Wöß, L. Stadler, G. Duboscq, C. Humer, G. Richards, D. Simon,
and M. Wolczko. One VM to rule them all. In Proceedings of Onward!, 2013.
Compiling the AST to machine code

View Slide

T. Würthinger, C. Wimmer, A. Wöß, L. Stadler, G. Duboscq, C. Humer, G. Richards, D. Simon,
and M. Wolczko. One VM to rule them all. In Proceedings of Onward!, 2013.
I
I I
G
G I
I I
G
G
Deoptimization
to AST Interpreter
D
I D
G
G D
I D
G
G
Node Rewriting to Update
Profiling Feedback
Recompilation using
Partial Evaluation
Deop?miza?on and recompila?on

View Slide

Copyright © 2016, Oracle and/or its aﬃliates. All rights reserved. | 27

View Slide


View Slide

Why is Ruby apparently so hard
to run fast?

View Slide


View Slide

require 'benchmark/ips'

Benchmark.ips do |x|
x.report( "direct", "14 + 2" )
x.report( "send-‐symbol", "14.send(:+, 2)" )
x.report( "send-‐string", "14.send('+', 2)" )
x.report( "eval", "eval('14 + 2')" )
x.compare!
end

View Slide

direct: 37299872.6 i/s
send-‐symbol: 13060179.1 i/s -‐ 2.86x slower
send-‐string: 4974575.3 i/s -‐ 7.50x slower
eval: 171835.9 i/s -‐ 217.07x slower

View Slide

Throwing away metaprogramming
•  Crystal – throws away metaprogramming en?rely to make a faster Ruby
•  Rubinius – doesn’t support set_trace_func
•  JRuby – doesn’t support set_trace_func or ObjectSpace
•  RubyMo?on – doesn’t support eval, Binding

View Slide

How does Truﬄe + Graal solve this?

View Slide

ParCal EvaluaCon
And DeopCmizaCon

View Slide

def min(a, b)
[a, b].sort[0]
end
puts min(2, 8)

View Slide

def min(a, b)
[a, b].sort[0]
end
puts [2, 8].sort[0]

View Slide

t0 = 2 <=> 8
t1 = t0 < 0 ? 2 : 8
t2 = t0 > 0 ? 8 : 2
t3 = [t1, t2]
puts t3[0]

View Slide

t0 = 2 <=> 8
t1 = t0 < 0 ? 2 : 8
t2 = t0 > 0 ? 8 : 2
t3 = [t1, t2]
puts t1

View Slide

t0 = -1
t1 = t0 < 0 ? 2 : 8
puts t1

View Slide

t0 = -1
t1 = -1 < 0 ? 2 : 8
puts t1

View Slide

t1 = true ? 2 : 8
puts t1

View Slide

t1 = 2
puts t1

View Slide

t1 = 2
puts 2

View Slide

require 'benchmark/ips'

Benchmark.ips do |x|
x.report( "direct", "14 + 2" )
x.report( "send-‐symbol", "14.send(:+, 2)" )
x.report( "send-‐string", "14.send('+', 2)" )
x.report( "eval", "eval('14 + 2')" )
x.compare!
end

View Slide

direct: 37299872.6 i/s
send-‐symbol: 13060179.1 i/s -‐ 2.86x slower
send-‐string: 4974575.3 i/s -‐ 7.50x slower
eval: 171835.9 i/s -‐ 217.07x slower

View Slide

direct: 73099792.5 i/s
send-‐symbol: 73458837.7 i/s – difference within err
send-‐string: 66882023.8 i/s -‐ difference within err
eval: 67024838.3 i/s -‐ difference within err

View Slide

An extreme example

View Slide


View Slide

= 22 !

View Slide


View Slide

Performance

View Slide

Speedup rela?ve to MRI

View Slide


View Slide

●
●
●
●
MRI 2.3
JRuby 9.0.4
Node.js
JRuby+Truffle
Java 1.8.0u66
1
5
10
25
50
75
12 Benchmarks using Objects, Closures and Arrays
from hPps://github.com/smarr/are-‐we-‐fast-‐yet
TruﬄeRuby

View Slide

Optcarrot: NES emulator benchmark

View Slide

Optcarrot: NES emulator benchmark – 9x faster than MRI
72

View Slide

Demo Time !

View Slide

CompaCbility

View Slide

Language specs
96% Core library specs
100%
92% C-‐extension API specs
Ruby specs
75

View Slide

But does it run
“Rails”, “Ruby on Rails”, and the Rails logo are registered trademarks of David Heinemeier Hansson
?

View Slide


View Slide

Rails tests
Ac?ve Model
Ac?ve Support
Ac?ve Record
Ac?on View
Ac?on Pack
Ac?on Mailer
Rail?es
Sprockets-‐Rails
Ac?ve Job
Spring
100%
100%
98%
Basic func0onality works
37%
78

View Slide

So then why can’t we run real applica?ons yet?
•  C extensions are s?ll a work in progress
– Almost no database drivers
– No openssl
– No nokogiri
– Prevents us running almost everything unfortunately
•  The specs don’t have perfect coverage
79

View Slide

Ongoing Research
•  Running C extension with Sulong (Truﬄe+LLVM)
•  ObjectSpace: stop the world and walk the frames
•  Represen?ng String as Ropes
•  Eﬃcient parallelism:
•  Thread-‐Safe Objects for Dynamic Languages
•  Thread-‐Safe and Scalable Ruby collec?ons

View Slide

Conclusions
Simple VM ideas can solve the major problems of
making Ruby fast

View Slide

Conclusions
Simple VM ideas can solve the major problems of
making Ruby fast

We don’t need to tell Ruby programmers to avoid
language features to get performance

View Slide

Where to get more info

View Slide


hPps://github.com/graalvm/truﬄeruby

(or just search for “truﬄeruby”)

View Slide

Search for ’graal otn’
www.oracle.com/technetwork/oracle-‐labs/program-‐languages

View Slide

Acknowledgements
Oracle
Danilo Ansaloni
Stefan Anzinger
Cosmin Basca
Daniele BonePa
MaPhias Brantner
Petr Chalupa
Jürgen Christ
Laurent Daynès
Gilles Duboscq
Mar?n Entlicher
Brandon Fish
Bas?an Hossbach
Chris?an Humer
Mick Jordan
Vojin Jovanovic
Peter Kessler
David Leopoldseder
Duncan MacGregor
Kevin Menard
Jakub Podlešák
Aleksandar Prokopec
Tom Rodriguez
Oracle (con0nued)
Roland Schatz
Chris Seaton
Doug Simon
Štěpán Šindelář
Zbyněk Šlajchrt
Lukas Stadler
Codrut Stancu
Jan Štola
Jaroslav Tulach
Michael Van De Vanter
Adam Welc
Chris?an Wimmer
Chris?an Wirth
Paul Wögerer
Mario Wolczko
Andreas Wöß
Thomas Würthinger
JKU Linz
Prof. Hanspeter Mössenböck
Benoit Daloze
Josef Eisl
Thomas Feich?nger
MaPhias Grimmer
Chris?an Häubl
Josef Haider
Chris?an Huber
Stefan Marr
Manuel Rigger
Stefan Rumzucker
Bernhard Urban

University of Edinburgh
Christophe Dubach
Juan José Fumero Alfonso
Ranjeet Singh
Toomas Remmelg

LaBRI
Floréal Morandat
University of California, Irvine
Prof. Michael Franz
Gulfem Savrun Yeniceri
Wei Zhang

Purdue University
Prof. Jan Vitek
Tomas Kalibera
Petr Maj
Lei Zhao

T. U. Dortmund
Prof. Peter Marwedel
Helena KoPhaus
Ingo Korb

University of California, Davis
Prof. Duncan Temple Lang
Nicholas Ulle

University of Lugano, Switzerland
Prof. Walter Binder
Sun Haiyang
Yudi Zheng
Oracle Interns
Brian Belleville
Miguel Garcia
Shams Imam
Alexey Karyakin
Stephen Kell
Andreas Kun…
Volker Lan?ng
Gero Leinemann
Julian LePner
Joe Nash
David Piorkowski
Gregor Richards
Robert Seilbeck
Rifat Shariyar

Alumni
Erik Eckstein
Michael Haupt
Christos Kotselidis
Hyunjin Lee
David Leibs
Chris Thalinger
Till Westmann

View Slide

Optimizing Ruby with TruffleRuby

Optimizing Ruby with TruffleRuby

Benoit Daloze

More Decks by Benoit Daloze

Other Decks in Programming

Featured

Transcript