An Introduction to Truffle and Graal

Copyright © 2015, Oracle and/or its aﬃliates. All rights reserved.
An Introduc@on to Truﬄe and Graal Or how to get a just-‐in-‐1me compiler by wri1ng an interpreter Benoit Daloze Johannes Kepler University 22 May 2017 1

Safe Harbor Statement The following is intended to outline our general product direc@on. 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. The development, release, and @ming of any features or func@onality described for Oracle’s products remains at the sole discre@on of Oracle. 2

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

Agenda GraalVM The Truﬄe Approach Performance Language Interoperability 1 2 3 4 4

GraalVM 5

Graal and GraalVM •  Graal: A new JIT compiler for Java wriXen in Java. •  GraalVM is based on the HotSpotVM. •  Designed for mul@-‐language performance. •  Graal is open source, as are many of the language projects. •  Binary release available from the Oracle Technology Network (OTN) with support for Java, JavaScript, Ruby, and R. 6

Guest Language Application Guest Language Implementation Host Services OS Written by: Application Developer VM Expert OS Expert Language Developer Guest Language Java Java / C++ Unmanaged Language (typically C or C++) Java Written in: Graal VM Truﬄe 7 7

Status of Graal •  The Graal compiler is open source – hXps://github.com/graalvm/graal – Full support of the Java speciﬁca@on •  JEP 243: Java-‐Level JVM Compiler Interface – hXp://openjdk.java.net/jeps/243 – Will be in JDK9 •  Graal JIT compiler can used with any JDK9 VM. 8

The Truﬄe Approach 9

Computer Language Benchmarks Game

Goal: Computer Language Benchmarks Game

“Write Your Own Language” 12 Prototype a new language Parser and language work to build syntax tree (AST), AST Interpreter Write a “real” VM In C/C++, s@ll using AST interpreter, spend a lot of @me implemen@ng run@me system, GC, … People start using it Deﬁne a bytecode format and write bytecode interpreter People complain about performance Write a JIT compiler, improve the garbage collector Performance is s@ll bad Prototype a new language in Java Parser and language work to build syntax tree (AST) Execute using AST interpreter People start using it And it is already fast And it integrates with other languages And it has tool support, e.g., a debugger Current situation How it should be

(a + b) * c •  Abstract syntax tree IS the interpreter •  Every node has an execute method •  Java program that interprets the guest language * + c a b Tradi@onal AST interpreter 13 Object execute(VirtualFrame frame) { Object a = left.execute(frame); Object b = right.execute(frame); return add(a, b); } Object add(Object a, Object b) { if (a instanceof Integer && b instanceof Integer) { return (int)a + (int)b; } else if (a instanceof String && b instanceof String) { return (String)a + (String)b; } else { return genericAdd(a, b); } } 13

Object add(Object a, Object b) { if (a instanceof Integer && b instanceof Integer) { return (int)a + (int)b; } else if (a instanceof String && b instanceof String) { return (String)a + (String)b; } else { return genericAdd(a, b); } } int add(int a, int b) { return a + b; } String add(String a, String b) { return a + b; } 14 14 Object add(Object a, Object b) { return genericAdd(a, b); }

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 15 15

16 Par@al Evalua@on and the 1st Futamura projec@on Par@allyEvaluate( Ruby interpreter, known Ruby program/method) => compiler graph => machine code Very similar to a Just-‐in-‐Time Compiler wriXen for Ruby speciﬁcally! •  Par@al evalua@on inlines the AST execute() methods and Graal op@mizes and translates the resul@ng graph to machine code •  Par@al evalua@on brings context to the compila@on: The AST of a method is a set of nodes instances which are compiled together

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 17 17

Performance 18

Performance Disclaimers •  All Truﬄe numbers reﬂect a development snapshot –  Subject to change at any @me (hopefully improve) –  You have to know a benchmark to understand why it is slow or fast •  We are not claiming to have complete language implementa@ons –  JavaScript: passes 100% of ECMAscript standard tests •  Working on full compa@bility with V8 for Node.JS –  Ruby: passing 100% of RubySpec language tests •  Passing around 94% of the core library tests –  R: prototype, but already complete enough and fast for a few selected workloads •  Benchmarks that are not shown –  may not run at all, or –  may not run fast 19

Performance: JavaScript 20 0 0.2 0.4 0.6 0.8 1 1.2 1.4 box2d Deltablue Crypto EarleyBoyer Gameboy NavierStokes Richards Raytrace Splay Geomean Speedup, higher is beBer Performance rela1ve to V8 JavaScript performance: similar to V8

Performance: R with Scalar Code 21 0 10 20 30 40 50 60 70 80 90 100 Speedup, higher is beBer Performance rela1ve to GNU R with bytecode interpreter 660x Huge speedups on scalar code, GNU R is only optimized for vector operations

Performance: Ruby Compute-‐Intensive Kernels 22 0 50 100 150 200 250 300 350 Speedup, higher is beBer Performance rela1ve to JRuby running with Java HotSpot server compiler Huge speedup because Truffle can optimize through Ruby metaprogramming

Speedup rela@ve to MRI

• • • • 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 hXps://github.com/smarr/are-‐we-‐fast-‐yet TruﬄeRuby

25 Optcarrot: NES emulator benchmark

Optcarrot: NES emulator benchmark – 9x faster than MRI 26

27 Mozart-‐Graal: {Pow X 7}

28 Mozart-‐Graal: {Map Xs fun {$ X} X + 3 end}

Language interoperability PuMng the pieces together 29

29 You can execute any language on the JVM / CLR -‐ as long as it looks like Java / C#. 30

31 Let’s write a small part of a program in a diﬀerent language! Ø There is no interface to the language I can use Ø I don’t want to write a lot of wrapper code Ø Performance will be bad #include<stdio.h> struct complex { double r; double i; } int main() { struct complex *a = …; struct complex *b = …; add(a, b); } function add(a, b) { var result = {r:0, i:0}; result.r = a.r + b.r; result.i = a.i + b.i; return result; } struct complex *a struct complex *b C JavaScript 31

High-‐Performance Language Interoperability Dynamic Compilation = a . obj obj is a C struct = a obj !!" ! JS Msg = a is C? -> Message Resolution Msg !! ! C JS-specific object access Language-independent object access C-specific object access var a = obj.value; value value Read obj value Map JS access to a message Map message to a C access JS JS JS JS C C JS JS Machine Code 32 32 Dynamic Compilation = a . obj obj is a C struct = a obj !!" ! JS Msg = a is C? -> Message Resolution Msg !! ! C JS-specific object access Language-independent object access C-specific object access var a = obj.value; value value Read obj value Map JS access to a message Map message to a C access JS JS JS JS C C JS JS Machine Code

High-‐Performance Language Interoperability obj is a Ruby object Msg Rb !!" ! C-specific object access Map message to Rb access Dynamic Compilation = a is C? -> obj value = a is C? -> obj value -> obj value is Rb? JS JS JS JS C C JS Rb JS Rb JS Machine Code C-specific and Rb-specific object accesses var a = obj.value; 33 33

34 Demo Time !

Conclusion 35

There would be much more to talk about! Language interoperability Paralleliza@on Database Driver Op@miza@ons Debuggability, Analyzability 36

Resources •  Graal on GitHub – hXps://github.com/graalvm/graal – Mailing list: graal-‐[email protected] •  Technology preview release of GraalVM – hXp://www.oracle.com/technetwork/oracle-‐labs/program-‐languages/ 37

An Introduction to Truffle and Graal

An Introduction to Truffle and Graal

Transcript

More Decks by Benoit Daloze

Other Decks in Research

Featured