Faster Ruby and JS with Graal/Truffle

Transcription

1

2 The Ruby Logo is Copyright (c) 2006, Yukihiro Matsumoto. It is licensed under the terms of the Creative Commons Attribution-ShareAlike 2.5 agreement. Faster Ruby and JS with Graal/Truffle Chris Seaton Oracle 11 April 2016

3 Safe Harbor Statement The following is intended to provide some insight into a line of research in Oracle Labs. It is intended for information purposes only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied upon in making purchasing decisions. Oracle reserves the right to alter its development plans and practices at any time, and the development, release, and timing of any features or functionality described in connection with any Oracle product or service remains at the sole discretion of Oracle. Any views expressed in this presentation are my own and do not necessarily reflect the views of Oracle.

4

5

6 One virtual machine to rule them all

7

8

9 Computer Language Benchmarks Game 3

10 Computer Language Benchmarks Game 3 Goal:

11 Current situation How it should be Prototype a new language Parser and language work to build syntax tree (AST), AST Interpreter Write a real VM In C/C++, still using AST interpreter, spend a lot of time implementing runtime system, GC, People start using it 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 People complain about performance Define a bytecode format and write bytecode interpreter Performance is still bad Write a JIT compiler Improve the garbage collector

12 Java internals basics

13 Java source code javac Java class files (bytecode) java

14 Java source code javac Java class files (bytecode) java x86 machine code (assembly) CPU

15

16 Truffle

17 Guest Language Bytecode JVM

18 Guest Language Java IR, machine code cache, invalidation and deoptimisation, optimisation phases, replacements, etc etc Graal VM

19 Guest Language Truffle AST interpreter Graal VM

20 x + y * z

21 x + y * z + x * y z

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

23 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

24 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

25 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

26 U Node Rewriting for Profiling Feedback G Compilation using Partial Evaluation G U U Node Transitions I G I G U U U Uninitialized Integer I I I I I AST Interpreter Uninitialized Nodes S String G D Double Generic AST Interpreter Rewritten Nodes Compiled Code 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.

27 Deoptimization to AST Interpreter G Node Rewriting to Update Profiling Feedback G Recompilation using Partial Evaluation G G I G I G D G D G I I I D I I I D 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.

28 Graal

29 Hotspot

30 Hotspot

31 Hotspot JIT

32 Hotspot JIT

33 JIT Hotspot JIT

34 JVM Garbage Collector JIT Compiler Classloader Everything else

35 JVM Garbage Collector Server Compiler Client Compiler Classloader Everything else

36 JVM Garbage Collector Server Compiler Graal Client Compiler Classloader Everything else

37 Bytecode Graal byte[] Metadata

38 C. Wimmer, Graal Tutorial, 2015.

39 JS R Ruby Truffle Java JVMCI (JVM Compiler Interface) Graal Hotspot C++

40 JS R Ruby Truffle via Maven etc Graal Hotspot Java 9

41 otn graal

42

43 Ruby

44 The JRuby logo is copyright (c) Tony Price 2011, licensed under the terms of Creative Commons Attribution-NoDerivs 3.0 Unported (CC BY-ND 3.0) JRuby JITs by emitting JVM bytecode VM in Java Core library mostly in Java

45 The Rubinius logo is copyright 2011 Shane Becker, licensed under the terms of Creative Commons Attribution-NoDerivatives 4.0 International CC BY-ND 4.0 Rubinius JITs by emitting LLVM code VM in C++ Core library mostly in Ruby

46 + Truffle and Graal

47 100% language Compatibility with the (spec/ruby)

48 90% library Compatibility with the core (spec/ruby)

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

50

51 Why is it apparently so hard to make Ruby fast?

52 How do people want to write Ruby?

53

54

55

56

57

58

59

60 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

61 direct: i/s send-symbol: i/s x slower send-string: i/s x slower eval: i/s x slower

62 Throwing away metaprogramming Crystal throws away metaprogramming entirely to make a faster Ruby Rubinius doesn t support set_trace_func JRuby doesn t support set_trace_func or ObjectSpace RubyMotion doesn t support eval, Binding A real shame, and not necessary

63 How does Truffle + Graal solve this?

64 Escape analysis and partial evaluation

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

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

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

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

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

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

71 t1 = true? 2 : 8 puts t1

72 t1 = 2 puts t1

73 t1 = 2 puts 2

74 puts 2

75 t0 = a <=> b t1 = t0 < 0? a : b puts t1

76 t0 = a <=> b t1 = (a <=> b) < 0? a : b puts t1

77 t1 = (a <=> b) < 0? a : b puts (a <=> b) < 0? a : b

78 puts (a <=> b) < 0? a : b

79 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

80 direct: i/s send-symbol: i/s x slower send-string: i/s x slower eval: i/s x slower

81 direct: i/s send-symbol: i/s difference within err send-string: i/s - difference within err eval: i/s - difference within err

82 An extreme example

83

84

85

86

87

88

89 = 22!

90

91

92

93 Node

94 Control flow Data flow Node

95

96

97

98 Polyglot

99

100

101

102 Conclusions We don t need to tell Ruby, JS, R etc programmers to avoid language features to get performance

103 Conclusions We don t need to tell Ruby, JS, R etc programmers to avoid language features to get performance We can make a better JIT compiler by writing it in Java

104 Where to get more info

105 otn graal

106

107

108

109 Acknowledgements Benoit Daloze Kevin Menard Petr Chalupa Oracle Labs Danilo Ansaloni Stefan Anzinger Daniele Bonetta Matthias Brantner Laurent Daynès Gilles Duboscq Michael Haupt Christian Humer Mick Jordan Peter Kessler Hyunjin Lee David Leibs Tom Rodriguez Roland Schatz Chris Seaton Doug Simon Lukas Stadler Oracle Labs (continued) Michael Van de Vanter Adam Welc Till Westmann Christian Wimmer Christian Wirth Paul Wögerer Mario Wolczko Andreas Wöß Thomas Würthinger Oracle Labs Interns Shams Imam Stephen Kell Gero Leinemann Julian Lettner Gregor Richards Robert Seilbeck Rifat Shariyar Oracle Labs Alumni Erik Eckstein Christos Kotselidi JKU Linz Prof. Hanspeter Mössenböck Josef Eisl Thomas Feichtinger Matthias Grimmer Christian Häub Josef Haider Christian Hube David Leopoltsederr Manuel Rigger Stefan Rumzucker Bernhard Urban University of California, Irvine Prof. Michael Franz Codrut Stancu Gulfem Savrun Yeniceri Wei Zhang University of Edinburgh Christophe Dubach Juan José Fumero Alfonso Ranjeet Singh Toomas Remmelg T. U. Dortmund Prof. Peter Marwedel Helena Kotthaus Ingo Korb LaBRI Floréal Morandat Purdue University Prof. Jan Vitek Tomas Kalibera Romand Tsegelskyi Prahlad Joshi Petr Maj Lei Zhao University of California, Davis Prof. Duncan Temple Lang Nicholas Ulle

110 @ChrisGSeaton (or just search for jruby truffle )

111 Safe Harbor Statement The preceding is intended to provide some insight into a line of research in Oracle Labs. It is intended for information purposes only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied upon in making purchasing decisions. Oracle reserves the right to alter its development plans and practices at any time, and the development, release, and timing of any features or functionality described in connection with any Oracle product or service remains at the sole discretion of Oracle. Any views expressed in this presentation are my own and do not necessarily reflect the views of Oracle.

112

113