Accelerating Ruby with LLVM

Accelerating Ruby
with LLVM
Evan Phoenix
Oct 2, 2009

Tuesday, October 6, 2009

RUBY


RUBY
Strongly, dynamically typed


RUBY
Uniﬁed Model


RUBY
Everything is an object


RUBY
3.class # => Fixnum


RUBY
Every code context is equal


RUBY
Every context is a method


RUBY
Garbage Collected


RUBY
A lot of syntax


RUBY
Strongly, dynamically typed Every code context is equal
Uniﬁed model Every context is a method
Everything is an object Garbage collected
3.class A lot of syntax


Rubinius


Rubinius
Started in 2006


Rubinius
Build a ruby environment for fun


Rubinius
Unlike most “scripting” languages,
write as much in ruby as possible


Rubinius
Core functionality of perl/python/ruby in C,
NOT in their respective language.


Rubinius
C => ruby => C => ruby


Rubinius
Language boundaries suck


Rubinius
Started in 2006
Built for fun
Turtles all the way down


Evolution


Evolution
100% ruby prototype running on 1.8


Evolution
Hand translated VM to C


Evolution
Rewrote VM in C++


Evolution
Switch away from stackless


Evolution
Experimented with handwritten
assembler for x86


Evolution
Switch to LLVM for JIT


Evolution
100% ruby prototype Switch away from stackless
Hand translated VM to C Experiment with assembler
Rewrote VM in C++ Switch to LLVM for JIT


Features


Features
Bytecode VM


Features
Simple interface to native code


Features
Accurate, generational garbage collector


Features
Integrated FFI API


Features
Bytecode VM Generational GC
Interface to native code Integrated FFI


Benchmarks


def foo()
ary = []
100.times { |i| ary << i }
end

300,000 times

Seconds
9

8.02

6.75

5.90
5.30
4.5

3.60

2.25 2.59

0
1.8 1.9 rbx rbx jit rbx jit +blocks


def foo()
hsh = {}
100.times { |i| hsh[i] = 0 }
end

100,000 times

Seconds
30

25.36

22.5

15

12.54
12.01

7.5

4.85 5.26

0


def foo()
hsh = { 47 => true }
100.times { |i| hsh[i] }
end

100,000 times

Seconds
7

6.26

5.25

3.5 3.64

2.68 2.66

2.09
1.75

0


Early LLVM Usage


Early LLVM Usage
Compiled all methods up front


Early LLVM Usage
Simple opcode-to-function translation
with inlining


Early LLVM Usage
Startup went from 0.3s to 80s


Early LLVM Usage
Compiled all methods upfront
Simple opcode-to-function translation
Startup from 0.3s to 80s


True JIT


True JIT
JIT Goals


True JIT
JIT Goals
Choose methods that beneﬁt the most


True JIT
JIT Goals
Compiling has minimum impact on performance


True JIT
JIT Goals
Ability to make intelligent frontend decisions


Choosing Methods


Choosing Methods
Simple call counters


Choosing Methods
When counter trips, the fun starts


Choosing Methods
Room for improvement


Choosing Methods
Increment counters in loops


Choosing Methods
Weigh different invocations differently


Choosing Methods
Simple counters Room for improvement
Trip the counters, do it Increment in loops
Weigh invocations


Which Method?


Which Method?
Leaf methods trip quickly


Which Methods?
Consider the whole callstack


Which Methods?
Pick a parent expecting inlining


Which Method?
Leaf methods trip
Consider the callstack
Find a parent


Minimal Impact


Minimal Impact
After the counters trip


Minimal Impact
Queue the method


Minimal Impact
Background thread drains queue


Minimal Impact
Frontend, passes, codegen in background


Minimal Impact
Install JIT’d function


Minimal Impact
Install JIT’d function
Requires GC interaction


Minimal Impact
Trip the counters Compile in background
Queue the method Install function pointer


Good Decisions


Good Decisions
Naive translation yields ﬁxed improvement


Good Decisions
Performance shifts to method dispatch


Good Decisions
Improve optimization horizon


Good Decisions
Inline using type feedback


Good Decisions
Naive translation sucks Performance in dispatch
Inline using type feedback Improve optimizations


Type Feedback


Type Feedback
Frontend translates to IR


Type Feedback
Read InlineCache information


Type Feedback
InlineCaches contain proﬁling info


Type Feedback
Use proﬁling to drive inlining!


Type Feedback
Frontend generates IR InlineCaches have proﬁling
Reads InlineCaches Use proﬁling to drive inlining!


Inlining


Inlining
Proﬁling info shows a dominant class


2
1%

1 class
98%


Inlining
Lookup method in compiler


Inlining
For native functions, emit direct call


Inlining
For FFI, inline conversions and call


Inlining
Find dominant class
Emit direct calls if possible
Lookup method


Inlining Ruby


Inlining Ruby
Policy decides on inlining


Inlining Ruby
Drive sub-frontend at call site


Inlining Ruby
All inlining occurs in the frontend


Inlining Ruby
Generated IR preserves runtime data


Inlining Ruby
Generated IR preserves runtime data
GC roots, backtraces, etc


Inlining Ruby
No AST between bytecode and IR


Inlining Ruby
No AST between bytecode and IR
Fast, but limits the ability to generate better IR


Inlining Ruby
Policy decides Preserve runtime data
Drive sub-frontend Generates fast, ugly IR


LLVM


LLVM
IR uses operand stack


LLVM
Highlevel data ﬂow not in SSA


LLVM
Passes eliminate redundencies


LLVM
Makes GC stack marking easy


LLVM
nocapture improves propagation


LLVM
Exceptions via sentinal value


LLVM
Nested handlers use branches for control


LLVM
Inlining exposes redundant checks


LLVM
Inline guards


LLVM
Inline guards
Simple type guards


if(obj->class->class_id ==
<integer constant>) {


LLVM
Inline guards
Custom AA pass for guard elimination


LLVM
Inline guards
Teach pointsToConstantMemory about...


LLVM
Maximizing constant propagation


LLVM
Type failures shouldn’t contribute values


if(obj->class->class_id == 0x33) {
val = 0x7;
} else {
val = send_msg(state, obj, ...);
}


if(obj->class->class_id == 0x33) {
val = 0x7;
} else {
return uncommon(state);
}


LLVM
Makes JIT similar to tracing


LLVM
Use overﬂow intrinsics


LLVM
Use overﬂow intrinsics
Custom pass to fold constants arguments


LLVM
AA knowledge for tagged pointers


LLVM
AA knowledge of tagged pointers
0x5 is 2 as a tagged pointer


LLVM
Not in SSA form Maximize constants
Simplistic exceptions Use overﬂow
Inlining guards Tagged pointer AA


Issues


Issues
How to link with LLVM?


Issues
How to link with LLVM?
An important SCM issue


Issues
Ugly, confusing IR from frontend


Issues
instcombine confuses basicaa


Issues
Operand stack confuses AA


Issues
Inability to communicate semantics


Object* new_object(state)


Returned pointer aliases nothing

Only modiﬁes state

If return value is unused, remove the call

Semi-pure?


Issues
Ugly IR AA confusion
Linking with LLVM Highlevel semantics


Thanks!

http://rubini.us
ephoenix@engineyard.com


Accelerating Ruby with LLVM

Recommended

More Related Content

Similar to Accelerating Ruby with LLVM (12)

More from evanphx (12)

Recently uploaded (20)

Accelerating Ruby with LLVM