SSA Form for the Java HotSpot™ Client Compiler

Sun Microsystems, Inc.

Institute for System Software

Johannes Kepler University Linz, Austria

Department of Computer Science

University of California, Irvine

Java HotSpot™ Client Compiler

SSA Form for the

Christian Wimmer [email protected]

www.christianwimmer.at

April 2009

2

Java is also an island…

Java Source Code

Java Compiler

Java Bytecodes

Just-in-Time Compiler

Garbage Collector

Java Class Library

Java Language

Specification

Java Virtual Machine

Specification

Java Virtual Machine

Java Specification

Requests

Java Enterprise Edition

Java Servlets

Java Server Faces

Enterprise JavaBeans

Java Persistence API

Class Loader

Interpreter

Java Micro Edition

Today’s Talk

3

Java HotSpot™ VM

Execution Engine

executes compiles generates

Garbage Collector collects

access memory

loads Class Loader

Interpreter JIT Compiler

Stop-and-Copy

Mark-and-Compact

Heap

Bytecodes Machine Code

Eden

Old Space collects

Deoptimization

Young Generation

Old Generation

From To

4

Client Compiler – Design

Compilation Speed Code Quality

Completeness

All Optimizations

Data Flow Analysis

Profile Data Simple Structure

Common Code Patterns

Fast Optimizations

JSR and Locking

Corner Cases

Unstructured Bytecodes

Design Decisions

for a JIT Compiler

Client Compiler Server Compiler

Interpreter

5

Bytecode

Parsing

Client Compiler – Structure

Bytecodes

HIR

Optimized HIR

LIR

Machine Code

Method Inlining

Constant Propagation

Local Value Numbering

Register Allocation

Global

Optimizations

LIR Generation

Code Generation

Back End

Bounds Check Elimination

HIR (SSA Form)

Global Value Numbering

Register Allocation

Escape Analysis

Null Check Elimination

Conditional Expr. Elim.

[Thomas Würthinger]

[Thomas Kotzmann]

Phi Function Placement

6

!""#$%&'(!"

)""#*%+,-()"

.""#/(#*01$2")3

4""#$%&'(!"

5""#$%&'(!"

3""#*%+,-()"6""#,78"

9""#+:%;2"!"#$%

)."#07$"

)<"#,-%=2()"

)>"?%-%")9

)3"#*%+,-()"

)6"#,-%=2()"

)9"#$%&'()"

.!"#=2-7=+"

!"#"$%&$'"&!"#$@$'"&+A"B"

""$'""1C"

""$(&@+"D")A"B"""""1"E"+"F"!"#$@+"G")AC"

""H")*!)"B"

""""1"E")C"

""H"

""+)",+'"1C"H"

#!" G"

#!" G" $-&

#!" G" $-&#!"

#!" G" $.&#!"#!"

#!" G" $/&#!"

#!" G" $0&#!"

#!" G" $1&

#!" $1&#!" #5"

#!" #6"

#!"#)!"

#!" G"

Phi Function Placement

7

!""#$%&'(!"

)""#*%+,-()"

.""#/(#*01$2")3

4""#$%&'(!"

5""#$%&'(!"

3""#*%+,-()"6""#,78"

9""#+:%;2"!"#$%

)."#07$"

)<"#,-%=2()"

)>"?%-%")9

)3"#*%+,-()"

)6"#,-%=2()"

)9"#$%&'()"

.!"#=2-7=+"

#!""&"'"()$)'%*%

#)"")"

".""#!"IE"#)"J"K."L"K)"

#)!"1M#N#5O"#6P"

"))"#=2-7=+"#)!"

#6"")"

"9""?%-%"K<"

#<"")"#>""#!"G"#<"

#4""#+:%;2"!"#$@#>A"

#5""#!"F"#4"

"3""?%-%"K<"

K!"

K)"

K."

K<"

Bytecodes Local Variables

and Operand Stack

HIR Instructions

Phi Function Placement – Loops

8

!""#*%+,-()"

)""#,-%=2()"

.""#*%+,-()"<""#,-%=2(."

>""#$%&'(."

4""#$%&'(!"

5""#/(#*01?-")9

9""#$%&'()"

)!"#$%&'(."

))"#07$")."#,-%=2()"

)<"##+*".")"

)5"?%-%">"

)9"#$%&'()"

.!"#=2-7=+"

!"#"$%&$'"&!"#$@$'""+A"B"

""$'"&1"E")C"

""(2+&@$'"&#"E")C"#"IE"+C"#QQA"B"""""1"E"1"F"#C"

""H"

""+)",+'&1C"

H"

#<" #3" #9"

#!" #)" #)"

#<" #>" #4"

#<" #>" #4"


9

!""#*%+,-()"

)""#,-%=2()"

.""#*%+,-()"<""#,-%=2(."

>""#$%&'(."

4""#$%&'(!"

5""#/(#*01?-")9

9""#$%&'()"

)!"#$%&'(."

))"#07$")."#,-%=2()"

)<"##+*".")"

)5"?%-%">"

)9"#$%&'()"

.!"#=2-7=+"

#!""&"'"()$)'%*%

#)"")"

".""?%-%"K)"

"))"#=2-7=+"#>"

#3""#>"F"#4"

#6"")"

#9""#4"Q"#6"")!"?%-%"K)"

#<""1M#N#!O"GP"

#>""1M#N#)O"GP"

#4""1M#N#)O"GP""5""#4"D"#<"J"K<"L"K."

K!"

K."

K<"

Bytecodes Local Variables HIR Instructions

#<""1M#N#!O"#<P"

#>""1M#N#)O"#3P"

#4""1M#N#)O"#9P""5""#4"D"#<"J"K<"L"K."

K)"

$-& #3" #9"

#!" #)" #)"

$-& #>" #4"

$-& #>" #4"


10

!""#*%+,-()"

)""#,-%=2()"

.""#*%+,-()"<""#,-%=2(."

>""#$%&'(."

4""#$%&'(!"

5""#/(#*01?-")9

9""#$%&'()"

)!"#$%&'(."

))"#07$")."#,-%=2()"

)<"##+*".")"

)5"?%-%">"

)9"#$%&'()"

.!"#=2-7=+"

#!""&"'"()$)'%*%

#)"")"

".""?%-%"K)"

"))"#=2-7=+"#>"

#3""#>"F"#4"

#6"")"

#9""#4"Q"#6"")!"?%-%"K)"

K!"

K."

K<"


Simplify Phi Functions

Eliminate #<"Replace with #!

#>""1M#N#)O"#3P"

#4""1M#N#)O"#9P""5""#4"D"$-"J"K<"L"K."

K)"

#!" #3" #9"

#!" #)" #)"

#!" #>" #4"

#!" #>" #4"


11

!""#*%+,-()"

)""#,-%=2()"

.""#*%+,-()"<""#,-%=2(."

>""#$%&'(."

4""#$%&'(!"

5""#/(#*01?-")9

9""#$%&'()"

)!"#$%&'(."

))"#07$")."#,-%=2()"

)<"##+*".")"

)5"?%-%">"

)9"#$%&'()"

.!"#=2-7=+"

#!""&"'"()$)'%*%

#)"")"

".""?%-%"K)"

"))"#=2-7=+"#>"

#3""#>"F"#4"

#6"")"

#9""#4"Q"#6"")!"?%-%"K)"

K!"

K."

K<"


#>""1M#N#)O"#3P"

#4""1M#N#)O"#9P""5""#4"D"#!"J"K<"L"K."

K)"

Create Phi Functions

Only for Variables

Modified in a Loop


12

SSA Construction

Detect Block Boundaries

Mark Loop Headers

field not hot

Mark Variables Modified in Loops

Reverse Postorder of Blocks

Fill Blocks with HIR Instructions

Create Phi Function

Variable Equal in

all Predecessors?

yes no

yes no no

Global Optimizations

…


Setup Phi Functions

Abstract Interpretation of Block

Save Variable State

Variable Modified

in any Loop?

yes

Phi Function Necessary for a Variable?

Loop Header?

yes

no Variable Alive?

No Phi Function Necessary

Phi Function Statistics

13

One Run of SPECjvm98 (All Benchmarks)

~1000 Methods Compiled

~ 210 KByte Code Compiled

Without Loop Optimization With Loop Optimization

Created Simplified Created Simplified

Loop Phi Functions 2843 2046 72% 1224 429 35%

Other Phi Functions 1930 376 20% 1685 108 6%

Total 4773 2422 50% 2909 537 18%

~ 122,000 HIR Instructions

SSA Deconstruction

14

#!""&"'"()$)'%*%

#)"")"

"))"#=2-7=+"#>"

#3""#>"F"#4"

#9""#4"Q"#)"

K!"

K."

K<"

#>""1M#N#)O"#3P"

#4""1M#N#)O"#9P"

"5""#4"D"#!"J"K<"L"K."

K)"

0%:2"2*RGDS)"

0%:2")GDS."

0%:2")GDS<"

0%:2"S.GD2&R"

=2-7=+"2&R"

0%:2"S.GDS>"

07$""S>OS<GDS>"

0%:2"S<GDS4"&''""S4O)GDS4"

0%:2"S>GDS."

0%:2"S4GDS<"

T701"K)"

*01""S<OS)"

T701"UVOK<"

LIR Operation HIR Instructions

R1

R2

R3

R4

R5

Intervals for Linear Scan Register Allocation

Moves in Predecessors

No Coalescing

Register Hints for

Linear Scan Register Allocator

0%:2")GD2&R"

0%:2")GD28R"

=2-7=+"2&R"

07$""2&RO28RGD2&R"

&''""28RO)GD28R"

T701"K)"

*01""28RO2*R"

T701"UVOK<"

eax

ecx

eax

ebx

ebx

LIR Register for Phi Function

Delete Unnecessary Moves

K!"

K."

K<"

K)"

Exception Handling

15

#!"")"

#)""?2-/#2$'"WWW"

#.""#&$%&'"WWW"#<""."

#>""*&$$"WWW"

#4""'#:"WWW"

WWW"

#)!"1M#N#!O#!O#<O#<P"

WWW"

Normal Control flow

All Critical Edges Split

Save Position to Insert Moves

Exception Edges

Start “In the Middle” of a Block

Many Phi Function Operands

Split Critical Exception Edges?

Would Lead to Many Blocks

Most of them Finally Empty

Overhead to Remove Them

Phi Functions in LIR

Known by Register Allocator

Create Adapter Blocks on Demand

Array Bounds Check Elimination

16

&!""&"'"()$)'%"%

#)""&"'"()$)'%*%

#.""!"

"<""?%-%"K)"

")!"=2-7=+"

"5""&!N#>P"E"#."

#3"")"

#6""#>"Q"#3""9""?%-%"K)"

K."

K<"

HIR Instructions

#>""1M#N#.O"#6P"

"4""#>"I"#)"J"K."L"K<"

K)"

!"#"$%&32$4"#+)"'@$'"NP"&O"$'""+A"B"

""(2+"@$'""#"E"!C"#"I"+C"#QQA"B"

""""&N#P"E"!C"""H"

H"

Java Source Code

Bounds for #> at instruction 5"

Lower Bound: #."

Increasing (because of #6)

Upper Bound: #)"

&!""&"'"()$)'%"%

#)""&"'"()$)'%*%

#.""!""))"*M2*;"#)"IE"&!W$2+?-M"

"<""?%-%"K)"

All Bounds Loop Invariant

K!"

Insert Check Before Loop

Remove Check Inside Loop

Java Exception Semantics

Integer Overflows

Escape Analysis

17

#!""&"'"()$)'%*%

#)"")"

".""#!"IE"#)"J"K."L"K)"

&9""+2X",*$)-)'%

")!"&9W6"E"#)"

"))"?%-%"K<"

&<""+2X",*$)-)'%

#>""#!"Y"#)"

#4""#+:%;2"!"#$@#>A"#5""#!"F"#4"

"3""&<W6"E"#5"

"6""?%-%"K<"

HIR Instructions

&)."1M#N&<O"&9P"

#)>"&).W6"")4"#=2-7=+"#)>"

K!"

K)"

K."

K<"

!"#"$%&$'"&!"#$@$'"&+A"B"

""Z+-2?2="1C"

""$(&@+"D")A"B"""""1"E"')5"Z+-2?2=@+"F"!"#$@+"G")AAC"

""H")*!)"B"

""""1"E"')5"Z+-2?2=@)AC"

""H"

""+)",+'"1W#+-[&$72@AC"H"

Java Source Code

#!" G"

#!" &<"

#!" &9"

Escape Analysis

18

#!""&"'"()$)'%*%

#)"")"

".""#!"IE"#)"J"K."L"K)"

&9""+2X",*$)-)'%

")!"&9W6"E"#)"

"))"?%-%"K<"

&<""+2X",*$)-)'%

#>""#!"Y"#)"

#4""#+:%;2"!"#$@#>A"#5""#!"F"#4"

"3""&<W6"E"#5"

"6""?%-%"K<"

HIR Instructions

#!"&)."

&9W6L#)"

&<W6L#5"

&).W6L#)<"&)."1M#N&<O"&9P"

#)<"1M#N#5O"#)P"

#)>"&).W6"")4"#=2-7=+"#)>"

"))"?%-%"K<"

#>""#!"Y"#)"

#4""#+:%;2"!"#$@#>A"#5""#!"F"#4"

"6""?%-%"K<"

#)<"1M#N#5O"#)P"

")4"#=2-7=+"#)<"

K!"

K)"

K."

K<"

Track Field Values

State Similar to Variables

Create Phi Functions

Track Escape State

Method Local

Thread Local

Escaping

Optimize Non-Escaping Objects

Eliminate Object Allocation

Eliminate Field Stores

Replace Field Loads

Alias Effects

Equi-Escape Sets

19

Performance Comparison

Speedup of JDK 6 Client and Server relative to JDK 5 Client

JDK 5 Client

JDK 6 Client

JDK 6 Server

Intel Pentium D processor 830 with 3.0 GHz, 2 GByte

main memory, Microsoft Windows XP Professional

Client and Server configuration use same garbage collector

First run

Last run

Selected Publications

!! Client Compiler, SSA Form, Linear Scan Register Allocation "! Thomas Kotzmann, Christian Wimmer, Hanspeter Mössenböck, Thomas Rodriguez, Kenneth Russell,

David Cox: Design of the Java HotSpot™ Client Compiler for Java 6. In ACM Transactions on Architecture and Code Optimization, volume 5, issue 1, article 7. ACM Press, 2008. doi:10.1145/1369396.1370017

"! Christian Wimmer, Hanspeter Mössenböck: Optimized Interval Splitting in a Linear Scan Register Allocator. In Proceedings of the ACM/USENIX International Conference on Virtual Execution Environments, pages 132-141. ACM Press, 2005. doi:10.1145/1064979.1064998

!! Array Bounds Check Elimination "! Thomas Würthinger, Christian Wimmer, Hanspeter Mössenböck: Array Bounds Check Elimination

in the Context of Deoptimization. In Science of Computer Programming, volume 74, issues 5-6, pages 279-295. Elsevier, 2009. doi:10.1016/j.scico.2009.01.002

!! Escape Analysis "! Thomas Kotzmann, Hanspeter Mössenböck: Escape Analysis in the Context of Dynamic

Compilation and Deoptimization. In Proceedings of the ACM/USENIX International Conference on Virtual Execution Environments, pages 111-120. ACM Press, 2005. doi:10.1145/1064979.1064996

"! Thomas Kotzmann, Hanspeter Mössenböck: Run-Time Support for Optimizations Based on Escape Analysis. In Proceedings of the International Symposium on Code Generation and Optimization, pages 49-60. IEEE Computer Society, 2007. doi:10.1109/CGO.2007.34

!! Complete List of Publications: "! http://wikis.sun.com/display/HotSpotInternals/Publications+JKU

20

SSA Form for the Java HotSpot™ Client Compiler

Documents

Transcript of SSA Form for the Java HotSpot™ Client Compiler