<Insert Picture Here> Maxine: A JVM Written in Java

Transcription

1

2 <Insert Picture Here> Maxine: A JVM Written in Java Michael Haupt Oracle Labs Potsdam, Germany

3 The following is intended to outline our general product direction. 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. The development, release, and timing of any features or functionality described for Oracle s products remains at the sole discretion of Oracle. 3

4 Speaker's Background Technische Universität Darmstadt, Doctoral research, Steamloom (virtual machine support for aspect-oriented programming) 4

5 Speaker's Background Hasso-Plattner-Institut Potsdam, Postdoc, virtual machine architecture disentangling, SOM family, NXTalk Technische Universität Darmstadt, Doctoral research, Steamloom (virtual machine support for aspect-oriented programming) 4

6 Speaker's Background Hasso-Plattner-Institut Potsdam, Postdoc, virtual machine architecture disentangling, SOM family, NXTalk Technische Universität Darmstadt, Doctoral research, Steamloom (virtual machine support for aspect-oriented programming) Oracle Labs, Potsdam, since 2011 Maxine team, virtual machine architecture, multi-language virtual machines, code cache management, JSR292 4

7 Who Needs a JVM? JVM 5

8 Who Needs a JVM? application developers, end users Java Applications JVM 5

9 Who Needs a JVM? application developers, end users researchers Java Applications Dynamic Compiler Garbage Collector Object Layout Runtime Optimisations Concurrent Language Specifics Thread Migration Target Architecture Parallel Tagging Concurrency Abstractions JVM 5

10 Who Needs a JVM? application developers, end users researchers programming language implementers Java, Ruby, Python, Java Smalltalk,... Applications Dynamic Compiler Garbage Collector Object Layout Runtime Optimisations Concurrent Language Specifics Thread Migration Target Architecture Parallel Tagging Concurrency Abstractions JVM 5

11 Who Needs a JVM? application developers, end users researchers programming language implementers Java, Ruby, Python, Java Smalltalk,... Applications Dynamic Compiler Garbage Collector Object Layout Runtime!!!! Optimisations Concurrent Language Specifics Thread Migration Target Architecture Parallel Tagging Concurrency Abstractions JVM 5

12 Who Needs a JVM? application developers, end users researchers programming language implementers Java, Ruby, Python, Java Smalltalk,... Applications Dynamic Compiler Garbage Collector Object Layout Runtime!!!! Optimisations Concurrent Language Specifics Thread Migration Target Architecture Parallel Tagging Concurrency Abstractions J?VM JVM 5

13 Maxine almost 100 % Java use unmodified JDK meta-circularity studies Home page: Source (GPLv2): Mailing list via: 6

14 Maxine almost 100 % Java use unmodified JDK meta-circularity studies supported on 64-bit x86 Solaris, Mac OS X, Linux Virtual Edition: Xen Home page: Source (GPLv2): Mailing list via: 6

15 Maxine almost 100 % Java use unmodified JDK meta-circularity studies supported on 64-bit x86 Solaris, Mac OS X, Linux Virtual Edition: Xen full IDE support (Eclipse, NetBeans) tooling: Inspector Home page: Source (GPLv2): Mailing list via: 6

16 Maxine almost 100 % Java use unmodified JDK meta-circularity studies supported on 64-bit x86 Solaris, Mac OS X, Linux Virtual Edition: Xen full IDE support (Eclipse, NetBeans) tooling: Inspector dynamic compilation only semi-space GC modular architecture Home page: Source (GPLv2): Mailing list via: 6

17 VM Research Group at Oracle Labs Doug Simon Thomas Würthinger Michael Haupt Christian Wimmer Laurent Daynès Mario Wolczko Director Mick Jordan Michael Van De Vanter Home page: Source (GPLv2): Mailing list via: 7

18 Why Java? language reflection annotations JDK rich and powerful API inclusion in VM VM architecture uniform representation same compiler chain ecosystem IDE support tooling 8

19 Bootstrapping Maxine Maxine Source Code 9

20 Bootstrapping Maxine Maxine Source Code javac Maxine Bytecodes Just-in-Time Compilers Memory Manager... others... 9

21 Bootstrapping Maxine Maxine Source Code javac Maxine Bytecodes Boot Image Generator Host VM (HotSpot) 9

22 Bootstrapping Maxine Maxine Source Code javac Maxine Bytecodes Just-in-Time Compilers Memory Manager... others... Boot Image Generator Host VM (HotSpot) 9

23 Bootstrapping Maxine Maxine Source Code javac Maxine Bytecodes Just-in-Time Compilers Memory Manager... others... Boot Image Generator Host VM (HotSpot) 9

24 Bootstrapping Maxine Maxine Source Code javac Maxine Bytecodes Just-in-Time Compilers Memory Manager... others... Boot Image Generator Host VM (HotSpot) 9

25 Bootstrapping Maxine Maxine Source Code javac Maxine Bytecodes Just-in-Time Compilers Memory Manager Maxine Boot Image... others... Boot Image Generator Host VM (HotSpot) 9

26 Bootstrapping Maxine Maxine Source Code javac Maxine Bytecodes Just-in-Time Compilers Memory Manager Maxine Boot Image Boot Loader... others... Boot Image Generator Host VM (HotSpot) 9

27 Maxine Scheme Abstractions Dynamic Compilers Garbage Collector Object Layout Runtime Maxine VM 10

28 Maxine Scheme Abstractions interface LayoutScheme Dynamic Compilers Garbage Collector Object Layout Runtime Maxine VM 10

29 Maxine Scheme Abstractions interface LayoutScheme interface ReferenceScheme Dynamic Compilers Garbage Collector Object Layout Runtime Maxine VM 10

30 Maxine Scheme Abstractions interface HeapScheme interface LayoutScheme interface ReferenceScheme Dynamic Compilers Garbage Collector Object Layout Runtime Maxine VM 10

31 Maxine Scheme Abstractions interface HeapScheme interface LayoutScheme interface ReferenceScheme Dynamic Compilers Garbage Collector Object Layout Runtime Maxine VM interface MonitorScheme 10

32 Maxine Scheme Abstractions interface HeapScheme interface LayoutScheme interface ReferenceScheme Dynamic Compilers Garbage Collector Object Layout Runtime Maxine VM interface MonitorScheme interface RunScheme 10

33 Maxine Scheme Abstractions interface HeapScheme interface LayoutScheme interface ReferenceScheme Dynamic Compilers Garbage Collector Object Layout Runtime Maxine VM interface CompilationBroker interface MonitorScheme interface RunScheme 10

34 Example: ReferenceScheme What you use in Maxine to access a field: int x = Reference.fromJava(obj).readInt(24); field offset (in bytes) is computed by LayoutScheme; we assume a constant offset in this example 11

35 Example: ReferenceScheme What you use in Maxine to access a field: int x = Reference.fromJava(obj).readInt(24); field offset (in bytes) is computed by LayoutScheme; we assume a constant offset in this example Actual call chain after forced inlining Reference ref = MaxineVM.vm.config.referenceScheme.toReference(obj); Pointer ptr = MaxineVM.vm.config.referenceScheme.toOrigin(ref); int x = ptr.readint(24); 11

36 Example: ReferenceScheme What you use in Maxine to access a field: int x = Reference.fromJava(obj).readInt(24); field offset (in bytes) is computed by LayoutScheme; we assume a constant offset in this example Actual call chain after forced inlining Reference ref = MaxineVM.vm.config.referenceScheme.toReference(obj); Pointer ptr = MaxineVM.vm.config.referenceScheme.toOrigin(ref); int x = ptr.readint(24); All fields are so the compiler knows the concrete scheme implementation used: class DirectReferenceScheme implements ReferenceScheme { native Reference toreference(object native Word toorigin(reference ref); 11

37 Example: ReferenceScheme What you use in Maxine to access a field: int x = Reference.fromJava(obj).readInt(24); field offset (in bytes) is computed by LayoutScheme; we assume a constant offset in this example Actual call chain after forced inlining Reference ref = MaxineVM.vm.config.referenceScheme.toReference(obj); Pointer ptr = MaxineVM.vm.config.referenceScheme.toOrigin(ref); int x = ptr.readint(24); All fields are so the compiler knows the concrete scheme implementation used: class DirectReferenceScheme implements ReferenceScheme { native Reference toreference(object native Word toorigin(reference ref); Final machine code: movl [rax + 24], rbx 11

38 Baseline Compiler: T1X job: assemble pre-generated bytecode templates written in Java compiled by optimising compiler at VM build time very small amount of glue assembly 12

39 Baseline Compiler: T1X job: assemble pre-generated bytecode templates written in Java compiled by optimising compiler at VM build time very small amount of glue assembly template code for integer array load static int Object int index) { ArrayAccess.checkIndex(array, index); } return result = ArrayAccess.getInt(array, static void checkindex(object array, int index) { int length = Layout.readArrayLength(Reference.fromJava(array)); if (UnsignedMath.aboveOrEqual(index, length)) { throw Throw.arrayIndexOutOfBoundsException(array, index); } static int getint(object array, int index) { return Layout.getInt(Reference.fromJava(array), index); } 12

40 Baseline Compiler: T1X instantiated template code for integer array load mov rdi, [rsp + 16] mov mov esi, [rsp] rax, rdi movsxd rax, [rax + 16] cmp esi, eax jb call nop mov call nop L1 Throw.indexOutOfBoundsException() rdi, rax MaxRuntimeCalls.runtimeUnwindException L1: movsxd rsi, esi movsxd rax, rdi[rsi * ] addq mov rsp, 0x10 [rsp], eax prologue (loading of operand stack slots) compiled template 8 bytes hub (class pointer) 8 bytes misc (locking, GC) 4 bytes array length 4 bytes alignment epilogue (storing of operand stack slots) 13

41 Inspector 14

42 Optimising Compilers: C1X and Graal C++ C1 / C2 Inline Assembly Compiler Interface Garbage Collector Object Locking Type Information Interpreter Runtime HotSpot VM 15

43 Optimising Compilers: C1X and Graal Java C++ C1 / C2 C1X Inline Assembly Compiler Interface XIR Compiler Interface Garbage Collector Type Information Garbage Collector Type Information Object Locking Interpreter Object Locking T1X Runtime HotSpot VM Runtime Maxine VM 15

44 Optimising Compilers: C1X and Graal Java C++ C1 / C2 C1X Inline Assembly Compiler Interface XIR Compiler Interface Garbage Collector Type Information Garbage Collector Type Information Object Locking Interpreter Object Locking T1X Runtime HotSpot VM Runtime Maxine VM 15

45 Optimising Compilers: C1X and Graal Java C++ C1 / C2 C1X Inline Assembly Compiler Interface XIR Compiler Interface Garbage Collector Type Information Garbage Collector Type Information Object Locking Interpreter Object Locking T1X Runtime HotSpot VM Runtime Maxine VM Graal Snippets Compiler Interface Garbage Collector Object Locking Type Information Interpreter Runtime Graal VM 15

46 Optimising Compilers: C1X and Graal Java C++ C1 / C2 C1X Inline Assembly Compiler Interface XIR Compiler Interface Garbage Collector Type Information Garbage Collector Type Information Object Locking Interpreter Object Locking T1X Runtime HotSpot VM Runtime Maxine VM Graal Snippets Compiler Interface Garbage Collector Object Locking Type Information Interpreter Runtime Graal VM 15

47 Optimising Compilers: C1X and Graal Java C++ C1 / C2 C1X Inline Assembly Compiler Interface XIR Compiler Interface Garbage Collector Type Information Garbage Collector Type Information Object Locking Interpreter Object Locking T1X Runtime HotSpot VM Runtime Maxine VM Graal Graal Snippets Compiler Interface Snippets Compiler Interface Garbage Collector Type Information Garbage Collector Type Information Object Locking Interpreter Object Locking T1X Runtime Graal VM Runtime Maxine VM with Graal 15

48 Optimising Compilers: C1X and Graal Java C++ C1 / C2 C1X Inline Assembly Compiler Interface XIR Compiler Interface Garbage Collector Type Information Garbage Collector Type Information Object Locking Interpreter Object Locking T1X Runtime HotSpot VM Runtime Maxine VM Graal Graal Snippets Compiler Interface Snippets Compiler Interface Garbage Collector Type Information Garbage Collector Type Information Object Locking Interpreter Object Locking T1X Runtime Graal VM Runtime Maxine VM with Graal 15

49 Graal Vision Statement Create an extensible, modular, dynamic, and aggressive compiler using object-oriented and reflective Java programming, a graph-based and visualizable intermediate representation, and Java snippets. -- Thomas Würthinger Collaboration with Johannes-Kepler-Universität Linz, Austria JKU Lukas Stadler, Gilles Duboscq, Christian Häubl, Prof. Hanspeter Mössenböck Home page: Source (GPLv2): Mailing list: graal-dev@openjdk.java.net 16

50 Graal Java Bytecodes Java Bytecode Parser Method Inlining Type Inference Escape Analysis Loop Optimisations VM Lowering Memory Optimisations Architecture Lowering Register Allocation Peephole Optimisations Code Generation Machine Code 17

51 Graal Java Bytecodes Java Bytecode Parser Method Inlining Type Inference Escape Analysis Loop Optimisations VM Lowering Memory Optimisations Architecture Lowering Register Allocation Peephole Optimisations Code Generation Machine Code 17

52 Graal Java Bytecodes Language dependent Language independent VM independent Architecture independent Java Bytecode Parser Method Inlining Type Inference Escape Analysis Loop Optimisations VM Lowering VM dependent Architecture independent Memory Optimisations Architecture Lowering VM dependent Architecture dependent Register Allocation Peephole Optimisations Code Generation Machine Code 17

53 Performance: Maxine/C1X 200% 191% 180% 160% 151% 157% 140% 138% 120% 119% 100% 80% 60% 71% 64% 58% 81% 54% 76% 74% 40% 20% 0% SPECjvm2008 DaCapo peak DaCapo startup SPECjbb2005 SciMark 2.0 JavaGrande Client Maxine/C1X Server 2-socket dual core AMD Opteron 2214, 2.2 GHz, 4 total cores 4 GByte main memory Oracle Enterprise Linux, version e15 18

54 Performance: GraalVM 120% 100% 80% 60% 72% 77% 48% 61% 66% 74% 40% 20% 0% DaCapo Scala-DaCapo SPECjvm2008 Client GraalVM Server Intel(R) Core(TM) 2.67GHz (4 cores) 8 GByte main memory Ubuntu Linux 11.10, kernel

55 Maxine: Code Cache Management T1X C1X Maxine VM 20

56 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() T1X Cache Maxine VM C1X Cache 20

57 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache 20

58 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache 20

59 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache to-space from-space T1X Cache with Semi-Space Memory Management Scheme 20

60 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache <clinit> m() n() to-space from-space T1X Cache with Semi-Space Memory Management Scheme 20

61 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache <clinit> m() n() o() to-space from-space T1X Cache with Semi-Space Memory Management Scheme 20

62 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache <clinit> m() n() to-space from-space T1X Cache with Semi-Space Memory Management Scheme 20

63 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache <clinit> m() n() from-space to-space from-space to-space T1X Cache with Semi-Space Memory Management Scheme 20

64 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache <clinit> m() n() from-space to-space from-space to-space T1X Cache with Semi-Space Memory Management Scheme 20

65 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache <clinit> n() m() from-space to-space from-space to-space T1X Cache with Semi-Space Memory Management Scheme 20

66 Maxine: Code Cache Management T1X C1X m() n() <clinit> o() o() T1X Cache Maxine VM m() C1X Cache <clinit> n() m() o() from-space to-space from-space to-space T1X Cache with Semi-Space Memory Management Scheme 20

67 Code Cache Management: Which Methods are Live? baseline (T1X) method optimised (C1X) method live method o() f() g() p() h() g() t1 call stack f() g() h() i() j() k() l() to-space from-space T1X Cache 21

68 Code Cache Management: Which Methods are Live? baseline (T1X) method optimised (C1X) method live method o() f() g() p() h() g() t1 call stack f() g() h() i() j() k() l() to-space from-space T1X Cache 21

69 Code Cache Management: Which Methods are Live? baseline (T1X) method optimised (C1X) method live method k() o() f() g() p() h() g() t1 call stack f() g() h() i() j() k() l() to-space from-space T1X Cache 21

70 Code Cache Management: Which Methods are Live? baseline (T1X) method optimised (C1X) method live method k() o() f() g() j() p() h() g() t1 call stack f() g() h() i() j() k() l() to-space from-space T1X Cache 21

71 Code Cache Management: Which Methods are Live? baseline (T1X) method optimised (C1X) method live method k() o() f() g() j() p() h() g() t1 call stack f() g() h() i() j() k() l() to-space from-space T1X Cache 21

72 Code Cache Management: Which Methods are Live? baseline (T1X) method optimised (C1X) method live method k() j() o() f() g() p() h() g() t1 call stack live methods methods on stack direct callees of on-stack methods others to protect from eviction freshly compiled, not yet installed invocation counter within optimisation threshold existing type profile f() g() h() i() j() k() l() to-space from-space T1X Cache 21

73 Code Cache Management: Eviction baseline (T1X) method optimised (C1X) method live method k() o() f() g() j() p() h() g() t1 call stack f() g() h() i() j() k() l() from-space to-space T1X Cache 22

74 Code Cache Management: Eviction baseline (T1X) method optimised (C1X) method live method k() j() o() f() g() p() h() g() t1 call stack course of action 1. invalidate unmarked methods (and vtable entries, direct calls) f() g() h() i() j() k() l() from-space to-space T1X Cache 22

75 Code Cache Management: Eviction baseline (T1X) method optimised (C1X) method live method k() j() o() f() g() p() h() g() t1 call stack course of action 1. invalidate unmarked methods (and vtable entries, direct calls) f() g() h() j() k() from-space to-space T1X Cache 22

76 Code Cache Management: Eviction baseline (T1X) method optimised (C1X) method live method k() j() o() f() g() p() h() g() t1 call stack course of action 1. invalidate unmarked methods (and vtable entries, direct calls) 2. move marked methods (update vtable entries, direct calls) f() g() h() j() k() from-space to-space T1X Cache 22

77 Code Cache Management: Eviction baseline (T1X) method optimised (C1X) method live method k() j() o() f() g() p() h() g() t1 call stack course of action 1. invalidate unmarked methods (and vtable entries, direct calls) 2. move marked methods (update vtable entries, direct calls) f() g() h() j() k() from-space T1X Cache to-space 22

78 Code Cache Management: Eviction baseline (T1X) method optimised (C1X) method live method k() j() o() f() g() p() h() g() t1 call stack course of action 1. invalidate unmarked methods (and vtable entries, direct calls) 2. move marked methods (update vtable entries, direct calls) 3. patch return addresses f() g() h() j() k() from-space T1X Cache to-space 22

79 Code Cache Management: Eviction baseline (T1X) method optimised (C1X) method live method k() j() o() f() g() p() h() g() t1 call stack course of action 1. invalidate unmarked methods (and vtable entries, direct calls) 2. move marked methods (update vtable entries, direct calls) 3. patch return addresses f() g() h() j() k() from-space T1X Cache to-space 22

80 Where Do Pointers to Machine Code Occur, and How Are They Handled? vtables/itables directly accessible via meta-level API (actors, hubs); invalidate (replace with trampoline calls) or relocate 23

81 Where Do Pointers to Machine Code Occur, and How Are They Handled? vtables/itables direct calls directly accessible via meta-level API (actors, hubs); invalidate (replace with trampoline calls) or relocate directly accessible via target method API (safepoints); invalidate (replace with trampoline calls) or relocate 23

82 Where Do Pointers to Machine Code Occur, and How Are They Handled? vtables/itables direct calls return addresses directly accessible via meta-level API (actors, hubs); invalidate (replace with trampoline calls) or relocate directly accessible via target method API (safepoints); invalidate (replace with trampoline calls) or relocate accessible during stack walking; relocate accordingly 23

83 Where Do Pointers to Machine Code Occur, and How Are They Handled? vtables/itables direct calls return addresses local variables member variables directly accessible via meta-level API (actors, hubs); invalidate (replace with trampoline calls) or relocate directly accessible via target method API (safepoints); invalidate (replace with trampoline calls) or relocate accessible during stack walking; relocate accordingly These might have to be relocated how to determine which Address is a code pointer? 23

84 Where Do Pointers to Machine Code Occur, and How Are They Handled? vtables/itables direct calls return addresses local variables member variables directly accessible via meta-level API (actors, hubs); invalidate (replace with trampoline calls) or relocate directly accessible via target method API (safepoints); invalidate (replace with trampoline calls) or relocate accessible during stack walking; relocate accordingly These might have to be relocated how to determine which Address is a code pointer? 23

85 Tagged Code Pointers Object CodePointer Word Address Offset Pointer Size 24

86 Tagged Code Pointers Object CodePointer Word 63 bits payload (offset from base code address) 1 tag bit Address Offset Pointer Size 24

87 Tagged Code Pointers class CodePointer public static CodePointer from(long value) { if (ishosted()) { return new CodePointer(tag(value)); } return UnsafeCast.asCodePointer(tag(value)); }... } CodePointer Object Word 63 bits payload (offset from base code address) 1 tag bit Address Offset Pointer Size 24

88 Tagged Code Pointers class CodePointer public static CodePointer from(long value) { if (ishosted()) { return new CodePointer(tag(value)); } return UnsafeCast.asCodePointer(tag(value)); }... } CodePointer Object Word 63 bits payload (offset from base code address) 1 tag bit Address public static CodePointer ascodepointer(long value) { return CodePointer.from(value); } Pointer Size 24

89 Stack Reference Maps and Tagged Code Pointers static void m(int i, MyClass o) { int j = 2 * i; o.f(i); } 0: iconst_2 1: iload_0 2: imul 3: istore_2 4: aload_1 5: iload_2 6: invokevirtual <f:(i)v> 9: return 25

90 Stack Reference Maps and Tagged Code Pointers static void m(int i, MyClass o) { int j = 2 * i; o.f(i); } 0: iconst_2 1: iload_0 2: imul 3: istore_2 4: aload_1 5: iload_2 6: invokevirtual <f:(i)v> 9: return idealised stack frame layout j o i operands locals arguments 0 <r> return address 25

91 Stack Reference Maps and Tagged Code Pointers static void m(int i, MyClass o) { int j = 2 * i; o.f(i); } 0: iconst_2 1: iload_0 2: imul 3: istore_2 4: aload_1 5: iload_2 6: invokevirtual <f:(i)v> 9: return idealised stack frame layout j operands locals stack reference map (at instruction 6) 2 1 o i arguments <r> return address 25

92 Stack Reference Maps and Tagged Code Pointers static void m(int i, MyClass o) { int j = 2 * i; o.f(i); } 0: iconst_2 1: iload_0 2: imul 3: istore_2 4: aload_1 5: iload_2 6: invokevirtual <f:(i)v> 9: return void x(...) {... MyClass o =...; CodePointer cp =...;... } idealised stack frame layout j operands locals stack reference map (at instruction 6) 2 1 o i arguments <r> return address 25

93 Stack Reference Maps and Tagged Code Pointers static void m(int i, MyClass o) { int j = 2 * i; o.f(i); } 0: iconst_2 1: iload_0 2: imul 3: istore_2 4: aload_1 5: iload_2 6: invokevirtual <f:(i)v> 9: return void x(...) {... MyClass o =...; CodePointer cp =...;... } idealised stack frame layout j operands locals stack reference map (at instruction 6) 2 1 o i arguments <r> return address 25

94 Stack Reference Maps and Tagged Code Pointers static void m(int i, MyClass o) { int j = 2 * i; o.f(i); } 0: iconst_2 1: iload_0 2: imul 3: istore_2 4: aload_1 5: iload_2 6: invokevirtual <f:(i)v> 9: return void x(...) {... MyClass o =...; CodePointer cp =...;... } idealised stack frame layout GC code eviction j operands locals stack reference map (at instruction 6) 2 1 o i arguments 0 <r> return address stack reference map builder Maxine VM 25

95 Stack Reference Maps and Tagged Code Pointers static void m(int i, MyClass o) { int j = 2 * i; o.f(i); } 0: iconst_2 1: iload_0 2: imul 3: istore_2 4: aload_1 5: iload_2 6: invokevirtual <f:(i)v> 9: return void x(...) {... MyClass o =...; CodePointer cp =...;... } idealised stack frame layout GC code eviction j operands locals stack reference map (at instruction 6) ignore tagged refs, scan others 2 1 o i arguments 0 <r> return address stack reference map builder Maxine VM 25

96 Stack Reference Maps and Tagged Code Pointers static void m(int i, MyClass o) { int j = 2 * i; o.f(i); } 0: iconst_2 1: iload_0 2: imul 3: istore_2 4: aload_1 5: iload_2 6: invokevirtual <f:(i)v> 9: return void x(...) {... MyClass o =...; CodePointer cp =...;... } idealised stack frame layout GC code eviction j operands locals stack reference map (at instruction 6) ignore tagged refs, scan others regard and relocate only tagged refs 2 1 o i arguments 0 <r> return address stack reference map builder Maxine VM 25

97 <Insert Thank you for your attention. Acknowledgements: these slides are joint work of the VM research group. Home page: Source (GPLv2): Mailing list via: 26

98