LLVM Tutorial 02 - わいわいswiftc

LLVMチュートリアルをやってみよう．
今回は第３章くらいまでです．

Transcript

1. Getting start with llvm
   Θ͍Θ͍swiftc
   Yuichi Yoshida
   Senior researcher, DENSO IT Laboratory, Inc.
   #conference
   @sonson_twit
   © 2018 DENSO IT Laboratory, Inc., All rights reserved. Redistribution or public display not permitted without written permission from DENSO IT Laboratory, Inc.
   LLVM Tutorial
   

   View Slide

2. https://peaks.cc/books/iOS12
   

   View Slide

3. LLVM IRಡΊ·͔͢ʁ
   @G = weak global i32 0 ; type of @G is i32*
   @H = weak global i32 0 ; type of @H is i32*
   define i32 @test(i1 %Condition) {
   entry:
   br i1 %Condition, label %cond_true, label %cond_false
   cond_true:
   %X.0 = load i32* @G
   br label %cond_next
   cond_false:
   %X.1 = load i32* @H
   br label %cond_next
   cond_next:
   %X.2 = phi i32 [ %X.1, %cond_false ], [ %X.0, %cond_true ]
   ret i32 %X.2
   }
   

   View Slide

4. LLVMͱ͸ʁ
   • ίϯύΠϥ։ൃج൫
   • ίϯύΠϥΛ࡞ΔͨΊͷϥΠϒϥϦ
   • C++Ͱॻ͔Ε͍ͯΔ
   • ෭࢈෺
   • LLDB
   • clang
   • GNU GCC͕ංେԽ͗ͨ͢͠ର߅അ
   • dragoneggͱݺ͹ΕΔgccͷόοΫΤϯυ΋͋Δ
   • libc++
   

   View Slide

5. $ݴޠ --7.*3 1BTT όΠφϦ
   ίϯύΠϥ 1BTT
   LLVMΛ࢖ͬͯ࡞Δ ͜ͷPass͕ॏཁ
   ৭ʑ༻ҙ͞Ε͍ͯΔ͠
   ࣗ෼Ͱ΋࡞ΕΔ
   x86, ARM, ia64, i386ɾɾɾͳͲ
   όΠφϦ΁ͷม׵͸LLVMͷϞδϡʔϧ
   

   View Slide

6. ΦϨΦϨ
   ݴޠ
   --7.*3 1BTT όΠφϦ
   ίϯύΠϥ 1BTT
   ͑͜͜͞࡞Ε͹
   ͋ͱ͸LLVMʹ೚ͤΒΕΔ
   

   View Slide

7. Swiftͷ৔߹
   4XJGU 4*-
   4*-
   ܕ੍໿
   4*-
   3"8
   4*-
   ࠷దԽ
   --7.*3
   4XJGU*OUFSNFEJBUF-BOHVBHF
   

   View Slide

8. Ԡ༻
   BSN " " 4
   

   View Slide

9. Ԡ༻
   BSN " " 4
   BSN
   "
   "
   4
   BSN
   "
   "
   4
   BSN
   "
   "
   4
   BSN
   "
   "
   4
   

   View Slide

10. Ԡ༻
    CJUDPEF
    " "
    4
    BSN
    --7.Λ࢖͍
    ֤σόΠε޲͚
    ʹ࠷దԽ
    

    View Slide

11. ͳͥɼLLVMʁ
    • ੈͷதͷྲྀΕ
    • ϜʔΞͷ๏ଇͷݶք
    • CPUͷྔతਐԽ͸ɼݶքʹ͍͍ۙͮͯΔʢ෺ཧతʣ
    • ճආ͢Δʹ͸ʁ
    • ฒྻԽ
    • CPU͸ສೳ͗͢ΔˠGPU, FPGA, ASIC
    • ͲΕ΋։ൃ͕ࠓ·Ͱͷͱҧ͍͗ͯ͢ࠔ೉ۃ·Δ
    ιϑτ΢ΣΞͷ։ൃํ๏͕৭ʑࢼ͞Ε͍ͯΔ
    

    View Slide

12. ͜͜਺೥ͷ࿩
    • Halide
    • LLVMΛ࢖͍ɼC/C++ʹ௚઀ฒྻίʔυΛ͔͚Δ
    • fixstars͕͜ΕͰHalide FPGAͳͲΛϦϦʔε
    • temp.split(y, y, yi, 8).parallel(y).vectorize(x, 4);
    • Tensorflow for Swift
    • SwiftͰػցֶशͷߦྻܭࢉ΍ฒྻܭࢉΛ΍Γ͍ͨ
    • Julia
    • ਺ֶతͳܭࢉΛඳ͖΍ͨ͘͢͠ݴޠ
    • LLVMΛ࢖༻
    ཁॴɼཁॴͰLLVM͕ग़ͯ͘Δ
    

    View Slide

13. [11]
    

    View Slide

14. Halide
    [10]
    • εϨου
    • SIMD
    • GPU
    • ͳͲͳͲɾɾɾ
    

    View Slide

15. Halide
    HalideͳΒɼ͜ͷίʔυͰશͯͷ؀ڥʹϙʔτͰ͖Δ
    [10]
    • εϨου
    • SIMD
    • GPU
    • ͳͲͳͲɾɾɾ
    

    View Slide

16. ·ͱΊΔͱ
    • ࠷ۙͷίϯύΠϥ͸LLVM
    • HalideͳͲΛར༻͢Δʹ΋LLVM
    • ΍͸ΓɼLLVMΛ஌Βͳ͍ͱલʹਐΊͳ͍
    

    View Slide

17. LLVM Tutorial
    • LLVMΛ࢖ͬͯΦϨΦϨݴޠͷίϯύΠϥΛ࡞Δ
    • C++ͰllvmΛϥΠϒϥϦͱͯ͠ɼίϯύΠϥΛ։ൃ
    • ಡΉͷʹඞཁͳεΩϧ
    • C++
    • ӳޠ
    • ͳΜ͔brokenͩͬͨΓɼ͚ͩͨ͘ײ͡
    • ਖ਼௚ɼจ͕௕͗ͯ͢ಡΊͳ͔ͬͨΓɼ͋Μ·ϝϯς
    φϯε͞Εͯͳ͍ͱ͜Ζ΋͋Δ
    

    View Slide

18. LLVM Tutorialͱ͸ʁ
    • Tutorials about using LLVM. Includes a tutorial about
    making a custom language with LLVM.
    

    View Slide

19. ϝϞॻ͖ͳͲ
    • https://github.com/sonsongithub/llvm-tutorial
    • LLVM Release Licenseʹج͖ͮɼ຋༁ϝϞ౳Λެ։
    ͍ͯ͠·͢
    

    View Slide

20. TutorialͷϏϧυʹ͍ͭͯ
    • 7.0͸Chapter4Ҏ߱ɼStandard libraryͷؔ਺Λ௚઀Ϧ
    ϯΫͯ͠Δαϯϓϧ͸࢖͑ͳ͍
    • ֎෦ͷAPIݺͼग़͠ʹࣦഊͯ͠Ϋϥογϡ͢Δ
    • 6.0͸OK
    • github͔ΒνΣοΫΞ΢τ͠ɼϏϧυ͢Δͷ͕ྑ͍
    • ֬ೝͨ͠؀ڥ
    • macOS, brewͱιʔε͔ΒϏϧυ
    • Linux, aptͷύοέʔδͱιʔε͔ΒϏϧυ
    

    View Slide

21. ίϯύΠϥͷதʹ
    /// ίϯύΠϥͷதͰɼKaleidoscope͔Βݺ΂Δؔ਺Λఆٛ͢Δ
    /// putchard - putchar that takes a double and returns 0.
    extern "C" DLLEXPORT double putchard(double X) {
    fputc((char)X, stderr);
    return 0;
    }
    /// printd - printf that takes a double prints it as "%f\n",
    returning 0.
    extern "C" DLLEXPORT double printd(double X) {
    fprintf(stderr, "%f\n", X);
    return 0;
    }
    

    View Slide

22. TutorialϏϧυํ๏
    • macOS
    • clang++ɾɾɾdeveloper tool
    • llvm-configɾɾɾιʔε͔ΒࣗલͰϏϧυͨ͠΋ͷ
    • linux
    • g++ɾɾɾbuild essentials
    • llvm-configɾɾɾaptͰπʔϧΛΠϯετʔϧ
    • -rdynamic,-Wl,–export-dynamic͕ඞཁͳ͜ͱ΋͋Δ
    clang++ ./toy.cpp `../../../build/bin/llvm-config --cxxflags --ldflags --libs --libfiles --system-libs`
    

    View Slide

23. ಺༁
    • ͸͡Ίʹ
    • ύʔαͱந৅ߏจ໦
    • LLVM IRίʔυੜ੒
    • JITͱ࠷దԽ
    • ੍ޚϑϩʔ
    • ԋࢉࢠͷఆٛ
    • mutableͷಋೖ
    • ίϯύΠϧ
    • σόοά৘ใ
    • ·ͱΊ
    

    View Slide

24. LLVM IRͷಛ௃
    • Ϩδελɼม਺͸ແݶʹ͋Δ
    • ̏൪஍ίʔυΈ͍ͨͳײ͡
    • SSA(Static Single Assignment form)
    • ੩త୯Ұ୅ೖɾɾɾม਺͸มԽ͠ͳ͍
    • ܕ͸ݫີ
    a = (b + c) ∗ d
    ʢʴɼ̱ɼ̲ɼtmpaddʣ
    ʢˎɼtmpaddɼ̳ɼaʣ
    

    View Slide

25. LLVM IRͷαϯϓϧ
    define double @baz(double %x) {
    entry:
    %ifcond = fcmp one double %x, 0.000000e+00
    br i1 %ifcond, label %then, label %else
    then: ; preds = %entry
    %calltmp = call double @foo()
    br label %ifcont
    else: ; preds = %entry
    %calltmp1 = call double @bar()
    br label %ifcont
    ifcont: ; preds = %else, %then
    %iftmp = phi double [ %calltmp, %then ], [ %calltmp1, %else ]
    ret double %iftmp
    }
    

    View Slide

26. ։ൃ͢ΔݴޠKaleidoscope
    • ໨ඪͱ࢓༷
    • doubleܕͷΈ
    • ؔ਺͕࢖͑Δ
    • Standard libraryͷؔ਺Λ࢖͑Δ
    • if/elseจ
    • forจ
    • Ϣʔβఆٛԋࢉࢠ
    • REPL/JIT
    

    View Slide

27. ࠷ॳ
    • ̍ߦΛίϯύΠϧ͢Δ
    • ม਺͸ͳ͍
    • ؔ਺͸͋Δ
    • math.hͱ͔͸࢖͑Δ
    • if/else/forจ͸ͳ͠ɾɾɾޙ͔Β௥Ճ
    • ͭ·ΓνϡʔϦϯά׬શͰ͸ͳ͍
    

    View Slide

28. ͜Ε͔ΒC++Ͱ࡞Δ΋ͷ
    ,BMFJEPTDPQFͷιʔείʔυ
    ΦϨΦϨݴޠͷιʔείʔυ
    
    ίϯύΠϥ
    --7.*3΍όΠφϦ
    ,BMFJEPTDPQFΛղऍ͠
    --7.Λ࢖ͬͯɼ--7.*3Λੜ੒͠ɼ
    όΠφϦΛు͘
    ͭ·ΓɼC++Ͱݴޠͷ࢓༷Λॻ͍͍ͯ͘ײ͡
    

    View Slide

29. ࠷ॳͷҰาʙޠኮղੳ(Lexer)
    • ·ͣιʔείʔυΛݻ·Γʹ෼ׂ͍ͯ͘͠
    extern cos(0.1);
    ֎෦એݴ
    tok_extern
    Կ͔ͷࣝผࢠ
    tok_identifier
    ਺஋τʔΫϯ
    tok_number
    Χοίͱηϛίϩϯ͸ޠኮʹ౰ͨΒͳ͍ͷͰɼ
    ͱΓ͋͑ͣඈ͹͢
    

    View Slide

30. ୈೋาʙந৅ߏจ໦
    def hoge(a b c) a+(b+c);
    FunctionAST
    Proto
    PrototypeAST
    Body
    BinaryExprAST - +
    LHS
    VariableExprAST - a
    RHS
    BinaryExprAST - +
    LHS
    VariableExprAST - b
    RHS
    VariableExprAST - c
    ԋࢉࢠͷ༏ઌ౓ʹج͖ͮɼਂ͞༏ઌͰߏจ໦Λ࡞͍ͬͯ͘
    

    View Slide

31. ந৅ߏจ໦ͷΫϥε
    class ExprAST {
    public:
    virtual ~ExprAST() = default;
    virtual Value *codegen() = 0;
    };
    

    View Slide

32. ม਺Λอ࣋͢Δϊʔυ
    class VariableExprAST : public ExprAST {
    std::string Name;
    public:
    VariableExprAST(const std::string &Name) : Name(Name) {}
    Value *codegen() override;
    };
    

    View Slide

33. ೋ߲ԋࢉࢠ
    class BinaryExprAST : public ExprAST {
    char Op;
    std::unique_ptr LHS, RHS;
    public:
    BinaryExprAST(char Op, std::unique_ptr LHS,
    std::unique_ptr RHS)
    : Op(Op), LHS(std::move(LHS)), RHS(std::move(RHS)) {}
    Value *codegen() override;
    };
    

    View Slide

34. ؔ਺એݴ
    class PrototypeAST {
    std::string Name;
    std::vector Args;
    public:
    PrototypeAST(const std::string &Name, std::vector Args)
    : Name(Name), Args(std::move(Args)) {}
    Function *codegen();
    const std::string &getName() const { return Name; }
    };
    

    View Slide

35. ؔ਺࣮૷
    class FunctionAST {
    std::unique_ptr Proto;
    std::unique_ptr Body;
    public:
    FunctionAST(std::unique_ptr Proto,
    std::unique_ptr Body)
    : Proto(std::move(Proto)), Body(std::move(Body)) {}
    Function *codegen();
    };
    

    View Slide

36. ந৅ߏจ໦
    def hoge(a b c) a+(b+c);
    FunctionAST
    Proto
    PrototypeAST
    Body
    BinaryExprAST - +
    LHS
    VariableExprAST - a
    RHS
    BinaryExprAST - +
    LHS
    VariableExprAST - b
    RHS
    VariableExprAST - c
    ԋࢉࢠͷ༏ઌ౓ʹج͖ͮɼਂ͞༏ઌͰߏจ໦Λ࡞͍ͬͯ͘
    

    View Slide

37. ந৅ߏจ໦
    def hoge(a b c) a+b+c;
    FunctionAST
    Proto
    PrototypeAST
    Body
    BinaryExprAST - +
    LHS
    VariableExprAST - c
    RHS
    BinaryExprAST - +
    LHS
    VariableExprAST - a
    RHS
    VariableExprAST - b
    ԋࢉࢠͷ༏ઌ౓ʹج͖ͮɼਂ͞༏ઌͰߏจ໦Λ࡞͍ͬͯ͘
    

    View Slide

38. LLVM IRΛੜ੒͢Δ
    class ExprAST {
    public:
    virtual ~ExprAST() = default;
    virtual Value *codegen() = 0;
    };
    ֤ϊʔυ͔ΒLLVM IRΛੜ੒͢Δ
    ߏจ໦ͷϊʔυ͝ͱʹੜ੒ͷͨΊͷίʔυΛ࣮૷͍ͯ͘͠
    

    View Slide

39. codegenͷ࢓ࣄ
    • llvm::ValueΫϥεΛฦ͢
    • ͜ͷΫϥε͕LLVM IRΛు͍ͯ͘ΕΔ
    • llvm::Value
    • ੩త୯Ұ୅ೖϨδελΛࣔ͢
    if (auto *FnIR = ProtoAST->codegen()) {
    fprintf(stderr, "Read extern: ");
    FnIR->print(errs());
    fprintf(stderr, "\n");
    }
    

    View Slide

40. ೋ߲ԋࢉࢠͷ৔߹
    Value *BinaryExprAST::codegen() {
    Value *L = LHS->codegen();
    Value *R = RHS->codegen();
    if (!L || !R)
    return nullptr;
    switch (Op) {
    case '+':
    return Builder.CreateFAdd(L, R, "addtmp");
    case '-':
    return Builder.CreateFSub(L, R, "subtmp");
    case '*':
    return Builder.CreateFMul(L, R, "multmp");
    case '<':
    L = Builder.CreateFCmpULT(L, R, "cmptmp");
    return Builder.CreateUIToFP(
    L,
    Type::getDoubleTy(TheContext), "booltmp");
    default:
    return LogErrorV("invalid binary operator");
    }
    }
    ܭࢉ݁ՌΛอଘ͢ΔϨδελ໊
    

    View Slide

41. ؔ਺એݴ
    Function *PrototypeAST::codegen() {
    std::vector Doubles(
    Args.size(),
    Type::getDoubleTy(TheContext)
    );
    FunctionType *FT = FunctionType::get(
    Type::getDoubleTy(TheContext),
    Doubles,
    false
    );
    Function *F = Function::Create(
    FT,
    Function::ExternalLinkage,
    Name,
    TheModule.get()
    );
    // Set names for all arguments.
    unsigned Idx = 0;
    for (auto &Arg : F->args())
    Arg.setName(Args[Idx++]);
    return F;
    }
    

    View Slide

42. ؔ਺࣮૷
    Function *FunctionAST::codegen() {
    Function *TheFunction = TheModule->getFunction(
    Proto->getName());
    if (!TheFunction)
    TheFunction = Proto->codegen();
    if (!TheFunction)
    return nullptr;
    // Create a new basic block to start insertion into.
    BasicBlock *BB = BasicBlock::Create(
    TheContext, "entry", TheFunction);
    Builder.SetInsertPoint(BB);
    

    View Slide

43. ؔ਺࣮૷
    // Record the function arguments in the NamedValues map.
    NamedValues.clear();
    for (auto &Arg : TheFunction->args())
    NamedValues[Arg.getName()] = &Arg;
    if (Value *RetVal = Body->codegen()) {
    // Finish off the function.
    Builder.CreateRet(RetVal);
    // Validate the generated code, checking for consistency.
    verifyFunction(*TheFunction);
    return TheFunction;
    }
    // Error reading body, remove function.
    TheFunction->eraseFromParent();
    return nullptr;
    }
    

    View Slide

44. ந৅ߏจ໦͔ΒLLVM IRΛ࡞Δ
    def hoge(a b c) a+b+c;
    FunctionAST
    Proto
    PrototypeAST
    Body
    BinaryExprAST - +
    LHS
    VariableExprAST - c
    RHS
    BinaryExprAST - +
    LHS
    VariableExprAST - a
    RHS
    VariableExprAST - b
    

    View Slide

45. codegen
    def hoge(a b c) a+b+c;
    FunctionAST
    Proto
    PrototypeAST
    Body
    BinaryExprAST - +
    LHS
    VariableExprAST - c
    RHS
    BinaryExprAST - +
    LHS
    VariableExprAST - a
    RHS
    VariableExprAST - b
    BEEUNQGBEEEPVCMFB C
    

    View Slide

46. codegen
    def hoge(a b c) a+b+c;
    FunctionAST
    Proto
    PrototypeAST
    Body
    BinaryExprAST - +
    LHS
    VariableExprAST - c
    RHS
    BinaryExprAST - +
    LHS
    VariableExprAST - a
    RHS
    VariableExprAST - b
    BEEUNQGBEEEPVCMFB C
    BEEUNQGBEEEPVCMFBEEUNQ D
    

    View Slide

47. codegen
    def hoge(a b c) a+b+c;
    FunctionAST
    Proto
    PrototypeAST
    Body
    BinaryExprAST - +
    LHS
    VariableExprAST - c
    RHS
    BinaryExprAST - +
    LHS
    VariableExprAST - a
    RHS
    VariableExprAST - b
    EFpOFEPVCMF!IPHF EPVCMFB EPVCMFC EPVCMFD
    \
    FOUSZ
    BEEUNQGBEEEPVCMFB C
    BEEUNQGBEEEPVCMFBEEUNQ D
    SFUEPVCMFBEEUNQ
    ^
    

    View Slide

48. ·ͱΊ
    • LLVMʹ͍ͭͯ
    • എܠɼྺ࢙ɼԠ༻
    • LLVM Tutorial
    • ந৅ߏจ໦
    • LLVM IRͷੜ੒
    • ͔͜͜Β͕͓΋͠Ζ͍
    • PassʹΑΔ࠷దԽ
    • ੍ޚϑϩʔͷ࣮૷
    • mutableม਺ͷಋೖ
    

    View Slide

49. ଓ͘
    

    View Slide