r217556 - Thread Safety Analysis: major update to thread safety TIL.
Delesley Hutchins
delesley at google.com
Thu Sep 11 15:02:43 PDT 2014
Thanks again! To answer your question, this code is intended for use both
inside and outside of clang, as part of a long term project to deploy
thread safety analysis to languages other than C++. It can be used
independently of clang with appropriate include magic, and I have test
frameworks that test it as such; the std::ostream stuff is a relic of that.
I did not mean to submit the commented code. :-)
-DeLesley
On Thu, Sep 11, 2014 at 1:03 PM, Justin Bogner <mail at justinbogner.com>
wrote:
> DeLesley Hutchins <delesley at google.com> writes:
> > Author: delesley
> > Date: Wed Sep 10 17:12:52 2014
> > New Revision: 217556
> >
> > URL: http://llvm.org/viewvc/llvm-project?rev=217556&view=rev
> > Log:
> > Thread Safety Analysis: major update to thread safety TIL.
>
> This was causing some bots to hang due to the implementation of
> operator<<(std::ostream&, StringRef) deciding to self recurse. I've
> attempted to fix that in r217621.
>
> I also noticed that you commented out a version of this same operator<<
> in ThreadSafetyTraverse.h that didn't have the recursion problem. Please
> don't commit commented out code - just remove it.
>
> Finally, is there a good reason this is using ostream rather than llvm's
> raw_ostream?
>
> > Numerous changes, including:
> > * Changed the way variables and instructions are handled in basic
> blocks to
> > be more efficient.
> > * Eliminated SExprRef.
> > * Simplified futures.
> > * Fixed documentation.
> > * Compute dominator and post dominator trees.
> >
> > Modified:
> > cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyCommon.h
> > cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyLogical.h
> > cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyOps.def
> > cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTIL.h
> > cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTraverse.h
> > cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyUtil.h
> > cfe/trunk/lib/Analysis/ThreadSafetyCommon.cpp
> > cfe/trunk/lib/Analysis/ThreadSafetyTIL.cpp
> >
> > Modified: cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyCommon.h
> > URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyCommon.h?rev=217556&r1=217555&r2=217556&view=diff
> >
> ==============================================================================
> > --- cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyCommon.h
> (original)
> > +++ cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyCommon.h Wed
> Sep 10 17:12:52 2014
> > @@ -477,9 +477,9 @@ private:
> > // Indexed by clang BlockID.
> >
> > LVarDefinitionMap CurrentLVarMap;
> > - std::vector<til::Variable*> CurrentArguments;
> > - std::vector<til::Variable*> CurrentInstructions;
> > - std::vector<til::Variable*> IncompleteArgs;
> > + std::vector<til::Phi*> CurrentArguments;
> > + std::vector<til::SExpr*> CurrentInstructions;
> > + std::vector<til::Phi*> IncompleteArgs;
> > til::BasicBlock *CurrentBB;
> > BlockInfo *CurrentBlockInfo;
> > };
> >
> > Modified: cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyLogical.h
> > URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyLogical.h?rev=217556&r1=217555&r2=217556&view=diff
> >
> ==============================================================================
> > --- cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyLogical.h
> (original)
> > +++ cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyLogical.h Wed
> Sep 10 17:12:52 2014
> > @@ -41,13 +41,13 @@ private:
> > };
> >
> > class Terminal : public LExpr {
> > - til::SExprRef Expr;
> > + til::SExpr *Expr;
> >
> > public:
> > Terminal(til::SExpr *Expr) : LExpr(LExpr::Terminal), Expr(Expr) {}
> >
> > - const til::SExpr *expr() const { return Expr.get(); }
> > - til::SExpr *expr() { return Expr.get(); }
> > + const til::SExpr *expr() const { return Expr; }
> > + til::SExpr *expr() { return Expr; }
> >
> > static bool classof(const LExpr *E) { return E->kind() ==
> LExpr::Terminal; }
> > };
> >
> > Modified: cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyOps.def
> > URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyOps.def?rev=217556&r1=217555&r2=217556&view=diff
> >
> ==============================================================================
> > --- cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyOps.def
> (original)
> > +++ cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyOps.def Wed
> Sep 10 17:12:52 2014
> > @@ -44,8 +44,11 @@ TIL_OPCODE_DEF(Cast)
> > TIL_OPCODE_DEF(SCFG)
> > TIL_OPCODE_DEF(BasicBlock)
> > TIL_OPCODE_DEF(Phi)
> > +
> > +// Terminator instructions
> > TIL_OPCODE_DEF(Goto)
> > TIL_OPCODE_DEF(Branch)
> > +TIL_OPCODE_DEF(Return)
> >
> > // pseudo-terms
> > TIL_OPCODE_DEF(Identifier)
> >
> > Modified: cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTIL.h
> > URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTIL.h?rev=217556&r1=217555&r2=217556&view=diff
> >
> ==============================================================================
> > --- cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTIL.h
> (original)
> > +++ cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTIL.h Wed Sep
> 10 17:12:52 2014
> > @@ -63,24 +63,27 @@ namespace threadSafety {
> > namespace til {
> >
> >
> > +/// Enum for the different distinct classes of SExpr
> > enum TIL_Opcode {
> > #define TIL_OPCODE_DEF(X) COP_##X,
> > #include "ThreadSafetyOps.def"
> > #undef TIL_OPCODE_DEF
> > };
> >
> > +/// Opcode for unary arithmetic operations.
> > enum TIL_UnaryOpcode : unsigned char {
> > UOP_Minus, // -
> > UOP_BitNot, // ~
> > UOP_LogicNot // !
> > };
> >
> > +/// Opcode for binary arithmetic operations.
> > enum TIL_BinaryOpcode : unsigned char {
> > + BOP_Add, // +
> > + BOP_Sub, // -
> > BOP_Mul, // *
> > BOP_Div, // /
> > BOP_Rem, // %
> > - BOP_Add, // +
> > - BOP_Sub, // -
> > BOP_Shl, // <<
> > BOP_Shr, // >>
> > BOP_BitAnd, // &
> > @@ -90,10 +93,11 @@ enum TIL_BinaryOpcode : unsigned char {
> > BOP_Neq, // !=
> > BOP_Lt, // <
> > BOP_Leq, // <=
> > - BOP_LogicAnd, // &&
> > - BOP_LogicOr // ||
> > + BOP_LogicAnd, // && (no short-circuit)
> > + BOP_LogicOr // || (no short-circuit)
> > };
> >
> > +/// Opcode for cast operations.
> > enum TIL_CastOpcode : unsigned char {
> > CAST_none = 0,
> > CAST_extendNum, // extend precision of numeric type
> > @@ -107,21 +111,24 @@ const TIL_Opcode COP_Min = COP_Fu
> > const TIL_Opcode COP_Max = COP_Branch;
> > const TIL_UnaryOpcode UOP_Min = UOP_Minus;
> > const TIL_UnaryOpcode UOP_Max = UOP_LogicNot;
> > -const TIL_BinaryOpcode BOP_Min = BOP_Mul;
> > +const TIL_BinaryOpcode BOP_Min = BOP_Add;
> > const TIL_BinaryOpcode BOP_Max = BOP_LogicOr;
> > const TIL_CastOpcode CAST_Min = CAST_none;
> > const TIL_CastOpcode CAST_Max = CAST_toInt;
> >
> > +/// Return the name of a unary opcode.
> > StringRef getUnaryOpcodeString(TIL_UnaryOpcode Op);
> > +
> > +/// Return the name of a binary opcode.
> > StringRef getBinaryOpcodeString(TIL_BinaryOpcode Op);
> >
> >
> > -// ValueTypes are data types that can actually be held in registers.
> > -// All variables and expressions must have a vBNF_Nonealue type.
> > -// Pointer types are further subdivided into the various heap-allocated
> > -// types, such as functions, records, etc.
> > -// Structured types that are passed by value (e.g. complex numbers)
> > -// require special handling; they use BT_ValueRef, and size ST_0.
> > +/// ValueTypes are data types that can actually be held in registers.
> > +/// All variables and expressions must have a value type.
> > +/// Pointer types are further subdivided into the various heap-allocated
> > +/// types, such as functions, records, etc.
> > +/// Structured types that are passed by value (e.g. complex numbers)
> > +/// require special handling; they use BT_ValueRef, and size ST_0.
> > struct ValueType {
> > enum BaseType : unsigned char {
> > BT_Void = 0,
> > @@ -247,8 +254,10 @@ inline ValueType ValueType::getValueType
> > }
> >
> >
> > +class BasicBlock;
> > +
> >
> > -// Base class for AST nodes in the typed intermediate language.
> > +/// Base class for AST nodes in the typed intermediate language.
> > class SExpr {
> > public:
> > TIL_Opcode opcode() const { return static_cast<TIL_Opcode>(Opcode); }
> > @@ -267,71 +276,47 @@ public:
> > // template <class C> typename C::CType compare(CType* E, C& Cmp) {
> > // compare all subexpressions, following the comparator interface
> > // }
> > -
> > void *operator new(size_t S, MemRegionRef &R) {
> > return ::operator new(S, R);
> > }
> >
> > - // SExpr objects cannot be deleted.
> > + /// SExpr objects cannot be deleted.
> > // This declaration is public to workaround a gcc bug that breaks
> building
> > // with REQUIRES_EH=1.
> > void operator delete(void *) LLVM_DELETED_FUNCTION;
> >
> > + /// Returns the instruction ID for this expression.
> > + /// All basic block instructions have a unique ID (i.e. virtual
> register).
> > + unsigned id() const { return SExprID; }
> > +
> > + /// Returns the block, if this is an instruction in a basic block,
> > + /// otherwise returns null.
> > + BasicBlock* block() const { return Block; }
> > +
> > + /// Set the basic block and instruction ID for this expression.
> > + void setID(BasicBlock *B, unsigned id) { Block = B; SExprID = id; }
> > +
> > protected:
> > - SExpr(TIL_Opcode Op) : Opcode(Op), Reserved(0), Flags(0) {}
> > - SExpr(const SExpr &E) : Opcode(E.Opcode), Reserved(0), Flags(E.Flags)
> {}
> > + SExpr(TIL_Opcode Op)
> > + : Opcode(Op), Reserved(0), Flags(0), SExprID(0), Block(nullptr) {}
> > + SExpr(const SExpr &E)
> > + : Opcode(E.Opcode), Reserved(0), Flags(E.Flags), SExprID(0),
> > + Block(nullptr) {}
> >
> > const unsigned char Opcode;
> > unsigned char Reserved;
> > unsigned short Flags;
> > + unsigned SExprID;
> > + BasicBlock* Block;
> >
> > private:
> > SExpr() LLVM_DELETED_FUNCTION;
> >
> > - // SExpr objects must be created in an arena.
> > + /// SExpr objects must be created in an arena.
> > void *operator new(size_t) LLVM_DELETED_FUNCTION;
> > };
> >
> >
> > -// Class for owning references to SExprs.
> > -// Includes attach/detach logic for counting variable references and
> lazy
> > -// rewriting strategies.
> > -class SExprRef {
> > -public:
> > - SExprRef() : Ptr(nullptr) { }
> > - SExprRef(std::nullptr_t P) : Ptr(nullptr) { }
> > - SExprRef(SExprRef &&R) : Ptr(R.Ptr) { R.Ptr = nullptr; }
> > -
> > - // Defined after Variable and Future, below.
> > - inline SExprRef(SExpr *P);
> > - inline ~SExprRef();
> > -
> > - SExpr *get() { return Ptr; }
> > - const SExpr *get() const { return Ptr; }
> > -
> > - SExpr *operator->() { return get(); }
> > - const SExpr *operator->() const { return get(); }
> > -
> > - SExpr &operator*() { return *Ptr; }
> > - const SExpr &operator*() const { return *Ptr; }
> > -
> > - bool operator==(const SExprRef &R) const { return Ptr == R.Ptr; }
> > - bool operator!=(const SExprRef &R) const { return !operator==(R); }
> > - bool operator==(const SExpr *P) const { return Ptr == P; }
> > - bool operator!=(const SExpr *P) const { return !operator==(P); }
> > - bool operator==(std::nullptr_t) const { return Ptr == nullptr; }
> > - bool operator!=(std::nullptr_t) const { return Ptr != nullptr; }
> > -
> > - inline void reset(SExpr *E);
> > -
> > -private:
> > - inline void attach();
> > - inline void detach();
> > -
> > - SExpr *Ptr;
> > -};
> > -
> > -
> > // Contains various helper functions for SExprs.
> > namespace ThreadSafetyTIL {
> > inline bool isTrivial(const SExpr *E) {
> > @@ -343,62 +328,64 @@ namespace ThreadSafetyTIL {
> > // Nodes which declare variables
> > class Function;
> > class SFunction;
> > -class BasicBlock;
> > class Let;
> >
> >
> > -// A named variable, e.g. "x".
> > -//
> > -// There are two distinct places in which a Variable can appear in the
> AST.
> > -// A variable declaration introduces a new variable, and can occur in 3
> places:
> > -// Let-expressions: (Let (x = t) u)
> > -// Functions: (Function (x : t) u)
> > -// Self-applicable functions (SFunction (x) t)
> > -//
> > -// If a variable occurs in any other location, it is a reference to an
> existing
> > -// variable declaration -- e.g. 'x' in (x * y + z). To save space, we
> don't
> > -// allocate a separate AST node for variable references; a reference is
> just a
> > -// pointer to the original declaration.
> > +/// A named variable, e.g. "x".
> > +///
> > +/// There are two distinct places in which a Variable can appear in the
> AST.
> > +/// A variable declaration introduces a new variable, and can occur in
> 3 places:
> > +/// Let-expressions: (Let (x = t) u)
> > +/// Functions: (Function (x : t) u)
> > +/// Self-applicable functions (SFunction (x) t)
> > +///
> > +/// If a variable occurs in any other location, it is a reference to an
> existing
> > +/// variable declaration -- e.g. 'x' in (x * y + z). To save space, we
> don't
> > +/// allocate a separate AST node for variable references; a reference
> is just a
> > +/// pointer to the original declaration.
> > class Variable : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Variable; }
> >
> > - // Let-variable, function parameter, or self-variable
> > enum VariableKind {
> > - VK_Let,
> > - VK_LetBB,
> > - VK_Fun,
> > - VK_SFun
> > + VK_Let, ///< Let-variable
> > + VK_Fun, ///< Function parameter
> > + VK_SFun ///< SFunction (self) parameter
> > };
> >
> > - // These are defined after SExprRef contructor, below
> > - inline Variable(SExpr *D, const clang::ValueDecl *Cvd = nullptr);
> > - inline Variable(StringRef s, SExpr *D = nullptr);
> > - inline Variable(const Variable &Vd, SExpr *D);
> > + Variable(StringRef s, SExpr *D = nullptr)
> > + : SExpr(COP_Variable), Name(s), Definition(D), Cvdecl(nullptr) {
> > + Flags = VK_Let;
> > + }
> > + Variable(SExpr *D, const clang::ValueDecl *Cvd = nullptr)
> > + : SExpr(COP_Variable), Name(Cvd ? Cvd->getName() : "_x"),
> > + Definition(D), Cvdecl(Cvd) {
> > + Flags = VK_Let;
> > + }
> > + Variable(const Variable &Vd, SExpr *D) // rewrite constructor
> > + : SExpr(Vd), Name(Vd.Name), Definition(D), Cvdecl(Vd.Cvdecl) {
> > + Flags = Vd.kind();
> > + }
> >
> > + /// Return the kind of variable (let, function param, or self)
> > VariableKind kind() const { return static_cast<VariableKind>(Flags); }
> >
> > + /// Return the name of the variable, if any.
> > StringRef name() const { return Name; }
> > +
> > + /// Return the clang declaration for this variable, if any.
> > const clang::ValueDecl *clangDecl() const { return Cvdecl; }
> >
> > - // Returns the definition (for let vars) or type (for parameter &
> self vars)
> > - SExpr *definition() { return Definition.get(); }
> > - const SExpr *definition() const { return Definition.get(); }
> > -
> > - void attachVar() const { ++NumUses; }
> > - void detachVar() const { assert(NumUses > 0); --NumUses; }
> > -
> > - unsigned getID() const { return Id; }
> > - unsigned getBlockID() const { return BlockID; }
> > -
> > - void setName(StringRef S) { Name = S; }
> > - void setID(unsigned Bid, unsigned I) {
> > - BlockID = static_cast<unsigned short>(Bid);
> > - Id = static_cast<unsigned short>(I);
> > - }
> > - void setClangDecl(const clang::ValueDecl *VD) { Cvdecl = VD; }
> > - void setDefinition(SExpr *E);
> > + /// Return the definition of the variable.
> > + /// For let-vars, this is the setting expression.
> > + /// For function and self parameters, it is the type of the variable.
> > + SExpr *definition() { return Definition; }
> > + const SExpr *definition() const { return Definition; }
> > +
> > + void setName(StringRef S) { Name = S; }
> > void setKind(VariableKind K) { Flags = K; }
> > + void setDefinition(SExpr *E) { Definition = E; }
> > + void setClangDecl(const clang::ValueDecl *VD) { Cvdecl = VD; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -418,17 +405,13 @@ private:
> > friend class Let;
> >
> > StringRef Name; // The name of the variable.
> > - SExprRef Definition; // The TIL type or definition
> > + SExpr* Definition; // The TIL type or definition
> > const clang::ValueDecl *Cvdecl; // The clang declaration for this
> variable.
> > -
> > - unsigned short BlockID;
> > - unsigned short Id;
> > - mutable unsigned NumUses;
> > };
> >
> >
> > -// Placeholder for an expression that has not yet been created.
> > -// Used to implement lazy copy and rewriting strategies.
> > +/// Placeholder for an expression that has not yet been created.
> > +/// Used to implement lazy copy and rewriting strategies.
> > class Future : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Future; }
> > @@ -439,22 +422,14 @@ public:
> > FS_done
> > };
> >
> > - Future() :
> > - SExpr(COP_Future), Status(FS_pending), Result(nullptr),
> Location(nullptr)
> > - {}
> > + Future() : SExpr(COP_Future), Status(FS_pending), Result(nullptr) {}
> > +
> > private:
> > virtual ~Future() LLVM_DELETED_FUNCTION;
> > -public:
> > -
> > - // Registers the location in the AST where this future is stored.
> > - // Forcing the future will automatically update the AST.
> > - static inline void registerLocation(SExprRef *Member) {
> > - if (Future *F = dyn_cast_or_null<Future>(Member->get()))
> > - F->Location = Member;
> > - }
> >
> > +public:
> > // A lazy rewriting strategy should subclass Future and override this
> method.
> > - virtual SExpr *create() { return nullptr; }
> > + virtual SExpr *compute() { return nullptr; }
> >
> > // Return the result of this future if it exists, otherwise return
> null.
> > SExpr *maybeGetResult() const {
> > @@ -465,8 +440,7 @@ public:
> > SExpr *result() {
> > switch (Status) {
> > case FS_pending:
> > - force();
> > - return Result;
> > + return force();
> > case FS_evaluating:
> > return nullptr; // infinite loop; illegal recursion.
> > case FS_done:
> > @@ -488,81 +462,14 @@ public:
> > }
> >
> > private:
> > - // Force the future.
> > - inline void force();
> > + SExpr* force();
> >
> > FutureStatus Status;
> > SExpr *Result;
> > - SExprRef *Location;
> > };
> >
> >
> > -inline void SExprRef::attach() {
> > - if (!Ptr)
> > - return;
> > -
> > - TIL_Opcode Op = Ptr->opcode();
> > - if (Op == COP_Variable) {
> > - cast<Variable>(Ptr)->attachVar();
> > - } else if (Op == COP_Future) {
> > - cast<Future>(Ptr)->registerLocation(this);
> > - }
> > -}
> > -
> > -inline void SExprRef::detach() {
> > - if (Ptr && Ptr->opcode() == COP_Variable) {
> > - cast<Variable>(Ptr)->detachVar();
> > - }
> > -}
> > -
> > -inline SExprRef::SExprRef(SExpr *P) : Ptr(P) {
> > - attach();
> > -}
> > -
> > -inline SExprRef::~SExprRef() {
> > - detach();
> > -}
> > -
> > -inline void SExprRef::reset(SExpr *P) {
> > - detach();
> > - Ptr = P;
> > - attach();
> > -}
> > -
> > -
> > -inline Variable::Variable(StringRef s, SExpr *D)
> > - : SExpr(COP_Variable), Name(s), Definition(D), Cvdecl(nullptr),
> > - BlockID(0), Id(0), NumUses(0) {
> > - Flags = VK_Let;
> > -}
> > -
> > -inline Variable::Variable(SExpr *D, const clang::ValueDecl *Cvd)
> > - : SExpr(COP_Variable), Name(Cvd ? Cvd->getName() : "_x"),
> > - Definition(D), Cvdecl(Cvd), BlockID(0), Id(0), NumUses(0) {
> > - Flags = VK_Let;
> > -}
> > -
> > -inline Variable::Variable(const Variable &Vd, SExpr *D) // rewrite
> constructor
> > - : SExpr(Vd), Name(Vd.Name), Definition(D), Cvdecl(Vd.Cvdecl),
> > - BlockID(0), Id(0), NumUses(0) {
> > - Flags = Vd.kind();
> > -}
> > -
> > -inline void Variable::setDefinition(SExpr *E) {
> > - Definition.reset(E);
> > -}
> > -
> > -void Future::force() {
> > - Status = FS_evaluating;
> > - SExpr *R = create();
> > - Result = R;
> > - if (Location)
> > - Location->reset(R);
> > - Status = FS_done;
> > -}
> > -
> > -
> > -// Placeholder for C++ expressions that cannot be represented in the
> TIL.
> > +/// Placeholder for expressions that cannot be represented in the TIL.
> > class Undefined : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Undefined; }
> > @@ -585,7 +492,7 @@ private:
> > };
> >
> >
> > -// Placeholder for a wildcard that matches any other expression.
> > +/// Placeholder for a wildcard that matches any other expression.
> > class Wildcard : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Wildcard; }
> > @@ -716,8 +623,8 @@ typename V::R_SExpr Literal::traverse(V
> > }
> >
> >
> > -// Literal pointer to an object allocated in memory.
> > -// At compile time, pointer literals are represented by symbolic names.
> > +/// A Literal pointer to an object allocated in memory.
> > +/// At compile time, pointer literals are represented by symbolic names.
> > class LiteralPtr : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_LiteralPtr; }
> > @@ -743,9 +650,9 @@ private:
> > };
> >
> >
> > -// A function -- a.k.a. lambda abstraction.
> > -// Functions with multiple arguments are created by currying,
> > -// e.g. (function (x: Int) (function (y: Int) (add x y)))
> > +/// A function -- a.k.a. lambda abstraction.
> > +/// Functions with multiple arguments are created by currying,
> > +/// e.g. (Function (x: Int) (Function (y: Int) (Code { return x + y })))
> > class Function : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Function; }
> > @@ -762,8 +669,8 @@ public:
> > Variable *variableDecl() { return VarDecl; }
> > const Variable *variableDecl() const { return VarDecl; }
> >
> > - SExpr *body() { return Body.get(); }
> > - const SExpr *body() const { return Body.get(); }
> > + SExpr *body() { return Body; }
> > + const SExpr *body() const { return Body; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -790,13 +697,13 @@ public:
> >
> > private:
> > Variable *VarDecl;
> > - SExprRef Body;
> > + SExpr* Body;
> > };
> >
> >
> > -// A self-applicable function.
> > -// A self-applicable function can be applied to itself. It's useful for
> > -// implementing objects and late binding
> > +/// A self-applicable function.
> > +/// A self-applicable function can be applied to itself. It's useful
> for
> > +/// implementing objects and late binding.
> > class SFunction : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_SFunction; }
> > @@ -805,20 +712,20 @@ public:
> > : SExpr(COP_SFunction), VarDecl(Vd), Body(B) {
> > assert(Vd->Definition == nullptr);
> > Vd->setKind(Variable::VK_SFun);
> > - Vd->Definition.reset(this);
> > + Vd->Definition = this;
> > }
> > SFunction(const SFunction &F, Variable *Vd, SExpr *B) // rewrite
> constructor
> > : SExpr(F), VarDecl(Vd), Body(B) {
> > assert(Vd->Definition == nullptr);
> > Vd->setKind(Variable::VK_SFun);
> > - Vd->Definition.reset(this);
> > + Vd->Definition = this;
> > }
> >
> > Variable *variableDecl() { return VarDecl; }
> > const Variable *variableDecl() const { return VarDecl; }
> >
> > - SExpr *body() { return Body.get(); }
> > - const SExpr *body() const { return Body.get(); }
> > + SExpr *body() { return Body; }
> > + const SExpr *body() const { return Body; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -842,11 +749,11 @@ public:
> >
> > private:
> > Variable *VarDecl;
> > - SExprRef Body;
> > + SExpr* Body;
> > };
> >
> >
> > -// A block of code -- e.g. the body of a function.
> > +/// A block of code -- e.g. the body of a function.
> > class Code : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() == COP_Code;
> }
> > @@ -855,11 +762,11 @@ public:
> > Code(const Code &C, SExpr *T, SExpr *B) // rewrite constructor
> > : SExpr(C), ReturnType(T), Body(B) {}
> >
> > - SExpr *returnType() { return ReturnType.get(); }
> > - const SExpr *returnType() const { return ReturnType.get(); }
> > + SExpr *returnType() { return ReturnType; }
> > + const SExpr *returnType() const { return ReturnType; }
> >
> > - SExpr *body() { return Body.get(); }
> > - const SExpr *body() const { return Body.get(); }
> > + SExpr *body() { return Body; }
> > + const SExpr *body() const { return Body; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -877,12 +784,12 @@ public:
> > }
> >
> > private:
> > - SExprRef ReturnType;
> > - SExprRef Body;
> > + SExpr* ReturnType;
> > + SExpr* Body;
> > };
> >
> >
> > -// A typed, writable location in memory
> > +/// A typed, writable location in memory
> > class Field : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Field; }
> > @@ -891,11 +798,11 @@ public:
> > Field(const Field &C, SExpr *R, SExpr *B) // rewrite constructor
> > : SExpr(C), Range(R), Body(B) {}
> >
> > - SExpr *range() { return Range.get(); }
> > - const SExpr *range() const { return Range.get(); }
> > + SExpr *range() { return Range; }
> > + const SExpr *range() const { return Range; }
> >
> > - SExpr *body() { return Body.get(); }
> > - const SExpr *body() const { return Body.get(); }
> > + SExpr *body() { return Body; }
> > + const SExpr *body() const { return Body; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -913,12 +820,16 @@ public:
> > }
> >
> > private:
> > - SExprRef Range;
> > - SExprRef Body;
> > + SExpr* Range;
> > + SExpr* Body;
> > };
> >
> >
> > -// Apply an argument to a function
> > +/// Apply an argument to a function.
> > +/// Note that this does not actually call the function. Functions are
> curried,
> > +/// so this returns a closure in which the first parameter has been
> applied.
> > +/// Once all parameters have been applied, Call can be used to invoke
> the
> > +/// function.
> > class Apply : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Apply; }
> > @@ -928,11 +839,11 @@ public:
> > : SExpr(A), Fun(F), Arg(Ar)
> > {}
> >
> > - SExpr *fun() { return Fun.get(); }
> > - const SExpr *fun() const { return Fun.get(); }
> > + SExpr *fun() { return Fun; }
> > + const SExpr *fun() const { return Fun; }
> >
> > - SExpr *arg() { return Arg.get(); }
> > - const SExpr *arg() const { return Arg.get(); }
> > + SExpr *arg() { return Arg; }
> > + const SExpr *arg() const { return Arg; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -950,12 +861,12 @@ public:
> > }
> >
> > private:
> > - SExprRef Fun;
> > - SExprRef Arg;
> > + SExpr* Fun;
> > + SExpr* Arg;
> > };
> >
> >
> > -// Apply a self-argument to a self-applicable function
> > +/// Apply a self-argument to a self-applicable function.
> > class SApply : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_SApply; }
> > @@ -964,18 +875,18 @@ public:
> > SApply(SApply &A, SExpr *Sf, SExpr *Ar = nullptr) // rewrite
> constructor
> > : SExpr(A), Sfun(Sf), Arg(Ar) {}
> >
> > - SExpr *sfun() { return Sfun.get(); }
> > - const SExpr *sfun() const { return Sfun.get(); }
> > + SExpr *sfun() { return Sfun; }
> > + const SExpr *sfun() const { return Sfun; }
> >
> > - SExpr *arg() { return Arg.get() ? Arg.get() : Sfun.get(); }
> > - const SExpr *arg() const { return Arg.get() ? Arg.get() : Sfun.get();
> }
> > + SExpr *arg() { return Arg ? Arg : Sfun; }
> > + const SExpr *arg() const { return Arg ? Arg : Sfun; }
> >
> > bool isDelegation() const { return Arg != nullptr; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > auto Nf = Vs.traverse(Sfun, Vs.subExprCtx(Ctx));
> > - typename V::R_SExpr Na = Arg.get() ? Vs.traverse(Arg,
> Vs.subExprCtx(Ctx))
> > + typename V::R_SExpr Na = Arg ? Vs.traverse(Arg, Vs.subExprCtx(Ctx))
> > : nullptr;
> > return Vs.reduceSApply(*this, Nf, Na);
> > }
> > @@ -989,12 +900,12 @@ public:
> > }
> >
> > private:
> > - SExprRef Sfun;
> > - SExprRef Arg;
> > + SExpr* Sfun;
> > + SExpr* Arg;
> > };
> >
> >
> > -// Project a named slot from a C++ struct or class.
> > +/// Project a named slot from a C++ struct or class.
> > class Project : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Project; }
> > @@ -1009,8 +920,8 @@ public:
> > : SExpr(P), Rec(R), SlotName(P.SlotName), Cvdecl(P.Cvdecl)
> > { }
> >
> > - SExpr *record() { return Rec.get(); }
> > - const SExpr *record() const { return Rec.get(); }
> > + SExpr *record() { return Rec; }
> > + const SExpr *record() const { return Rec; }
> >
> > const clang::ValueDecl *clangDecl() const { return Cvdecl; }
> >
> > @@ -1042,13 +953,13 @@ public:
> > }
> >
> > private:
> > - SExprRef Rec;
> > + SExpr* Rec;
> > StringRef SlotName;
> > const clang::ValueDecl *Cvdecl;
> > };
> >
> >
> > -// Call a function (after all arguments have been applied).
> > +/// Call a function (after all arguments have been applied).
> > class Call : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() == COP_Call;
> }
> > @@ -1057,8 +968,8 @@ public:
> > : SExpr(COP_Call), Target(T), Cexpr(Ce) {}
> > Call(const Call &C, SExpr *T) : SExpr(C), Target(T), Cexpr(C.Cexpr) {}
> >
> > - SExpr *target() { return Target.get(); }
> > - const SExpr *target() const { return Target.get(); }
> > + SExpr *target() { return Target; }
> > + const SExpr *target() const { return Target; }
> >
> > const clang::CallExpr *clangCallExpr() const { return Cexpr; }
> >
> > @@ -1074,12 +985,12 @@ public:
> > }
> >
> > private:
> > - SExprRef Target;
> > + SExpr* Target;
> > const clang::CallExpr *Cexpr;
> > };
> >
> >
> > -// Allocate memory for a new value on the heap or stack.
> > +/// Allocate memory for a new value on the heap or stack.
> > class Alloc : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() == COP_Call;
> }
> > @@ -1094,8 +1005,8 @@ public:
> >
> > AllocKind kind() const { return static_cast<AllocKind>(Flags); }
> >
> > - SExpr *dataType() { return Dtype.get(); }
> > - const SExpr *dataType() const { return Dtype.get(); }
> > + SExpr *dataType() { return Dtype; }
> > + const SExpr *dataType() const { return Dtype; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1112,11 +1023,11 @@ public:
> > }
> >
> > private:
> > - SExprRef Dtype;
> > + SExpr* Dtype;
> > };
> >
> >
> > -// Load a value from memory.
> > +/// Load a value from memory.
> > class Load : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() == COP_Load;
> }
> > @@ -1124,8 +1035,8 @@ public:
> > Load(SExpr *P) : SExpr(COP_Load), Ptr(P) {}
> > Load(const Load &L, SExpr *P) : SExpr(L), Ptr(P) {}
> >
> > - SExpr *pointer() { return Ptr.get(); }
> > - const SExpr *pointer() const { return Ptr.get(); }
> > + SExpr *pointer() { return Ptr; }
> > + const SExpr *pointer() const { return Ptr; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1139,12 +1050,12 @@ public:
> > }
> >
> > private:
> > - SExprRef Ptr;
> > + SExpr* Ptr;
> > };
> >
> >
> > -// Store a value to memory.
> > -// Source is a pointer, destination is the value to store.
> > +/// Store a value to memory.
> > +/// The destination is a pointer to a field, the source is the value to
> store.
> > class Store : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Store; }
> > @@ -1152,11 +1063,11 @@ public:
> > Store(SExpr *P, SExpr *V) : SExpr(COP_Store), Dest(P), Source(V) {}
> > Store(const Store &S, SExpr *P, SExpr *V) : SExpr(S), Dest(P),
> Source(V) {}
> >
> > - SExpr *destination() { return Dest.get(); } // Address to store to
> > - const SExpr *destination() const { return Dest.get(); }
> > + SExpr *destination() { return Dest; } // Address to store to
> > + const SExpr *destination() const { return Dest; }
> >
> > - SExpr *source() { return Source.get(); } // Value to store
> > - const SExpr *source() const { return Source.get(); }
> > + SExpr *source() { return Source; } // Value to store
> > + const SExpr *source() const { return Source; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1174,13 +1085,13 @@ public:
> > }
> >
> > private:
> > - SExprRef Dest;
> > - SExprRef Source;
> > + SExpr* Dest;
> > + SExpr* Source;
> > };
> >
> >
> > -// If p is a reference to an array, then first(p) is a reference to the
> first
> > -// element. The usual array notation p[i] becomes first(p + i).
> > +/// If p is a reference to an array, then p[i] is a reference to the
> i'th
> > +/// element of the array.
> > class ArrayIndex : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_ArrayIndex; }
> > @@ -1189,11 +1100,11 @@ public:
> > ArrayIndex(const ArrayIndex &E, SExpr *A, SExpr *N)
> > : SExpr(E), Array(A), Index(N) {}
> >
> > - SExpr *array() { return Array.get(); }
> > - const SExpr *array() const { return Array.get(); }
> > + SExpr *array() { return Array; }
> > + const SExpr *array() const { return Array; }
> >
> > - SExpr *index() { return Index.get(); }
> > - const SExpr *index() const { return Index.get(); }
> > + SExpr *index() { return Index; }
> > + const SExpr *index() const { return Index; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1211,14 +1122,14 @@ public:
> > }
> >
> > private:
> > - SExprRef Array;
> > - SExprRef Index;
> > + SExpr* Array;
> > + SExpr* Index;
> > };
> >
> >
> > -// Pointer arithmetic, restricted to arrays only.
> > -// If p is a reference to an array, then p + n, where n is an integer,
> is
> > -// a reference to a subarray.
> > +/// Pointer arithmetic, restricted to arrays only.
> > +/// If p is a reference to an array, then p + n, where n is an integer,
> is
> > +/// a reference to a subarray.
> > class ArrayAdd : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_ArrayAdd; }
> > @@ -1227,11 +1138,11 @@ public:
> > ArrayAdd(const ArrayAdd &E, SExpr *A, SExpr *N)
> > : SExpr(E), Array(A), Index(N) {}
> >
> > - SExpr *array() { return Array.get(); }
> > - const SExpr *array() const { return Array.get(); }
> > + SExpr *array() { return Array; }
> > + const SExpr *array() const { return Array; }
> >
> > - SExpr *index() { return Index.get(); }
> > - const SExpr *index() const { return Index.get(); }
> > + SExpr *index() { return Index; }
> > + const SExpr *index() const { return Index; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1249,12 +1160,13 @@ public:
> > }
> >
> > private:
> > - SExprRef Array;
> > - SExprRef Index;
> > + SExpr* Array;
> > + SExpr* Index;
> > };
> >
> >
> > -// Simple unary operation -- e.g. !, ~, etc.
> > +/// Simple arithmetic unary operations, e.g. negate and not.
> > +/// These operations have no side-effects.
> > class UnaryOp : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_UnaryOp; }
> > @@ -1268,8 +1180,8 @@ public:
> > return static_cast<TIL_UnaryOpcode>(Flags);
> > }
> >
> > - SExpr *expr() { return Expr0.get(); }
> > - const SExpr *expr() const { return Expr0.get(); }
> > + SExpr *expr() { return Expr0; }
> > + const SExpr *expr() const { return Expr0; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1287,11 +1199,12 @@ public:
> > }
> >
> > private:
> > - SExprRef Expr0;
> > + SExpr* Expr0;
> > };
> >
> >
> > -// Simple binary operation -- e.g. +, -, etc.
> > +/// Simple arithmetic binary operations, e.g. +, -, etc.
> > +/// These operations have no side effects.
> > class BinaryOp : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_BinaryOp; }
> > @@ -1309,11 +1222,11 @@ public:
> > return static_cast<TIL_BinaryOpcode>(Flags);
> > }
> >
> > - SExpr *expr0() { return Expr0.get(); }
> > - const SExpr *expr0() const { return Expr0.get(); }
> > + SExpr *expr0() { return Expr0; }
> > + const SExpr *expr0() const { return Expr0; }
> >
> > - SExpr *expr1() { return Expr1.get(); }
> > - const SExpr *expr1() const { return Expr1.get(); }
> > + SExpr *expr1() { return Expr1; }
> > + const SExpr *expr1() const { return Expr1; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1335,12 +1248,14 @@ public:
> > }
> >
> > private:
> > - SExprRef Expr0;
> > - SExprRef Expr1;
> > + SExpr* Expr0;
> > + SExpr* Expr1;
> > };
> >
> >
> > -// Cast expression
> > +/// Cast expressions.
> > +/// Cast expressions are essentially unary operations, but we treat them
> > +/// as a distinct AST node because they only change the type of the
> result.
> > class Cast : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() == COP_Cast;
> }
> > @@ -1352,8 +1267,8 @@ public:
> > return static_cast<TIL_CastOpcode>(Flags);
> > }
> >
> > - SExpr *expr() { return Expr0.get(); }
> > - const SExpr *expr() const { return Expr0.get(); }
> > + SExpr *expr() { return Expr0; }
> > + const SExpr *expr() const { return Expr0; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1371,16 +1286,18 @@ public:
> > }
> >
> > private:
> > - SExprRef Expr0;
> > + SExpr* Expr0;
> > };
> >
> >
> > class SCFG;
> >
> >
> > +/// Phi Node, for code in SSA form.
> > +/// Each Phi node has an array of possible values that it can take,
> > +/// depending on where control flow comes from.
> > class Phi : public SExpr {
> > public:
> > - // TODO: change to SExprRef
> > typedef SimpleArray<SExpr *> ValArray;
> >
> > // In minimal SSA form, all Phi nodes are MultiVal.
> > @@ -1394,9 +1311,12 @@ public:
> >
> > static bool classof(const SExpr *E) { return E->opcode() == COP_Phi; }
> >
> > - Phi() : SExpr(COP_Phi) {}
> > - Phi(MemRegionRef A, unsigned Nvals) : SExpr(COP_Phi), Values(A,
> Nvals) {}
> > - Phi(const Phi &P, ValArray &&Vs) : SExpr(P), Values(std::move(Vs))
> {}
> > + Phi()
> > + : SExpr(COP_Phi), Cvdecl(nullptr) {}
> > + Phi(MemRegionRef A, unsigned Nvals)
> > + : SExpr(COP_Phi), Values(A, Nvals), Cvdecl(nullptr) {}
> > + Phi(const Phi &P, ValArray &&Vs)
> > + : SExpr(P), Values(std::move(Vs)), Cvdecl(nullptr) {}
> >
> > const ValArray &values() const { return Values; }
> > ValArray &values() { return Values; }
> > @@ -1404,6 +1324,12 @@ public:
> > Status status() const { return static_cast<Status>(Flags); }
> > void setStatus(Status s) { Flags = s; }
> >
> > + /// Return the clang declaration of the variable for this Phi node,
> if any.
> > + const clang::ValueDecl *clangDecl() const { return Cvdecl; }
> > +
> > + /// Set the clang variable associated with this Phi node.
> > + void setClangDecl(const clang::ValueDecl *Cvd) { Cvdecl = Cvd; }
> > +
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > typename V::template Container<typename V::R_SExpr>
> > @@ -1423,65 +1349,260 @@ public:
> >
> > private:
> > ValArray Values;
> > + const clang::ValueDecl* Cvdecl;
> > +};
> > +
> > +
> > +/// Base class for basic block terminators: Branch, Goto, and Return.
> > +class Terminator : public SExpr {
> > +public:
> > + static bool classof(const SExpr *E) {
> > + return E->opcode() >= COP_Goto && E->opcode() <= COP_Return;
> > + }
> > +
> > +protected:
> > + Terminator(TIL_Opcode Op) : SExpr(Op) {}
> > + Terminator(const SExpr &E) : SExpr(E) {}
> > +
> > +public:
> > + /// Return the list of basic blocks that this terminator can branch
> to.
> > + ArrayRef<BasicBlock*> successors();
> > +
> > + ArrayRef<BasicBlock*> successors() const {
> > + return const_cast<const Terminator*>(this)->successors();
> > + }
> > +};
> > +
> > +
> > +/// Jump to another basic block.
> > +/// A goto instruction is essentially a tail-recursive call into another
> > +/// block. In addition to the block pointer, it specifies an index
> into the
> > +/// phi nodes of that block. The index can be used to retrieve the
> "arguments"
> > +/// of the call.
> > +class Goto : public Terminator {
> > +public:
> > + static bool classof(const SExpr *E) { return E->opcode() == COP_Goto;
> }
> > +
> > + Goto(BasicBlock *B, unsigned I)
> > + : Terminator(COP_Goto), TargetBlock(B), Index(I) {}
> > + Goto(const Goto &G, BasicBlock *B, unsigned I)
> > + : Terminator(COP_Goto), TargetBlock(B), Index(I) {}
> > +
> > + const BasicBlock *targetBlock() const { return TargetBlock; }
> > + BasicBlock *targetBlock() { return TargetBlock; }
> > +
> > + /// Returns the index into the
> > + unsigned index() const { return Index; }
> > +
> > + /// Return the list of basic blocks that this terminator can branch
> to.
> > + ArrayRef<BasicBlock*> successors() {
> > + return ArrayRef<BasicBlock*>(&TargetBlock, 1);
> > + }
> > +
> > + template <class V>
> > + typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > + BasicBlock *Ntb = Vs.reduceBasicBlockRef(TargetBlock);
> > + return Vs.reduceGoto(*this, Ntb);
> > + }
> > +
> > + template <class C>
> > + typename C::CType compare(const Goto *E, C &Cmp) const {
> > + // TODO: implement CFG comparisons
> > + return Cmp.comparePointers(this, E);
> > + }
> > +
> > +private:
> > + BasicBlock *TargetBlock;
> > + unsigned Index;
> > +};
> > +
> > +
> > +/// A conditional branch to two other blocks.
> > +/// Note that unlike Goto, Branch does not have an index. The target
> blocks
> > +/// must be child-blocks, and cannot have Phi nodes.
> > +class Branch : public Terminator {
> > +public:
> > + static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Branch; }
> > +
> > + Branch(SExpr *C, BasicBlock *T, BasicBlock *E)
> > + : Terminator(COP_Branch), Condition(C) {
> > + Branches[0] = T;
> > + Branches[1] = E;
> > + }
> > + Branch(const Branch &Br, SExpr *C, BasicBlock *T, BasicBlock *E)
> > + : Terminator(Br), Condition(C) {
> > + Branches[0] = T;
> > + Branches[1] = E;
> > + }
> > +
> > + const SExpr *condition() const { return Condition; }
> > + SExpr *condition() { return Condition; }
> > +
> > + const BasicBlock *thenBlock() const { return Branches[0]; }
> > + BasicBlock *thenBlock() { return Branches[0]; }
> > +
> > + const BasicBlock *elseBlock() const { return Branches[1]; }
> > + BasicBlock *elseBlock() { return Branches[1]; }
> > +
> > + /// Return the list of basic blocks that this terminator can branch
> to.
> > + ArrayRef<BasicBlock*> successors() {
> > + return ArrayRef<BasicBlock*>(Branches, 2);
> > + }
> > +
> > + template <class V>
> > + typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > + auto Nc = Vs.traverse(Condition, Vs.subExprCtx(Ctx));
> > + BasicBlock *Ntb = Vs.reduceBasicBlockRef(Branches[0]);
> > + BasicBlock *Nte = Vs.reduceBasicBlockRef(Branches[1]);
> > + return Vs.reduceBranch(*this, Nc, Ntb, Nte);
> > + }
> > +
> > + template <class C>
> > + typename C::CType compare(const Branch *E, C &Cmp) const {
> > + // TODO: implement CFG comparisons
> > + return Cmp.comparePointers(this, E);
> > + }
> > +
> > +private:
> > + SExpr* Condition;
> > + BasicBlock *Branches[2];
> > +};
> > +
> > +
> > +/// Return from the enclosing function, passing the return value to the
> caller.
> > +/// Only the exit block should end with a return statement.
> > +class Return : public Terminator {
> > +public:
> > + static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Return; }
> > +
> > + Return(SExpr* Rval) : Terminator(COP_Return), Retval(Rval) {}
> > + Return(const Return &R, SExpr* Rval) : Terminator(R), Retval(Rval) {}
> > +
> > + /// Return an empty list.
> > + ArrayRef<BasicBlock*> successors() {
> > + return ArrayRef<BasicBlock*>();
> > + }
> > +
> > + SExpr *returnValue() { return Retval; }
> > + const SExpr *returnValue() const { return Retval; }
> > +
> > + template <class V>
> > + typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > + auto Ne = Vs.traverse(Retval, Vs.subExprCtx(Ctx));
> > + return Vs.reduceReturn(*this, Ne);
> > + }
> > +
> > + template <class C>
> > + typename C::CType compare(const Return *E, C &Cmp) const {
> > + return Cmp.compare(Retval, E->Retval);
> > + }
> > +
> > +private:
> > + SExpr* Retval;
> > };
> >
> >
> > -// A basic block is part of an SCFG, and can be treated as a function in
> > -// continuation passing style. It consists of a sequence of phi nodes,
> which
> > -// are "arguments" to the function, followed by a sequence of
> instructions.
> > -// Both arguments and instructions define new variables. It ends with a
> > -// branch or goto to another basic block in the same SCFG.
> > +inline ArrayRef<BasicBlock*> Terminator::successors() {
> > + switch (opcode()) {
> > + case COP_Goto: return cast<Goto>(this)->successors();
> > + case COP_Branch: return cast<Branch>(this)->successors();
> > + case COP_Return: return cast<Return>(this)->successors();
> > + default:
> > + return ArrayRef<BasicBlock*>();
> > + }
> > +}
> > +
> > +
> > +/// A basic block is part of an SCFG. It can be treated as a function
> in
> > +/// continuation passing style. A block consists of a sequence of phi
> nodes,
> > +/// which are "arguments" to the function, followed by a sequence of
> > +/// instructions. It ends with a Terminator, which is a Branch or Goto
> to
> > +/// another basic block in the same SCFG.
> > class BasicBlock : public SExpr {
> > public:
> > - typedef SimpleArray<Variable*> VarArray;
> > + typedef SimpleArray<SExpr*> InstrArray;
> > typedef SimpleArray<BasicBlock*> BlockArray;
> >
> > + // TopologyNodes are used to overlay tree structures on top of the
> CFG,
> > + // such as dominator and postdominator trees. Each block is assigned
> an
> > + // ID in the tree according to a depth-first search. Tree traversals
> are
> > + // always up, towards the parents.
> > + struct TopologyNode {
> > + TopologyNode() : NodeID(0), SizeOfSubTree(0), Parent(nullptr) {}
> > +
> > + bool isParentOf(const TopologyNode& OtherNode) {
> > + return OtherNode.NodeID > NodeID &&
> > + OtherNode.NodeID < NodeID + SizeOfSubTree;
> > + }
> > +
> > + bool isParentOfOrEqual(const TopologyNode& OtherNode) {
> > + return OtherNode.NodeID >= NodeID &&
> > + OtherNode.NodeID < NodeID + SizeOfSubTree;
> > + }
> > +
> > + int NodeID;
> > + int SizeOfSubTree; // Includes this node, so must be > 1.
> > + BasicBlock *Parent; // Pointer to parent.
> > + };
> > +
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_BasicBlock; }
> >
> > - explicit BasicBlock(MemRegionRef A, BasicBlock* P = nullptr)
> > + explicit BasicBlock(MemRegionRef A)
> > : SExpr(COP_BasicBlock), Arena(A), CFGPtr(nullptr), BlockID(0),
> > - Parent(P), Terminator(nullptr)
> > - { }
> > - BasicBlock(BasicBlock &B, VarArray &&As, VarArray &&Is, SExpr *T)
> > - : SExpr(COP_BasicBlock), Arena(B.Arena), CFGPtr(nullptr),
> BlockID(0),
> > - Parent(nullptr), Args(std::move(As)), Instrs(std::move(Is)),
> > - Terminator(T)
> > - { }
> > -
> > - unsigned blockID() const { return BlockID; }
> > - unsigned numPredecessors() const { return Predecessors.size(); }
> > + Visited(0), TermInstr(nullptr) {}
> > + BasicBlock(BasicBlock &B, MemRegionRef A, InstrArray &&As, InstrArray
> &&Is,
> > + Terminator *T)
> > + : SExpr(COP_BasicBlock), Arena(A), CFGPtr(nullptr),
> BlockID(0),Visited(0),
> > + Args(std::move(As)), Instrs(std::move(Is)), TermInstr(T) {}
> > +
> > + /// Returns the block ID. Every block has a unique ID in the CFG.
> > + int blockID() const { return BlockID; }
> > +
> > + /// Returns the number of predecessors.
> > + size_t numPredecessors() const { return Predecessors.size(); }
> > + size_t numSuccessors() const { return successors().size(); }
> >
> > const SCFG* cfg() const { return CFGPtr; }
> > SCFG* cfg() { return CFGPtr; }
> >
> > - const BasicBlock *parent() const { return Parent; }
> > - BasicBlock *parent() { return Parent; }
> > + const BasicBlock *parent() const { return DominatorNode.Parent; }
> > + BasicBlock *parent() { return DominatorNode.Parent; }
> >
> > - const VarArray &arguments() const { return Args; }
> > - VarArray &arguments() { return Args; }
> > + const InstrArray &arguments() const { return Args; }
> > + InstrArray &arguments() { return Args; }
> >
> > - const VarArray &instructions() const { return Instrs; }
> > - VarArray &instructions() { return Instrs; }
> > + InstrArray &instructions() { return Instrs; }
> > + const InstrArray &instructions() const { return Instrs; }
> >
> > - const BlockArray &predecessors() const { return Predecessors; }
> > + /// Returns a list of predecessors.
> > + /// The order of predecessors in the list is important; each phi node
> has
> > + /// exactly one argument for each precessor, in the same order.
> > BlockArray &predecessors() { return Predecessors; }
> > + const BlockArray &predecessors() const { return Predecessors; }
> > +
> > + ArrayRef<BasicBlock*> successors() { return TermInstr->successors(); }
> > + ArrayRef<BasicBlock*> successors() const { return
> TermInstr->successors(); }
> >
> > - const SExpr *terminator() const { return Terminator.get(); }
> > - SExpr *terminator() { return Terminator.get(); }
> > + const Terminator *terminator() const { return TermInstr; }
> > + Terminator *terminator() { return TermInstr; }
> >
> > - void setBlockID(unsigned i) { BlockID = i; }
> > - void setParent(BasicBlock *P) { Parent = P; }
> > - void setTerminator(SExpr *E) { Terminator.reset(E); }
> > -
> > - // Add a new argument. V must define a phi-node.
> > - void addArgument(Variable *V) {
> > - V->setKind(Variable::VK_LetBB);
> > + void setTerminator(Terminator *E) { TermInstr = E; }
> > +
> > + bool Dominates(const BasicBlock &Other) {
> > + return DominatorNode.isParentOfOrEqual(Other.DominatorNode);
> > + }
> > +
> > + bool PostDominates(const BasicBlock &Other) {
> > + return PostDominatorNode.isParentOfOrEqual(Other.PostDominatorNode);
> > + }
> > +
> > + /// Add a new argument.
> > + void addArgument(Phi *V) {
> > Args.reserveCheck(1, Arena);
> > Args.push_back(V);
> > }
> > - // Add a new instruction.
> > - void addInstruction(Variable *V) {
> > - V->setKind(Variable::VK_LetBB);
> > + /// Add a new instruction.
> > + void addInstruction(SExpr *V) {
> > Instrs.reserveCheck(1, Arena);
> > Instrs.push_back(V);
> > }
> > @@ -1498,34 +1619,29 @@ public:
> > // Reserve space for NumPreds predecessors, including space in phi
> nodes.
> > void reservePredecessors(unsigned NumPreds);
> >
> > - // Return the index of BB, or Predecessors.size if BB is not a
> predecessor.
> > + /// Return the index of BB, or Predecessors.size if BB is not a
> predecessor.
> > unsigned findPredecessorIndex(const BasicBlock *BB) const {
> > auto I = std::find(Predecessors.cbegin(), Predecessors.cend(), BB);
> > return std::distance(Predecessors.cbegin(), I);
> > }
> >
> > - // Set id numbers for variables.
> > - void renumberVars();
> > -
> > template <class V>
> > typename V::R_BasicBlock traverse(V &Vs, typename V::R_Ctx Ctx) {
> > - typename V::template Container<Variable*> Nas(Vs, Args.size());
> > - typename V::template Container<Variable*> Nis(Vs, Instrs.size());
> > + typename V::template Container<SExpr*> Nas(Vs, Args.size());
> > + typename V::template Container<SExpr*> Nis(Vs, Instrs.size());
> >
> > // Entering the basic block should do any scope initialization.
> > Vs.enterBasicBlock(*this);
> >
> > - for (auto *A : Args) {
> > - auto Ne = Vs.traverse(A->Definition, Vs.subExprCtx(Ctx));
> > - Variable *Nvd = Vs.enterScope(*A, Ne);
> > - Nas.push_back(Nvd);
> > + for (auto *E : Args) {
> > + auto Ne = Vs.traverse(E, Vs.subExprCtx(Ctx));
> > + Nas.push_back(Ne);
> > }
> > - for (auto *I : Instrs) {
> > - auto Ne = Vs.traverse(I->Definition, Vs.subExprCtx(Ctx));
> > - Variable *Nvd = Vs.enterScope(*I, Ne);
> > - Nis.push_back(Nvd);
> > + for (auto *E : Instrs) {
> > + auto Ne = Vs.traverse(E, Vs.subExprCtx(Ctx));
> > + Nis.push_back(Ne);
> > }
> > - auto Nt = Vs.traverse(Terminator, Ctx);
> > + auto Nt = Vs.traverse(TermInstr, Ctx);
> >
> > // Exiting the basic block should handle any scope cleanup.
> > Vs.exitBasicBlock(*this);
> > @@ -1542,22 +1658,32 @@ public:
> > private:
> > friend class SCFG;
> >
> > - MemRegionRef Arena;
> > -
> > - SCFG *CFGPtr; // The CFG that contains this block.
> > - unsigned BlockID; // unique id for this BB in the containing
> CFG
> > - BasicBlock *Parent; // The parent block is the enclosing
> lexical scope.
> > - // The parent dominates this block.
> > - BlockArray Predecessors; // Predecessor blocks in the CFG.
> > - VarArray Args; // Phi nodes. One argument per predecessor.
> > - VarArray Instrs; // Instructions.
> > - SExprRef Terminator; // Branch or Goto
> > + int renumberInstrs(int id); // assign unique ids to all instructions
> > + int topologicalSort(SimpleArray<BasicBlock*>& Blocks, int ID);
> > + int topologicalFinalSort(SimpleArray<BasicBlock*>& Blocks, int ID);
> > + void computeDominator();
> > + void computePostDominator();
> > +
> > +private:
> > + MemRegionRef Arena; // The arena used to allocate this block.
> > + SCFG *CFGPtr; // The CFG that contains this block.
> > + int BlockID : 31; // unique id for this BB in the containing
> CFG.
> > + // IDs are in topological order.
> > + int Visited : 1; // Bit to determine if a block has been
> visited
> > + // during a traversal.
> > + BlockArray Predecessors; // Predecessor blocks in the CFG.
> > + InstrArray Args; // Phi nodes. One argument per
> predecessor.
> > + InstrArray Instrs; // Instructions.
> > + Terminator* TermInstr; // Terminating instruction
> > +
> > + TopologyNode DominatorNode; // The dominator tree
> > + TopologyNode PostDominatorNode; // The post-dominator tree
> > };
> >
> >
> > -// An SCFG is a control-flow graph. It consists of a set of basic
> blocks, each
> > -// of which terminates in a branch to another basic block. There is one
> > -// entry point, and one exit point.
> > +/// An SCFG is a control-flow graph. It consists of a set of basic
> blocks,
> > +/// each of which terminates in a branch to another basic block. There
> is one
> > +/// entry point, and one exit point.
> > class SCFG : public SExpr {
> > public:
> > typedef SimpleArray<BasicBlock *> BlockArray;
> > @@ -1568,20 +1694,29 @@ public:
> >
> > SCFG(MemRegionRef A, unsigned Nblocks)
> > : SExpr(COP_SCFG), Arena(A), Blocks(A, Nblocks),
> > - Entry(nullptr), Exit(nullptr) {
> > - Entry = new (A) BasicBlock(A, nullptr);
> > - Exit = new (A) BasicBlock(A, Entry);
> > - auto *V = new (A) Variable(new (A) Phi());
> > + Entry(nullptr), Exit(nullptr), NumInstructions(0), Normal(false) {
> > + Entry = new (A) BasicBlock(A);
> > + Exit = new (A) BasicBlock(A);
> > + auto *V = new (A) Phi();
> > Exit->addArgument(V);
> > + Exit->setTerminator(new (A) Return(V));
> > add(Entry);
> > add(Exit);
> > }
> > SCFG(const SCFG &Cfg, BlockArray &&Ba) // steals memory from Ba
> > : SExpr(COP_SCFG), Arena(Cfg.Arena), Blocks(std::move(Ba)),
> > - Entry(nullptr), Exit(nullptr) {
> > + Entry(nullptr), Exit(nullptr), NumInstructions(0),
> Normal(false) {
> > // TODO: set entry and exit!
> > }
> >
> > + /// Return true if this CFG is valid.
> > + bool valid() const { return Entry && Exit && Blocks.size() > 0; }
> > +
> > + /// Return true if this CFG has been normalized.
> > + /// After normalization, blocks are in topological order, and block
> and
> > + /// instruction IDs have been assigned.
> > + bool normal() const { return Normal; }
> > +
> > iterator begin() { return Blocks.begin(); }
> > iterator end() { return Blocks.end(); }
> >
> > @@ -1596,9 +1731,17 @@ public:
> > const BasicBlock *exit() const { return Exit; }
> > BasicBlock *exit() { return Exit; }
> >
> > + /// Return the number of blocks in the CFG.
> > + /// Block::blockID() will return a number less than numBlocks();
> > + size_t numBlocks() const { return Blocks.size(); }
> > +
> > + /// Return the total number of instructions in the CFG.
> > + /// This is useful for building instruction side-tables;
> > + /// A call to SExpr::id() will return a number less than
> numInstructions().
> > + unsigned numInstructions() { return NumInstructions; }
> > +
> > inline void add(BasicBlock *BB) {
> > - assert(BB->CFGPtr == nullptr || BB->CFGPtr == this);
> > - BB->setBlockID(Blocks.size());
> > + assert(BB->CFGPtr == nullptr);
> > BB->CFGPtr = this;
> > Blocks.reserveCheck(1, Arena);
> > Blocks.push_back(BB);
> > @@ -1607,13 +1750,13 @@ public:
> > void setEntry(BasicBlock *BB) { Entry = BB; }
> > void setExit(BasicBlock *BB) { Exit = BB; }
> >
> > - // Set varable ids in all blocks.
> > - void renumberVars();
> > + void computeNormalForm();
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > Vs.enterCFG(*this);
> > typename V::template Container<BasicBlock *> Bbs(Vs, Blocks.size());
> > +
> > for (auto *B : Blocks) {
> > Bbs.push_back( B->traverse(Vs, Vs.subExprCtx(Ctx)) );
> > }
> > @@ -1623,101 +1766,26 @@ public:
> >
> > template <class C>
> > typename C::CType compare(const SCFG *E, C &Cmp) const {
> > - // TODO -- implement CFG comparisons
> > + // TODO: implement CFG comparisons
> > return Cmp.comparePointers(this, E);
> > }
> >
> > private:
> > + void renumberInstrs(); // assign unique ids to all instructions
> > +
> > +private:
> > MemRegionRef Arena;
> > BlockArray Blocks;
> > BasicBlock *Entry;
> > BasicBlock *Exit;
> > + unsigned NumInstructions;
> > + bool Normal;
> > };
> >
> >
> > -class Goto : public SExpr {
> > -public:
> > - static bool classof(const SExpr *E) { return E->opcode() == COP_Goto;
> }
> > -
> > - Goto(BasicBlock *B, unsigned I)
> > - : SExpr(COP_Goto), TargetBlock(B), Index(I) {}
> > - Goto(const Goto &G, BasicBlock *B, unsigned I)
> > - : SExpr(COP_Goto), TargetBlock(B), Index(I) {}
> > -
> > - const BasicBlock *targetBlock() const { return TargetBlock; }
> > - BasicBlock *targetBlock() { return TargetBlock; }
> > -
> > - unsigned index() const { return Index; }
> > -
> > - template <class V>
> > - typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > - BasicBlock *Ntb = Vs.reduceBasicBlockRef(TargetBlock);
> > - return Vs.reduceGoto(*this, Ntb);
> > - }
> > -
> > - template <class C>
> > - typename C::CType compare(const Goto *E, C &Cmp) const {
> > - // TODO -- implement CFG comparisons
> > - return Cmp.comparePointers(this, E);
> > - }
> > -
> > -private:
> > - BasicBlock *TargetBlock;
> > - unsigned Index; // Index into Phi nodes of target block.
> > -};
> > -
> > -
> > -class Branch : public SExpr {
> > -public:
> > - static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Branch; }
> > -
> > - Branch(SExpr *C, BasicBlock *T, BasicBlock *E, unsigned TI, unsigned
> EI)
> > - : SExpr(COP_Branch), Condition(C), ThenBlock(T), ElseBlock(E),
> > - ThenIndex(TI), ElseIndex(EI)
> > - {}
> > - Branch(const Branch &Br, SExpr *C, BasicBlock *T, BasicBlock *E,
> > - unsigned TI, unsigned EI)
> > - : SExpr(COP_Branch), Condition(C), ThenBlock(T), ElseBlock(E),
> > - ThenIndex(TI), ElseIndex(EI)
> > - {}
> > -
> > - const SExpr *condition() const { return Condition; }
> > - SExpr *condition() { return Condition; }
> > -
> > - const BasicBlock *thenBlock() const { return ThenBlock; }
> > - BasicBlock *thenBlock() { return ThenBlock; }
> > -
> > - const BasicBlock *elseBlock() const { return ElseBlock; }
> > - BasicBlock *elseBlock() { return ElseBlock; }
> > -
> > - unsigned thenIndex() const { return ThenIndex; }
> > - unsigned elseIndex() const { return ElseIndex; }
> > -
> > - template <class V>
> > - typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > - auto Nc = Vs.traverse(Condition, Vs.subExprCtx(Ctx));
> > - BasicBlock *Ntb = Vs.reduceBasicBlockRef(ThenBlock);
> > - BasicBlock *Nte = Vs.reduceBasicBlockRef(ElseBlock);
> > - return Vs.reduceBranch(*this, Nc, Ntb, Nte);
> > - }
> >
> > - template <class C>
> > - typename C::CType compare(const Branch *E, C &Cmp) const {
> > - // TODO -- implement CFG comparisons
> > - return Cmp.comparePointers(this, E);
> > - }
> > -
> > -private:
> > - SExpr *Condition;
> > - BasicBlock *ThenBlock;
> > - BasicBlock *ElseBlock;
> > - unsigned ThenIndex;
> > - unsigned ElseIndex;
> > -};
> > -
> > -
> > -// An identifier, e.g. 'foo' or 'x'.
> > -// This is a pseduo-term; it will be lowered to a variable or
> projection.
> > +/// An identifier, e.g. 'foo' or 'x'.
> > +/// This is a pseduo-term; it will be lowered to a variable or
> projection.
> > class Identifier : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_Identifier; }
> > @@ -1742,8 +1810,8 @@ private:
> > };
> >
> >
> > -// An if-then-else expression.
> > -// This is a pseduo-term; it will be lowered to a branch in a CFG.
> > +/// An if-then-else expression.
> > +/// This is a pseduo-term; it will be lowered to a branch in a CFG.
> > class IfThenElse : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() ==
> COP_IfThenElse; }
> > @@ -1755,14 +1823,14 @@ public:
> > : SExpr(I), Condition(C), ThenExpr(T), ElseExpr(E)
> > { }
> >
> > - SExpr *condition() { return Condition.get(); } // Address to store
> to
> > - const SExpr *condition() const { return Condition.get(); }
> > + SExpr *condition() { return Condition; } // Address to store to
> > + const SExpr *condition() const { return Condition; }
> >
> > - SExpr *thenExpr() { return ThenExpr.get(); } // Value to store
> > - const SExpr *thenExpr() const { return ThenExpr.get(); }
> > + SExpr *thenExpr() { return ThenExpr; } // Value to store
> > + const SExpr *thenExpr() const { return ThenExpr; }
> >
> > - SExpr *elseExpr() { return ElseExpr.get(); } // Value to store
> > - const SExpr *elseExpr() const { return ElseExpr.get(); }
> > + SExpr *elseExpr() { return ElseExpr; } // Value to store
> > + const SExpr *elseExpr() const { return ElseExpr; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1784,14 +1852,14 @@ public:
> > }
> >
> > private:
> > - SExprRef Condition;
> > - SExprRef ThenExpr;
> > - SExprRef ElseExpr;
> > + SExpr* Condition;
> > + SExpr* ThenExpr;
> > + SExpr* ElseExpr;
> > };
> >
> >
> > -// A let-expression, e.g. let x=t; u.
> > -// This is a pseduo-term; it will be lowered to instructions in a CFG.
> > +/// A let-expression, e.g. let x=t; u.
> > +/// This is a pseduo-term; it will be lowered to instructions in a CFG.
> > class Let : public SExpr {
> > public:
> > static bool classof(const SExpr *E) { return E->opcode() == COP_Let; }
> > @@ -1806,8 +1874,8 @@ public:
> > Variable *variableDecl() { return VarDecl; }
> > const Variable *variableDecl() const { return VarDecl; }
> >
> > - SExpr *body() { return Body.get(); }
> > - const SExpr *body() const { return Body.get(); }
> > + SExpr *body() { return Body; }
> > + const SExpr *body() const { return Body; }
> >
> > template <class V>
> > typename V::R_SExpr traverse(V &Vs, typename V::R_Ctx Ctx) {
> > @@ -1834,14 +1902,14 @@ public:
> >
> > private:
> > Variable *VarDecl;
> > - SExprRef Body;
> > + SExpr* Body;
> > };
> >
> >
> >
> > const SExpr *getCanonicalVal(const SExpr *E);
> > SExpr* simplifyToCanonicalVal(SExpr *E);
> > -void simplifyIncompleteArg(Variable *V, til::Phi *Ph);
> > +void simplifyIncompleteArg(til::Phi *Ph);
> >
> >
> > } // end namespace til
> >
> > Modified:
> cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTraverse.h
> > URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTraverse.h?rev=217556&r1=217555&r2=217556&view=diff
> >
> ==============================================================================
> > --- cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTraverse.h
> (original)
> > +++ cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyTraverse.h Wed
> Sep 10 17:12:52 2014
> > @@ -58,11 +58,16 @@ public:
> > // Traverse an expression -- returning a result of type R_SExpr.
> > // Override this method to do something for every expression,
> regardless
> > // of which kind it is.
> > - typename R::R_SExpr traverse(SExprRef &E, typename R::R_Ctx Ctx) {
> > - return traverse(E.get(), Ctx);
> > + // E is a reference, so this can be use for in-place updates.
> > + // The type T must be a subclass of SExpr.
> > + template <class T>
> > + typename R::R_SExpr traverse(T* &E, typename R::R_Ctx Ctx) {
> > + return traverseSExpr(E, Ctx);
> > }
> >
> > - typename R::R_SExpr traverse(SExpr *E, typename R::R_Ctx Ctx) {
> > + // Override this method to do something for every expression.
> > + // Does not allow in-place updates.
> > + typename R::R_SExpr traverseSExpr(SExpr *E, typename R::R_Ctx Ctx) {
> > return traverseByCase(E, Ctx);
> > }
> >
> > @@ -75,6 +80,7 @@ public:
> > #include "ThreadSafetyOps.def"
> > #undef TIL_OPCODE_DEF
> > }
> > + return self()->reduceNull();
> > }
> >
> > // Traverse e, by static dispatch on the type "X" of e.
> > @@ -92,10 +98,10 @@ public:
> > class SimpleReducerBase {
> > public:
> > enum TraversalKind {
> > - TRV_Normal,
> > - TRV_Decl,
> > - TRV_Lazy,
> > - TRV_Type
> > + TRV_Normal, // ordinary subexpressions
> > + TRV_Decl, // declarations (e.g. function bodies)
> > + TRV_Lazy, // expressions that require lazy evaluation
> > + TRV_Type // type expressions
> > };
> >
> > // R_Ctx defines a "context" for the traversal, which encodes
> information
> > @@ -147,153 +153,6 @@ protected:
> > };
> >
> >
> > -// Implements a traversal that makes a deep copy of an SExpr.
> > -// The default behavior of reduce##X(...) is to create a copy of the
> original.
> > -// Subclasses can override reduce##X to implement non-destructive
> rewriting
> > -// passes.
> > -template<class Self>
> > -class CopyReducer : public Traversal<Self, CopyReducerBase>,
> > - public CopyReducerBase {
> > -public:
> > - CopyReducer(MemRegionRef A) : CopyReducerBase(A) {}
> > -
> > -public:
> > - R_SExpr reduceNull() {
> > - return nullptr;
> > - }
> > - // R_SExpr reduceFuture(...) is never used.
> > -
> > - R_SExpr reduceUndefined(Undefined &Orig) {
> > - return new (Arena) Undefined(Orig);
> > - }
> > - R_SExpr reduceWildcard(Wildcard &Orig) {
> > - return new (Arena) Wildcard(Orig);
> > - }
> > -
> > - R_SExpr reduceLiteral(Literal &Orig) {
> > - return new (Arena) Literal(Orig);
> > - }
> > - template<class T>
> > - R_SExpr reduceLiteralT(LiteralT<T> &Orig) {
> > - return new (Arena) LiteralT<T>(Orig);
> > - }
> > - R_SExpr reduceLiteralPtr(LiteralPtr &Orig) {
> > - return new (Arena) LiteralPtr(Orig);
> > - }
> > -
> > - R_SExpr reduceFunction(Function &Orig, Variable *Nvd, R_SExpr E0) {
> > - return new (Arena) Function(Orig, Nvd, E0);
> > - }
> > - R_SExpr reduceSFunction(SFunction &Orig, Variable *Nvd, R_SExpr E0) {
> > - return new (Arena) SFunction(Orig, Nvd, E0);
> > - }
> > - R_SExpr reduceCode(Code &Orig, R_SExpr E0, R_SExpr E1) {
> > - return new (Arena) Code(Orig, E0, E1);
> > - }
> > - R_SExpr reduceField(Field &Orig, R_SExpr E0, R_SExpr E1) {
> > - return new (Arena) Field(Orig, E0, E1);
> > - }
> > -
> > - R_SExpr reduceApply(Apply &Orig, R_SExpr E0, R_SExpr E1) {
> > - return new (Arena) Apply(Orig, E0, E1);
> > - }
> > - R_SExpr reduceSApply(SApply &Orig, R_SExpr E0, R_SExpr E1) {
> > - return new (Arena) SApply(Orig, E0, E1);
> > - }
> > - R_SExpr reduceProject(Project &Orig, R_SExpr E0) {
> > - return new (Arena) Project(Orig, E0);
> > - }
> > - R_SExpr reduceCall(Call &Orig, R_SExpr E0) {
> > - return new (Arena) Call(Orig, E0);
> > - }
> > -
> > - R_SExpr reduceAlloc(Alloc &Orig, R_SExpr E0) {
> > - return new (Arena) Alloc(Orig, E0);
> > - }
> > - R_SExpr reduceLoad(Load &Orig, R_SExpr E0) {
> > - return new (Arena) Load(Orig, E0);
> > - }
> > - R_SExpr reduceStore(Store &Orig, R_SExpr E0, R_SExpr E1) {
> > - return new (Arena) Store(Orig, E0, E1);
> > - }
> > - R_SExpr reduceArrayIndex(ArrayIndex &Orig, R_SExpr E0, R_SExpr E1) {
> > - return new (Arena) ArrayIndex(Orig, E0, E1);
> > - }
> > - R_SExpr reduceArrayAdd(ArrayAdd &Orig, R_SExpr E0, R_SExpr E1) {
> > - return new (Arena) ArrayAdd(Orig, E0, E1);
> > - }
> > - R_SExpr reduceUnaryOp(UnaryOp &Orig, R_SExpr E0) {
> > - return new (Arena) UnaryOp(Orig, E0);
> > - }
> > - R_SExpr reduceBinaryOp(BinaryOp &Orig, R_SExpr E0, R_SExpr E1) {
> > - return new (Arena) BinaryOp(Orig, E0, E1);
> > - }
> > - R_SExpr reduceCast(Cast &Orig, R_SExpr E0) {
> > - return new (Arena) Cast(Orig, E0);
> > - }
> > -
> > - R_SExpr reduceSCFG(SCFG &Orig, Container<BasicBlock *> &Bbs) {
> > - return nullptr; // FIXME: implement CFG rewriting
> > - }
> > - R_BasicBlock reduceBasicBlock(BasicBlock &Orig, Container<Variable *>
> &As,
> > - Container<Variable *> &Is, R_SExpr T) {
> > - return nullptr; // FIXME: implement CFG rewriting
> > - }
> > - R_SExpr reducePhi(Phi &Orig, Container<R_SExpr> &As) {
> > - return new (Arena) Phi(Orig, std::move(As.Elems));
> > - }
> > - R_SExpr reduceGoto(Goto &Orig, BasicBlock *B) {
> > - return new (Arena) Goto(Orig, B, 0); // FIXME: set index
> > - }
> > - R_SExpr reduceBranch(Branch &O, R_SExpr C, BasicBlock *B0, BasicBlock
> *B1) {
> > - return new (Arena) Branch(O, C, B0, B1, 0, 0); // FIXME: set
> indices
> > - }
> > -
> > - R_SExpr reduceIdentifier(Identifier &Orig) {
> > - return new (Arena) Identifier(Orig);
> > - }
> > - R_SExpr reduceIfThenElse(IfThenElse &Orig, R_SExpr C, R_SExpr T,
> R_SExpr E) {
> > - return new (Arena) IfThenElse(Orig, C, T, E);
> > - }
> > - R_SExpr reduceLet(Let &Orig, Variable *Nvd, R_SExpr B) {
> > - return new (Arena) Let(Orig, Nvd, B);
> > - }
> > -
> > - // Create a new variable from orig, and push it onto the lexical
> scope.
> > - Variable *enterScope(Variable &Orig, R_SExpr E0) {
> > - return new (Arena) Variable(Orig, E0);
> > - }
> > - // Exit the lexical scope of orig.
> > - void exitScope(const Variable &Orig) {}
> > -
> > - void enterCFG(SCFG &Cfg) {}
> > - void exitCFG(SCFG &Cfg) {}
> > - void enterBasicBlock(BasicBlock &BB) {}
> > - void exitBasicBlock(BasicBlock &BB) {}
> > -
> > - // Map Variable references to their rewritten definitions.
> > - Variable *reduceVariableRef(Variable *Ovd) { return Ovd; }
> > -
> > - // Map BasicBlock references to their rewritten definitions.
> > - BasicBlock *reduceBasicBlockRef(BasicBlock *Obb) { return Obb; }
> > -};
> > -
> > -
> > -class SExprCopier : public CopyReducer<SExprCopier> {
> > -public:
> > - typedef SExpr *R_SExpr;
> > -
> > - SExprCopier(MemRegionRef A) : CopyReducer(A) { }
> > -
> > - // Create a copy of e in region a.
> > - static SExpr *copy(SExpr *E, MemRegionRef A) {
> > - SExprCopier Copier(A);
> > - return Copier.traverse(E, TRV_Normal);
> > - }
> > -};
> > -
> > -
> > -
> > // Base class for visit traversals.
> > class VisitReducerBase : public SimpleReducerBase {
> > public:
> > @@ -368,8 +227,8 @@ public:
> > R_SExpr reduceSCFG(SCFG &Orig, Container<BasicBlock *> Bbs) {
> > return Bbs.Success;
> > }
> > - R_BasicBlock reduceBasicBlock(BasicBlock &Orig, Container<Variable *>
> &As,
> > - Container<Variable *> &Is, R_SExpr T) {
> > + R_BasicBlock reduceBasicBlock(BasicBlock &Orig, Container<R_SExpr>
> &As,
> > + Container<R_SExpr> &Is, R_SExpr T) {
> > return (As.Success && Is.Success && T);
> > }
> > R_SExpr reducePhi(Phi &Orig, Container<R_SExpr> &As) {
> > @@ -381,6 +240,9 @@ public:
> > R_SExpr reduceBranch(Branch &O, R_SExpr C, BasicBlock *B0, BasicBlock
> *B1) {
> > return C;
> > }
> > + R_SExpr reduceReturn(Return &O, R_SExpr E) {
> > + return E;
> > + }
> >
> > R_SExpr reduceIdentifier(Identifier &Orig) {
> > return true;
> > @@ -433,7 +295,7 @@ public:
> > #include "ThreadSafetyOps.def"
> > #undef TIL_OPCODE_DEF
> > }
> > - llvm_unreachable("invalid enum");
> > + return false;
> > }
> > };
> >
> > @@ -514,9 +376,9 @@ public:
> >
> >
> >
> > -inline std::ostream& operator<<(std::ostream& SS, llvm::StringRef R) {
> > - return SS.write(R.data(), R.size());
> > -}
> > +// inline std::ostream& operator<<(std::ostream& SS, StringRef R) {
> > +// return SS.write(R.data(), R.size());
> > +// }
>
> The commented code I mentioned.
>
> >
> > // Pretty printer for TIL expressions
> > template <typename Self, typename StreamType>
> > @@ -587,6 +449,7 @@ protected:
> > case COP_Phi: return Prec_Atom;
> > case COP_Goto: return Prec_Atom;
> > case COP_Branch: return Prec_Atom;
> > + case COP_Return: return Prec_Other;
> >
> > case COP_Identifier: return Prec_Atom;
> > case COP_IfThenElse: return Prec_Other;
> > @@ -595,22 +458,29 @@ protected:
> > return Prec_MAX;
> > }
> >
> > - void printBlockLabel(StreamType & SS, const BasicBlock *BB, unsigned
> index) {
> > + void printBlockLabel(StreamType & SS, const BasicBlock *BB, int
> index) {
> > if (!BB) {
> > SS << "BB_null";
> > return;
> > }
> > SS << "BB_";
> > SS << BB->blockID();
> > - SS << ":";
> > - SS << index;
> > + if (index >= 0) {
> > + SS << ":";
> > + SS << index;
> > + }
> > }
> >
> > - void printSExpr(const SExpr *E, StreamType &SS, unsigned P) {
> > +
> > + void printSExpr(const SExpr *E, StreamType &SS, unsigned P, bool
> Sub=true) {
> > if (!E) {
> > self()->printNull(SS);
> > return;
> > }
> > + if (Sub && E->block() && E->opcode() != COP_Variable) {
> > + SS << "_x" << E->id();
> > + return;
> > + }
> > if (self()->precedence(E) > P) {
> > // Wrap expr in () if necessary.
> > SS << "(";
> > @@ -740,20 +610,11 @@ protected:
> > SS << E->clangDecl()->getNameAsString();
> > }
> >
> > - void printVariable(const Variable *V, StreamType &SS, bool IsVarDecl
> = false) {
> > - if (!IsVarDecl && Cleanup) {
> > - const SExpr* E = getCanonicalVal(V);
> > - if (E != V) {
> > - printSExpr(E, SS, Prec_Atom);
> > - return;
> > - }
> > - }
> > - if (V->kind() == Variable::VK_LetBB)
> > - SS << V->name() << V->getBlockID() << "_" << V->getID();
> > - else if (CStyle && V->kind() == Variable::VK_SFun)
> > + void printVariable(const Variable *V, StreamType &SS, bool
> IsVarDecl=false) {
> > + if (CStyle && V->kind() == Variable::VK_SFun)
> > SS << "this";
> > else
> > - SS << V->name() << V->getID();
> > + SS << V->name() << V->id();
> > }
> >
> > void printFunction(const Function *E, StreamType &SS, unsigned
> sugared = 0) {
> > @@ -927,32 +788,38 @@ protected:
> > newline(SS);
> > }
> >
> > +
> > + void printBBInstr(const SExpr *E, StreamType &SS) {
> > + bool Sub = false;
> > + if (E->opcode() == COP_Variable) {
> > + auto *V = cast<Variable>(E);
> > + SS << "let " << V->name() << V->id() << " = ";
> > + E = V->definition();
> > + Sub = true;
> > + }
> > + else if (E->opcode() != COP_Store) {
> > + SS << "let _x" << E->id() << " = ";
> > + }
> > + self()->printSExpr(E, SS, Prec_MAX, Sub);
> > + SS << ";";
> > + newline(SS);
> > + }
> > +
> > void printBasicBlock(const BasicBlock *E, StreamType &SS) {
> > SS << "BB_" << E->blockID() << ":";
> > if (E->parent())
> > SS << " BB_" << E->parent()->blockID();
> > newline(SS);
> > - for (auto *A : E->arguments()) {
> > - SS << "let ";
> > - self()->printVariable(A, SS, true);
> > - SS << " = ";
> > - self()->printSExpr(A->definition(), SS, Prec_MAX);
> > - SS << ";";
> > - newline(SS);
> > - }
> > - for (auto *I : E->instructions()) {
> > - if (I->definition()->opcode() != COP_Store) {
> > - SS << "let ";
> > - self()->printVariable(I, SS, true);
> > - SS << " = ";
> > - }
> > - self()->printSExpr(I->definition(), SS, Prec_MAX);
> > - SS << ";";
> > - newline(SS);
> > - }
> > +
> > + for (auto *A : E->arguments())
> > + printBBInstr(A, SS);
> > +
> > + for (auto *I : E->instructions())
> > + printBBInstr(I, SS);
> > +
> > const SExpr *T = E->terminator();
> > if (T) {
> > - self()->printSExpr(T, SS, Prec_MAX);
> > + self()->printSExpr(T, SS, Prec_MAX, false);
> > SS << ";";
> > newline(SS);
> > }
> > @@ -983,9 +850,14 @@ protected:
> > SS << "branch (";
> > self()->printSExpr(E->condition(), SS, Prec_MAX);
> > SS << ") ";
> > - printBlockLabel(SS, E->thenBlock(), E->thenIndex());
> > + printBlockLabel(SS, E->thenBlock(), -1);
> > SS << " ";
> > - printBlockLabel(SS, E->elseBlock(), E->elseIndex());
> > + printBlockLabel(SS, E->elseBlock(), -1);
> > + }
> > +
> > + void printReturn(const Return *E, StreamType &SS) {
> > + SS << "return ";
> > + self()->printSExpr(E->returnValue(), SS, Prec_Other);
> > }
> >
> > void printIdentifier(const Identifier *E, StreamType &SS) {
> >
> > Modified: cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyUtil.h
> > URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyUtil.h?rev=217556&r1=217555&r2=217556&view=diff
> >
> ==============================================================================
> > --- cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyUtil.h
> (original)
> > +++ cfe/trunk/include/clang/Analysis/Analyses/ThreadSafetyUtil.h Wed Sep
> 10 17:12:52 2014
> > @@ -142,20 +142,35 @@ public:
> > assert(i < Size && "Array index out of bounds.");
> > return Data[i];
> > }
> > + T &back() {
> > + assert(Size && "No elements in the array.");
> > + return Data[Size - 1];
> > + }
> > + const T &back() const {
> > + assert(Size && "No elements in the array.");
> > + return Data[Size - 1];
> > + }
> >
> > iterator begin() { return Data; }
> > + iterator end() { return Data + Size; }
> > +
> > const_iterator begin() const { return Data; }
> > - iterator end() { return Data + Size; }
> > - const_iterator end() const { return Data + Size; }
> > + const_iterator end() const { return Data + Size; }
> >
> > const_iterator cbegin() const { return Data; }
> > - const_iterator cend() const { return Data + Size; }
> > + const_iterator cend() const { return Data + Size; }
> >
> > void push_back(const T &Elem) {
> > assert(Size < Capacity);
> > Data[Size++] = Elem;
> > }
> >
> > + // drop last n elements from array
> > + void drop(unsigned n = 0) {
> > + assert(Size > n);
> > + Size -= n;
> > + }
> > +
> > void setValues(unsigned Sz, const T& C) {
> > assert(Sz <= Capacity);
> > Size = Sz;
> > @@ -173,6 +188,37 @@ public:
> > return J - Osz;
> > }
> >
> > + // An adaptor to reverse a simple array
> > + class ReverseAdaptor {
> > + public:
> > + ReverseAdaptor(SimpleArray &Array) : Array(Array) {}
> > + // A reverse iterator used by the reverse adaptor
> > + class Iterator {
> > + public:
> > + Iterator(T *Data) : Data(Data) {}
> > + T &operator*() { return *Data; }
> > + const T &operator*() const { return *Data; }
> > + Iterator &operator++() {
> > + --Data;
> > + return *this;
> > + }
> > + bool operator!=(Iterator Other) { return Data != Other.Data; }
> > +
> > + private:
> > + T *Data;
> > + };
> > + Iterator begin() { return Array.end() - 1; }
> > + Iterator end() { return Array.begin() - 1; }
> > + const Iterator begin() const { return Array.end() - 1; }
> > + const Iterator end() const { return Array.begin() - 1; }
> > +
> > + private:
> > + SimpleArray &Array;
> > + };
> > +
> > + const ReverseAdaptor reverse() const { return ReverseAdaptor(*this); }
> > + ReverseAdaptor reverse() { return ReverseAdaptor(*this); }
> > +
> > private:
> > // std::max is annoying here, because it requires a reference,
> > // thus forcing InitialCapacity to be initialized outside the .h file.
> > @@ -187,6 +233,7 @@ private:
> > size_t Capacity;
> > };
> >
> > +
> > } // end namespace til
> >
> >
> > @@ -312,6 +359,12 @@ private:
> > };
> >
> >
> > +inline std::ostream& operator<<(std::ostream& ss, const StringRef str) {
> > + ss << str.data();
> > + return ss;
> > +}
> > +
> > +
> > } // end namespace threadSafety
> > } // end namespace clang
> >
> >
> > Modified: cfe/trunk/lib/Analysis/ThreadSafetyCommon.cpp
> > URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Analysis/ThreadSafetyCommon.cpp?rev=217556&r1=217555&r2=217556&view=diff
> >
> ==============================================================================
> > --- cfe/trunk/lib/Analysis/ThreadSafetyCommon.cpp (original)
> > +++ cfe/trunk/lib/Analysis/ThreadSafetyCommon.cpp Wed Sep 10 17:12:52
> 2014
> > @@ -63,11 +63,9 @@ std::string getSourceLiteralString(const
> > namespace til {
> >
> > // Return true if E is a variable that points to an incomplete Phi node.
> > -static bool isIncompleteVar(const SExpr *E) {
> > - if (const auto *V = dyn_cast<Variable>(E)) {
> > - if (const auto *Ph = dyn_cast<Phi>(V->definition()))
> > - return Ph->status() == Phi::PH_Incomplete;
> > - }
> > +static bool isIncompletePhi(const SExpr *E) {
> > + if (const auto *Ph = dyn_cast<Phi>(E))
> > + return Ph->status() == Phi::PH_Incomplete;
> > return false;
> > }
> >
> > @@ -320,6 +318,8 @@ til::SExpr *SExprBuilder::translateCXXTh
> > const ValueDecl *getValueDeclFromSExpr(const til::SExpr *E) {
> > if (auto *V = dyn_cast<til::Variable>(E))
> > return V->clangDecl();
> > + if (auto *Ph = dyn_cast<til::Phi>(E))
> > + return Ph->clangDecl();
> > if (auto *P = dyn_cast<til::Project>(E))
> > return P->clangDecl();
> > if (auto *L = dyn_cast<til::LiteralPtr>(E))
> > @@ -641,14 +641,14 @@ SExprBuilder::translateDeclStmt(const De
> > // If E is trivial returns E.
> > til::SExpr *SExprBuilder::addStatement(til::SExpr* E, const Stmt *S,
> > const ValueDecl *VD) {
> > - if (!E || !CurrentBB || til::ThreadSafetyTIL::isTrivial(E))
> > + if (!E || !CurrentBB || E->block() ||
> til::ThreadSafetyTIL::isTrivial(E))
> > return E;
> > -
> > - til::Variable *V = new (Arena) til::Variable(E, VD);
> > - CurrentInstructions.push_back(V);
> > + if (VD)
> > + E = new (Arena) til::Variable(E, VD);
> > + CurrentInstructions.push_back(E);
> > if (S)
> > - insertStmt(S, V);
> > - return V;
> > + insertStmt(S, E);
> > + return E;
> > }
> >
> >
> > @@ -705,11 +705,11 @@ void SExprBuilder::makePhiNodeVar(unsign
> > unsigned ArgIndex = CurrentBlockInfo->ProcessedPredecessors;
> > assert(ArgIndex > 0 && ArgIndex < NPreds);
> >
> > - til::Variable *V = dyn_cast<til::Variable>(CurrentLVarMap[i].second);
> > - if (V && V->getBlockID() == CurrentBB->blockID()) {
> > + til::SExpr *CurrE = CurrentLVarMap[i].second;
> > + if (CurrE->block() == CurrentBB) {
> > // We already have a Phi node in the current block,
> > // so just add the new variable to the Phi node.
> > - til::Phi *Ph = dyn_cast<til::Phi>(V->definition());
> > + til::Phi *Ph = dyn_cast<til::Phi>(CurrE);
> > assert(Ph && "Expecting Phi node.");
> > if (E)
> > Ph->values()[ArgIndex] = E;
> > @@ -718,27 +718,26 @@ void SExprBuilder::makePhiNodeVar(unsign
> >
> > // Make a new phi node: phi(..., E)
> > // All phi args up to the current index are set to the current value.
> > - til::SExpr *CurrE = CurrentLVarMap[i].second;
> > til::Phi *Ph = new (Arena) til::Phi(Arena, NPreds);
> > Ph->values().setValues(NPreds, nullptr);
> > for (unsigned PIdx = 0; PIdx < ArgIndex; ++PIdx)
> > Ph->values()[PIdx] = CurrE;
> > if (E)
> > Ph->values()[ArgIndex] = E;
> > + Ph->setClangDecl(CurrentLVarMap[i].first);
> > // If E is from a back-edge, or either E or CurrE are incomplete, then
> > // mark this node as incomplete; we may need to remove it later.
> > - if (!E || isIncompleteVar(E) || isIncompleteVar(CurrE)) {
> > + if (!E || isIncompletePhi(E) || isIncompletePhi(CurrE)) {
> > Ph->setStatus(til::Phi::PH_Incomplete);
> > }
> >
> > // Add Phi node to current block, and update CurrentLVarMap[i]
> > - auto *Var = new (Arena) til::Variable(Ph, CurrentLVarMap[i].first);
> > - CurrentArguments.push_back(Var);
> > + CurrentArguments.push_back(Ph);
> > if (Ph->status() == til::Phi::PH_Incomplete)
> > - IncompleteArgs.push_back(Var);
> > + IncompleteArgs.push_back(Ph);
> >
> > CurrentLVarMap.makeWritable();
> > - CurrentLVarMap.elem(i).second = Var;
> > + CurrentLVarMap.elem(i).second = Ph;
> > }
> >
> >
> > @@ -812,15 +811,13 @@ void SExprBuilder::mergePhiNodesBackEdge
> > unsigned ArgIndex = BBInfo[Blk->getBlockID()].ProcessedPredecessors;
> > assert(ArgIndex > 0 && ArgIndex < BB->numPredecessors());
> >
> > - for (til::Variable *V : BB->arguments()) {
> > - til::Phi *Ph = dyn_cast_or_null<til::Phi>(V->definition());
> > + for (til::SExpr *PE : BB->arguments()) {
> > + til::Phi *Ph = dyn_cast_or_null<til::Phi>(PE);
> > assert(Ph && "Expecting Phi Node.");
> > assert(Ph->values()[ArgIndex] == nullptr && "Wrong index for back
> edge.");
> > - assert(V->clangDecl() && "No local variable for Phi node.");
> >
> > - til::SExpr *E = lookupVarDecl(V->clangDecl());
> > + til::SExpr *E = lookupVarDecl(Ph->clangDecl());
> > assert(E && "Couldn't find local variable for Phi node.");
> > -
> > Ph->values()[ArgIndex] = E;
> > }
> > }
> > @@ -899,8 +896,8 @@ void SExprBuilder::enterCFGBlockBody(con
> > // Push those arguments onto the basic block.
> > CurrentBB->arguments().reserve(
> > static_cast<unsigned>(CurrentArguments.size()), Arena);
> > - for (auto *V : CurrentArguments)
> > - CurrentBB->addArgument(V);
> > + for (auto *A : CurrentArguments)
> > + CurrentBB->addArgument(A);
> > }
> >
> >
> > @@ -934,7 +931,7 @@ void SExprBuilder::exitCFGBlockBody(cons
> > til::BasicBlock *BB = *It ? lookupBlock(*It) : nullptr;
> > // TODO: set index
> > unsigned Idx = BB ? BB->findPredecessorIndex(CurrentBB) : 0;
> > - til::SExpr *Tm = new (Arena) til::Goto(BB, Idx);
> > + auto *Tm = new (Arena) til::Goto(BB, Idx);
> > CurrentBB->setTerminator(Tm);
> > }
> > else if (N == 2) {
> > @@ -942,9 +939,8 @@ void SExprBuilder::exitCFGBlockBody(cons
> > til::BasicBlock *BB1 = *It ? lookupBlock(*It) : nullptr;
> > ++It;
> > til::BasicBlock *BB2 = *It ? lookupBlock(*It) : nullptr;
> > - unsigned Idx1 = BB1 ? BB1->findPredecessorIndex(CurrentBB) : 0;
> > - unsigned Idx2 = BB2 ? BB2->findPredecessorIndex(CurrentBB) : 0;
> > - til::SExpr *Tm = new (Arena) til::Branch(C, BB1, BB2, Idx1, Idx2);
> > + // FIXME: make sure these arent' critical edges.
> > + auto *Tm = new (Arena) til::Branch(C, BB1, BB2);
> > CurrentBB->setTerminator(Tm);
> > }
> > }
> > @@ -971,10 +967,9 @@ void SExprBuilder::exitCFGBlock(const CF
> >
> >
> > void SExprBuilder::exitCFG(const CFGBlock *Last) {
> > - for (auto *V : IncompleteArgs) {
> > - til::Phi *Ph = dyn_cast<til::Phi>(V->definition());
> > - if (Ph && Ph->status() == til::Phi::PH_Incomplete)
> > - simplifyIncompleteArg(V, Ph);
> > + for (auto *Ph : IncompleteArgs) {
> > + if (Ph->status() == til::Phi::PH_Incomplete)
> > + simplifyIncompleteArg(Ph);
> > }
> >
> > CurrentArguments.clear();
> >
> > Modified: cfe/trunk/lib/Analysis/ThreadSafetyTIL.cpp
> > URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Analysis/ThreadSafetyTIL.cpp?rev=217556&r1=217555&r2=217556&view=diff
> >
> ==============================================================================
> > --- cfe/trunk/lib/Analysis/ThreadSafetyTIL.cpp (original)
> > +++ cfe/trunk/lib/Analysis/ThreadSafetyTIL.cpp Wed Sep 10 17:12:52 2014
> > @@ -48,12 +48,20 @@ StringRef getBinaryOpcodeString(TIL_Bina
> > }
> >
> >
> > +SExpr* Future::force() {
> > + Status = FS_evaluating;
> > + Result = compute();
> > + Status = FS_done;
> > + return Result;
> > +}
> > +
> > +
> > unsigned BasicBlock::addPredecessor(BasicBlock *Pred) {
> > unsigned Idx = Predecessors.size();
> > Predecessors.reserveCheck(1, Arena);
> > Predecessors.push_back(Pred);
> > - for (Variable *V : Args) {
> > - if (Phi* Ph = dyn_cast<Phi>(V->definition())) {
> > + for (SExpr *E : Args) {
> > + if (Phi* Ph = dyn_cast<Phi>(E)) {
> > Ph->values().reserveCheck(1, Arena);
> > Ph->values().push_back(nullptr);
> > }
> > @@ -61,105 +69,73 @@ unsigned BasicBlock::addPredecessor(Basi
> > return Idx;
> > }
> >
> > +
> > void BasicBlock::reservePredecessors(unsigned NumPreds) {
> > Predecessors.reserve(NumPreds, Arena);
> > - for (Variable *V : Args) {
> > - if (Phi* Ph = dyn_cast<Phi>(V->definition())) {
> > + for (SExpr *E : Args) {
> > + if (Phi* Ph = dyn_cast<Phi>(E)) {
> > Ph->values().reserve(NumPreds, Arena);
> > }
> > }
> > }
> >
> > -void BasicBlock::renumberVars() {
> > - unsigned VID = 0;
> > - for (Variable *V : Args) {
> > - V->setID(BlockID, VID++);
> > - }
> > - for (Variable *V : Instrs) {
> > - V->setID(BlockID, VID++);
> > - }
> > -}
> > -
> > -void SCFG::renumberVars() {
> > - for (BasicBlock *B : Blocks) {
> > - B->renumberVars();
> > - }
> > -}
> > -
> > -
> >
> > // If E is a variable, then trace back through any aliases or redundant
> > // Phi nodes to find the canonical definition.
> > const SExpr *getCanonicalVal(const SExpr *E) {
> > - while (auto *V = dyn_cast<Variable>(E)) {
> > - const SExpr *D;
> > - do {
> > - if (V->kind() != Variable::VK_Let)
> > - return V;
> > - D = V->definition();
> > - auto *V2 = dyn_cast<Variable>(D);
> > - if (V2)
> > - V = V2;
> > - else
> > - break;
> > - } while (true);
> > -
> > - if (ThreadSafetyTIL::isTrivial(D))
> > - return D;
> > -
> > - if (const Phi *Ph = dyn_cast<Phi>(D)) {
> > + while (true) {
> > + if (auto *V = dyn_cast<Variable>(E)) {
> > + if (V->kind() == Variable::VK_Let) {
> > + E = V->definition();
> > + continue;
> > + }
> > + }
> > + if (const Phi *Ph = dyn_cast<Phi>(E)) {
> > if (Ph->status() == Phi::PH_SingleVal) {
> > E = Ph->values()[0];
> > continue;
> > }
> > }
> > - return V;
> > + break;
> > }
> > return E;
> > }
> >
> >
> > -
> > // If E is a variable, then trace back through any aliases or redundant
> > // Phi nodes to find the canonical definition.
> > // The non-const version will simplify incomplete Phi nodes.
> > SExpr *simplifyToCanonicalVal(SExpr *E) {
> > - while (auto *V = dyn_cast<Variable>(E)) {
> > - SExpr *D;
> > - do {
> > + while (true) {
> > + if (auto *V = dyn_cast<Variable>(E)) {
> > if (V->kind() != Variable::VK_Let)
> > return V;
> > - D = V->definition();
> > - auto *V2 = dyn_cast<Variable>(D);
> > - if (V2)
> > - V = V2;
> > - else
> > - break;
> > - } while (true);
> > -
> > - if (ThreadSafetyTIL::isTrivial(D))
> > - return D;
> > -
> > - if (Phi *Ph = dyn_cast<Phi>(D)) {
> > + // Eliminate redundant variables, e.g. x = y, or x = 5,
> > + // but keep anything more complicated.
> > + if (til::ThreadSafetyTIL::isTrivial(V->definition())) {
> > + E = V->definition();
> > + continue;
> > + }
> > + return V;
> > + }
> > + if (auto *Ph = dyn_cast<Phi>(E)) {
> > if (Ph->status() == Phi::PH_Incomplete)
> > - simplifyIncompleteArg(V, Ph);
> > -
> > + simplifyIncompleteArg(Ph);
> > + // Eliminate redundant Phi nodes.
> > if (Ph->status() == Phi::PH_SingleVal) {
> > E = Ph->values()[0];
> > continue;
> > }
> > }
> > - return V;
> > + return E;
> > }
> > - return E;
> > }
> >
> >
> > -
> > // Trace the arguments of an incomplete Phi node to see if they have
> the same
> > // canonical definition. If so, mark the Phi node as redundant.
> > // getCanonicalVal() will recursively call simplifyIncompletePhi().
> > -void simplifyIncompleteArg(Variable *V, til::Phi *Ph) {
> > +void simplifyIncompleteArg(til::Phi *Ph) {
> > assert(Ph && Ph->status() == Phi::PH_Incomplete);
> >
> > // eliminate infinite recursion -- assume that this node is not
> redundant.
> > @@ -168,18 +144,200 @@ void simplifyIncompleteArg(Variable *V,
> > SExpr *E0 = simplifyToCanonicalVal(Ph->values()[0]);
> > for (unsigned i=1, n=Ph->values().size(); i<n; ++i) {
> > SExpr *Ei = simplifyToCanonicalVal(Ph->values()[i]);
> > - if (Ei == V)
> > + if (Ei == Ph)
> > continue; // Recursive reference to itself. Don't count.
> > if (Ei != E0) {
> > return; // Status is already set to MultiVal.
> > }
> > }
> > Ph->setStatus(Phi::PH_SingleVal);
> > - // Eliminate Redundant Phi node.
> > - V->setDefinition(Ph->values()[0]);
> > }
> >
> >
> > +// Renumbers the arguments and instructions to have unique, sequential
> IDs.
> > +int BasicBlock::renumberInstrs(int ID) {
> > + for (auto *Arg : Args)
> > + Arg->setID(this, ID++);
> > + for (auto *Instr : Instrs)
> > + Instr->setID(this, ID++);
> > + TermInstr->setID(this, ID++);
> > + return ID;
> > +}
> > +
> > +// Sorts the CFGs blocks using a reverse post-order depth-first
> traversal.
> > +// Each block will be written into the Blocks array in order, and its
> BlockID
> > +// will be set to the index in the array. Sorting should start from
> the entry
> > +// block, and ID should be the total number of blocks.
> > +int BasicBlock::topologicalSort(SimpleArray<BasicBlock*>& Blocks, int
> ID) {
> > + if (Visited) return ID;
> > + Visited = 1;
> > + for (auto *Block : successors())
> > + ID = Block->topologicalSort(Blocks, ID);
> > + // set ID and update block array in place.
> > + // We may lose pointers to unreachable blocks.
> > + assert(ID > 0);
> > + BlockID = --ID;
> > + Blocks[BlockID] = this;
> > + return ID;
> > +}
> > +
> > +// Performs a reverse topological traversal, starting from the exit
> block and
> > +// following back-edges. The dominator is serialized before any
> predecessors,
> > +// which guarantees that all blocks are serialized after their
> dominator and
> > +// before their post-dominator (because it's a reverse topological
> traversal).
> > +// ID should be initially set to 0.
> > +//
> > +// This sort assumes that (1) dominators have been computed, (2) there
> are no
> > +// critical edges, and (3) the entry block is reachable from the exit
> block
> > +// and no blocks are accessable via traversal of back-edges from the
> exit that
> > +// weren't accessable via forward edges from the entry.
> > +int BasicBlock::topologicalFinalSort(SimpleArray<BasicBlock*>& Blocks,
> int ID) {
> > + // Visited is assumed to have been set by the topologicalSort. This
> pass
> > + // assumes !Visited means that we've visited this node before.
> > + if (!Visited) return ID;
> > + Visited = 0;
> > + if (DominatorNode.Parent)
> > + ID = DominatorNode.Parent->topologicalFinalSort(Blocks, ID);
> > + for (auto *Pred : Predecessors)
> > + ID = Pred->topologicalFinalSort(Blocks, ID);
> > + assert(ID < Blocks.size());
> > + BlockID = ID++;
> > + Blocks[BlockID] = this;
> > + return ID;
> > +}
> > +
> > +// Computes the immediate dominator of the current block. Assumes that
> all of
> > +// its predecessors have already computed their dominators. This is
> achieved
> > +// by visiting the nodes in topological order.
> > +void BasicBlock::computeDominator() {
> > + BasicBlock *Candidate = nullptr;
> > + // Walk backwards from each predecessor to find the common dominator
> node.
> > + for (auto *Pred : Predecessors) {
> > + // Skip back-edges
> > + if (Pred->BlockID >= BlockID) continue;
> > + // If we don't yet have a candidate for dominator yet, take this
> one.
> > + if (Candidate == nullptr) {
> > + Candidate = Pred;
> > + continue;
> > + }
> > + // Walk the alternate and current candidate back to find a common
> ancestor.
> > + auto *Alternate = Pred;
> > + while (Alternate != Candidate) {
> > + if (Candidate->BlockID > Alternate->BlockID)
> > + Candidate = Candidate->DominatorNode.Parent;
> > + else
> > + Alternate = Alternate->DominatorNode.Parent;
> > + }
> > + }
> > + DominatorNode.Parent = Candidate;
> > + DominatorNode.SizeOfSubTree = 1;
> > +}
> > +
> > +// Computes the immediate post-dominator of the current block. Assumes
> that all
> > +// of its successors have already computed their post-dominators. This
> is
> > +// achieved visiting the nodes in reverse topological order.
> > +void BasicBlock::computePostDominator() {
> > + BasicBlock *Candidate = nullptr;
> > + // Walk back from each predecessor to find the common post-dominator
> node.
> > + for (auto *Succ : successors()) {
> > + // Skip back-edges
> > + if (Succ->BlockID <= BlockID) continue;
> > + // If we don't yet have a candidate for post-dominator yet, take
> this one.
> > + if (Candidate == nullptr) {
> > + Candidate = Succ;
> > + continue;
> > + }
> > + // Walk the alternate and current candidate back to find a common
> ancestor.
> > + auto *Alternate = Succ;
> > + while (Alternate != Candidate) {
> > + if (Candidate->BlockID < Alternate->BlockID)
> > + Candidate = Candidate->PostDominatorNode.Parent;
> > + else
> > + Alternate = Alternate->PostDominatorNode.Parent;
> > + }
> > + }
> > + PostDominatorNode.Parent = Candidate;
> > + PostDominatorNode.SizeOfSubTree = 1;
> > +}
> > +
> > +
> > +// Renumber instructions in all blocks
> > +void SCFG::renumberInstrs() {
> > + int InstrID = 0;
> > + for (auto *Block : Blocks)
> > + InstrID = Block->renumberInstrs(InstrID);
> > +}
> > +
> > +
> > +static inline void computeNodeSize(BasicBlock *B,
> > + BasicBlock::TopologyNode
> BasicBlock::*TN) {
> > + BasicBlock::TopologyNode *N = &(B->*TN);
> > + if (N->Parent) {
> > + BasicBlock::TopologyNode *P = &(N->Parent->*TN);
> > + // Initially set ID relative to the (as yet uncomputed) parent ID
> > + N->NodeID = P->SizeOfSubTree;
> > + P->SizeOfSubTree += N->SizeOfSubTree;
> > + }
> > +}
> > +
> > +static inline void computeNodeID(BasicBlock *B,
> > + BasicBlock::TopologyNode
> BasicBlock::*TN) {
> > + BasicBlock::TopologyNode *N = &(B->*TN);
> > + if (N->Parent) {
> > + BasicBlock::TopologyNode *P = &(N->Parent->*TN);
> > + N->NodeID += P->NodeID; // Fix NodeIDs relative to starting node.
> > + }
> > +}
> > +
> > +
> > +// Normalizes a CFG. Normalization has a few major components:
> > +// 1) Removing unreachable blocks.
> > +// 2) Computing dominators and post-dominators
> > +// 3) Topologically sorting the blocks into the "Blocks" array.
> > +void SCFG::computeNormalForm() {
> > + // Topologically sort the blocks starting from the entry block.
> > + int NumUnreachableBlocks = Entry->topologicalSort(Blocks,
> Blocks.size());
> > + if (NumUnreachableBlocks > 0) {
> > + // If there were unreachable blocks shift everything down, and
> delete them.
> > + for (size_t I = NumUnreachableBlocks, E = Blocks.size(); I < E;
> ++I) {
> > + size_t NI = I - NumUnreachableBlocks;
> > + Blocks[NI] = Blocks[I];
> > + Blocks[NI]->BlockID = NI;
> > + // FIXME: clean up predecessor pointers to unreachable blocks?
> > + }
> > + Blocks.drop(NumUnreachableBlocks);
> > + }
> > +
> > + // Compute dominators.
> > + for (auto *Block : Blocks)
> > + Block->computeDominator();
> > +
> > + // Once dominators have been computed, the final sort may be
> performed.
> > + int NumBlocks = Exit->topologicalFinalSort(Blocks, 0);
> > + assert(NumBlocks == Blocks.size());
> > + (void) NumBlocks;
> > +
> > + // Renumber the instructions now that we have a final sort.
> > + renumberInstrs();
> > +
> > + // Compute post-dominators and compute the sizes of each node in the
> > + // dominator tree.
> > + for (auto *Block : Blocks.reverse()) {
> > + Block->computePostDominator();
> > + computeNodeSize(Block, &BasicBlock::DominatorNode);
> > + }
> > + // Compute the sizes of each node in the post-dominator tree and
> assign IDs in
> > + // the dominator tree.
> > + for (auto *Block : Blocks) {
> > + computeNodeID(Block, &BasicBlock::DominatorNode);
> > + computeNodeSize(Block, &BasicBlock::PostDominatorNode);
> > + }
> > + // Assign IDs in the post-dominator tree.
> > + for (auto *Block : Blocks.reverse()) {
> > + computeNodeID(Block, &BasicBlock::PostDominatorNode);
> > + }
> > +}
> > +
> > } // end namespace til
> > } // end namespace threadSafety
> > } // end namespace clang
> >
> >
> > _______________________________________________
> > cfe-commits mailing list
> > cfe-commits at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
--
DeLesley Hutchins | Software Engineer | delesley at google.com | 505-206-0315
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140911/1aab84cf/attachment.html>
More information about the cfe-commits
mailing list