1 //===-- LLParser.h - Parser Class -------------------------------*- C++ -*-===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 //  This file defines the parser class for .ll files.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #ifndef LLVM_LIB_ASMPARSER_LLPARSER_H
15 #define LLVM_LIB_ASMPARSER_LLPARSER_H
16 
17 #include "LLLexer.h"
18 #include "llvm/ADT/DenseMap.h"
19 #include "llvm/ADT/StringMap.h"
20 #include "llvm/IR/Attributes.h"
21 #include "llvm/IR/Instructions.h"
22 #include "llvm/IR/Module.h"
23 #include "llvm/IR/Operator.h"
24 #include "llvm/IR/Type.h"
25 #include "llvm/IR/ValueHandle.h"
26 #include <map>
27 
28 namespace llvm {
29   class Module;
30   class OpaqueType;
31   class Function;
32   class Value;
33   class BasicBlock;
34   class Instruction;
35   class Constant;
36   class GlobalValue;
37   class Comdat;
38   class MDString;
39   class MDNode;
40   class StructType;
41 
42   /// ValID - Represents a reference of a definition of some sort with no type.
43   /// There are several cases where we have to parse the value but where the
44   /// type can depend on later context.  This may either be a numeric reference
45   /// or a symbolic (%var) reference.  This is just a discriminated union.
46   struct ValID {
47     enum {
48       t_LocalID, t_GlobalID,      // ID in UIntVal.
49       t_LocalName, t_GlobalName,  // Name in StrVal.
50       t_APSInt, t_APFloat,        // Value in APSIntVal/APFloatVal.
51       t_Null, t_Undef, t_Zero,    // No value.
52       t_EmptyArray,               // No value:  []
53       t_Constant,                 // Value in ConstantVal.
54       t_InlineAsm,                // Value in StrVal/StrVal2/UIntVal.
55       t_ConstantStruct,           // Value in ConstantStructElts.
56       t_PackedConstantStruct      // Value in ConstantStructElts.
57     } Kind;
58 
59     LLLexer::LocTy Loc;
60     unsigned UIntVal;
61     std::string StrVal, StrVal2;
62     APSInt APSIntVal;
63     APFloat APFloatVal;
64     Constant *ConstantVal;
65     Constant **ConstantStructElts;
66 
ValIDValID67     ValID() : Kind(t_LocalID), APFloatVal(0.0) {}
~ValIDValID68     ~ValID() {
69       if (Kind == t_ConstantStruct || Kind == t_PackedConstantStruct)
70         delete [] ConstantStructElts;
71     }
72 
73     bool operator<(const ValID &RHS) const {
74       if (Kind == t_LocalID || Kind == t_GlobalID)
75         return UIntVal < RHS.UIntVal;
76       assert((Kind == t_LocalName || Kind == t_GlobalName ||
77               Kind == t_ConstantStruct || Kind == t_PackedConstantStruct) &&
78              "Ordering not defined for this ValID kind yet");
79       return StrVal < RHS.StrVal;
80     }
81   };
82 
83   class LLParser {
84   public:
85     typedef LLLexer::LocTy LocTy;
86   private:
87     LLVMContext &Context;
88     LLLexer Lex;
89     Module *M;
90 
91     // Instruction metadata resolution.  Each instruction can have a list of
92     // MDRef info associated with them.
93     //
94     // The simpler approach of just creating temporary MDNodes and then calling
95     // RAUW on them when the definition is processed doesn't work because some
96     // instruction metadata kinds, such as dbg, get stored in the IR in an
97     // "optimized" format which doesn't participate in the normal value use
98     // lists. This means that RAUW doesn't work, even on temporary MDNodes
99     // which otherwise support RAUW. Instead, we defer resolving MDNode
100     // references until the definitions have been processed.
101     struct MDRef {
102       SMLoc Loc;
103       unsigned MDKind, MDSlot;
104     };
105 
106     SmallVector<Instruction*, 64> InstsWithTBAATag;
107 
108     // Type resolution handling data structures.  The location is set when we
109     // have processed a use of the type but not a definition yet.
110     StringMap<std::pair<Type*, LocTy> > NamedTypes;
111     std::map<unsigned, std::pair<Type*, LocTy> > NumberedTypes;
112 
113     std::map<unsigned, TrackingMDNodeRef> NumberedMetadata;
114     std::map<unsigned, std::pair<TempMDTuple, LocTy>> ForwardRefMDNodes;
115 
116     // Global Value reference information.
117     std::map<std::string, std::pair<GlobalValue*, LocTy> > ForwardRefVals;
118     std::map<unsigned, std::pair<GlobalValue*, LocTy> > ForwardRefValIDs;
119     std::vector<GlobalValue*> NumberedVals;
120 
121     // Comdat forward reference information.
122     std::map<std::string, LocTy> ForwardRefComdats;
123 
124     // References to blockaddress.  The key is the function ValID, the value is
125     // a list of references to blocks in that function.
126     std::map<ValID, std::map<ValID, GlobalValue *>> ForwardRefBlockAddresses;
127     class PerFunctionState;
128     /// Reference to per-function state to allow basic blocks to be
129     /// forward-referenced by blockaddress instructions within the same
130     /// function.
131     PerFunctionState *BlockAddressPFS;
132 
133     // Attribute builder reference information.
134     std::map<Value*, std::vector<unsigned> > ForwardRefAttrGroups;
135     std::map<unsigned, AttrBuilder> NumberedAttrBuilders;
136 
137   public:
LLParser(StringRef F,SourceMgr & SM,SMDiagnostic & Err,Module * m)138     LLParser(StringRef F, SourceMgr &SM, SMDiagnostic &Err, Module *m)
139         : Context(m->getContext()), Lex(F, SM, Err, m->getContext()), M(m),
140           BlockAddressPFS(nullptr) {}
141     bool Run();
142 
getContext()143     LLVMContext &getContext() { return Context; }
144 
145   private:
146 
Error(LocTy L,const Twine & Msg)147     bool Error(LocTy L, const Twine &Msg) const {
148       return Lex.Error(L, Msg);
149     }
TokError(const Twine & Msg)150     bool TokError(const Twine &Msg) const {
151       return Error(Lex.getLoc(), Msg);
152     }
153 
154     /// GetGlobalVal - Get a value with the specified name or ID, creating a
155     /// forward reference record if needed.  This can return null if the value
156     /// exists but does not have the right type.
157     GlobalValue *GetGlobalVal(const std::string &N, Type *Ty, LocTy Loc);
158     GlobalValue *GetGlobalVal(unsigned ID, Type *Ty, LocTy Loc);
159 
160     /// Get a Comdat with the specified name, creating a forward reference
161     /// record if needed.
162     Comdat *getComdat(const std::string &N, LocTy Loc);
163 
164     // Helper Routines.
165     bool ParseToken(lltok::Kind T, const char *ErrMsg);
EatIfPresent(lltok::Kind T)166     bool EatIfPresent(lltok::Kind T) {
167       if (Lex.getKind() != T) return false;
168       Lex.Lex();
169       return true;
170     }
171 
EatFastMathFlagsIfPresent()172     FastMathFlags EatFastMathFlagsIfPresent() {
173       FastMathFlags FMF;
174       while (true)
175         switch (Lex.getKind()) {
176         case lltok::kw_fast: FMF.setUnsafeAlgebra();   Lex.Lex(); continue;
177         case lltok::kw_nnan: FMF.setNoNaNs();          Lex.Lex(); continue;
178         case lltok::kw_ninf: FMF.setNoInfs();          Lex.Lex(); continue;
179         case lltok::kw_nsz:  FMF.setNoSignedZeros();   Lex.Lex(); continue;
180         case lltok::kw_arcp: FMF.setAllowReciprocal(); Lex.Lex(); continue;
181         default: return FMF;
182         }
183       return FMF;
184     }
185 
186     bool ParseOptionalToken(lltok::Kind T, bool &Present,
187                             LocTy *Loc = nullptr) {
188       if (Lex.getKind() != T) {
189         Present = false;
190       } else {
191         if (Loc)
192           *Loc = Lex.getLoc();
193         Lex.Lex();
194         Present = true;
195       }
196       return false;
197     }
198     bool ParseStringConstant(std::string &Result);
199     bool ParseUInt32(unsigned &Val);
ParseUInt32(unsigned & Val,LocTy & Loc)200     bool ParseUInt32(unsigned &Val, LocTy &Loc) {
201       Loc = Lex.getLoc();
202       return ParseUInt32(Val);
203     }
204     bool ParseUInt64(uint64_t &Val);
ParseUInt64(uint64_t & Val,LocTy & Loc)205     bool ParseUInt64(uint64_t &Val, LocTy &Loc) {
206       Loc = Lex.getLoc();
207       return ParseUInt64(Val);
208     }
209 
210     bool ParseTLSModel(GlobalVariable::ThreadLocalMode &TLM);
211     bool ParseOptionalThreadLocal(GlobalVariable::ThreadLocalMode &TLM);
parseOptionalUnnamedAddr(bool & UnnamedAddr)212     bool parseOptionalUnnamedAddr(bool &UnnamedAddr) {
213       return ParseOptionalToken(lltok::kw_unnamed_addr, UnnamedAddr);
214     }
215     bool ParseOptionalAddrSpace(unsigned &AddrSpace);
216     bool ParseOptionalParamAttrs(AttrBuilder &B);
217     bool ParseOptionalReturnAttrs(AttrBuilder &B);
218     bool ParseOptionalLinkage(unsigned &Linkage, bool &HasLinkage);
ParseOptionalLinkage(unsigned & Linkage)219     bool ParseOptionalLinkage(unsigned &Linkage) {
220       bool HasLinkage; return ParseOptionalLinkage(Linkage, HasLinkage);
221     }
222     bool ParseOptionalVisibility(unsigned &Visibility);
223     bool ParseOptionalDLLStorageClass(unsigned &DLLStorageClass);
224     bool ParseOptionalCallingConv(unsigned &CC);
225     bool ParseOptionalAlignment(unsigned &Alignment);
226     bool ParseOptionalDerefAttrBytes(lltok::Kind AttrKind, uint64_t &Bytes);
227     bool ParseScopeAndOrdering(bool isAtomic, SynchronizationScope &Scope,
228                                AtomicOrdering &Ordering);
229     bool ParseOrdering(AtomicOrdering &Ordering);
230     bool ParseOptionalStackAlignment(unsigned &Alignment);
231     bool ParseOptionalCommaAlign(unsigned &Alignment, bool &AteExtraComma);
232     bool ParseOptionalCommaInAlloca(bool &IsInAlloca);
233     bool ParseIndexList(SmallVectorImpl<unsigned> &Indices,bool &AteExtraComma);
ParseIndexList(SmallVectorImpl<unsigned> & Indices)234     bool ParseIndexList(SmallVectorImpl<unsigned> &Indices) {
235       bool AteExtraComma;
236       if (ParseIndexList(Indices, AteExtraComma)) return true;
237       if (AteExtraComma)
238         return TokError("expected index");
239       return false;
240     }
241 
242     // Top-Level Entities
243     bool ParseTopLevelEntities();
244     bool ValidateEndOfModule();
245     bool ParseTargetDefinition();
246     bool ParseModuleAsm();
247     bool ParseDepLibs();        // FIXME: Remove in 4.0.
248     bool ParseUnnamedType();
249     bool ParseNamedType();
250     bool ParseDeclare();
251     bool ParseDefine();
252 
253     bool ParseGlobalType(bool &IsConstant);
254     bool ParseUnnamedGlobal();
255     bool ParseNamedGlobal();
256     bool ParseGlobal(const std::string &Name, LocTy Loc, unsigned Linkage,
257                      bool HasLinkage, unsigned Visibility,
258                      unsigned DLLStorageClass,
259                      GlobalVariable::ThreadLocalMode TLM, bool UnnamedAddr);
260     bool ParseAlias(const std::string &Name, LocTy Loc, unsigned Linkage,
261                     unsigned Visibility, unsigned DLLStorageClass,
262                     GlobalVariable::ThreadLocalMode TLM, bool UnnamedAddr);
263     bool parseComdat();
264     bool ParseStandaloneMetadata();
265     bool ParseNamedMetadata();
266     bool ParseMDString(MDString *&Result);
267     bool ParseMDNodeID(MDNode *&Result);
268     bool ParseUnnamedAttrGrp();
269     bool ParseFnAttributeValuePairs(AttrBuilder &B,
270                                     std::vector<unsigned> &FwdRefAttrGrps,
271                                     bool inAttrGrp, LocTy &BuiltinLoc);
272 
273     // Type Parsing.
274     bool ParseType(Type *&Result, const Twine &Msg, bool AllowVoid = false);
275     bool ParseType(Type *&Result, bool AllowVoid = false) {
276       return ParseType(Result, "expected type", AllowVoid);
277     }
278     bool ParseType(Type *&Result, const Twine &Msg, LocTy &Loc,
279                    bool AllowVoid = false) {
280       Loc = Lex.getLoc();
281       return ParseType(Result, Msg, AllowVoid);
282     }
283     bool ParseType(Type *&Result, LocTy &Loc, bool AllowVoid = false) {
284       Loc = Lex.getLoc();
285       return ParseType(Result, AllowVoid);
286     }
287     bool ParseAnonStructType(Type *&Result, bool Packed);
288     bool ParseStructBody(SmallVectorImpl<Type*> &Body);
289     bool ParseStructDefinition(SMLoc TypeLoc, StringRef Name,
290                                std::pair<Type*, LocTy> &Entry,
291                                Type *&ResultTy);
292 
293     bool ParseArrayVectorType(Type *&Result, bool isVector);
294     bool ParseFunctionType(Type *&Result);
295 
296     // Function Semantic Analysis.
297     class PerFunctionState {
298       LLParser &P;
299       Function &F;
300       std::map<std::string, std::pair<Value*, LocTy> > ForwardRefVals;
301       std::map<unsigned, std::pair<Value*, LocTy> > ForwardRefValIDs;
302       std::vector<Value*> NumberedVals;
303 
304       /// FunctionNumber - If this is an unnamed function, this is the slot
305       /// number of it, otherwise it is -1.
306       int FunctionNumber;
307     public:
308       PerFunctionState(LLParser &p, Function &f, int FunctionNumber);
309       ~PerFunctionState();
310 
getFunction()311       Function &getFunction() const { return F; }
312 
313       bool FinishFunction();
314 
315       /// GetVal - Get a value with the specified name or ID, creating a
316       /// forward reference record if needed.  This can return null if the value
317       /// exists but does not have the right type.
318       Value *GetVal(const std::string &Name, Type *Ty, LocTy Loc);
319       Value *GetVal(unsigned ID, Type *Ty, LocTy Loc);
320 
321       /// SetInstName - After an instruction is parsed and inserted into its
322       /// basic block, this installs its name.
323       bool SetInstName(int NameID, const std::string &NameStr, LocTy NameLoc,
324                        Instruction *Inst);
325 
326       /// GetBB - Get a basic block with the specified name or ID, creating a
327       /// forward reference record if needed.  This can return null if the value
328       /// is not a BasicBlock.
329       BasicBlock *GetBB(const std::string &Name, LocTy Loc);
330       BasicBlock *GetBB(unsigned ID, LocTy Loc);
331 
332       /// DefineBB - Define the specified basic block, which is either named or
333       /// unnamed.  If there is an error, this returns null otherwise it returns
334       /// the block being defined.
335       BasicBlock *DefineBB(const std::string &Name, LocTy Loc);
336 
337       bool resolveForwardRefBlockAddresses();
338     };
339 
340     bool ConvertValIDToValue(Type *Ty, ValID &ID, Value *&V,
341                              PerFunctionState *PFS);
342 
343     bool ParseValue(Type *Ty, Value *&V, PerFunctionState *PFS);
ParseValue(Type * Ty,Value * & V,PerFunctionState & PFS)344     bool ParseValue(Type *Ty, Value *&V, PerFunctionState &PFS) {
345       return ParseValue(Ty, V, &PFS);
346     }
ParseValue(Type * Ty,Value * & V,LocTy & Loc,PerFunctionState & PFS)347     bool ParseValue(Type *Ty, Value *&V, LocTy &Loc,
348                     PerFunctionState &PFS) {
349       Loc = Lex.getLoc();
350       return ParseValue(Ty, V, &PFS);
351     }
352 
353     bool ParseTypeAndValue(Value *&V, PerFunctionState *PFS);
ParseTypeAndValue(Value * & V,PerFunctionState & PFS)354     bool ParseTypeAndValue(Value *&V, PerFunctionState &PFS) {
355       return ParseTypeAndValue(V, &PFS);
356     }
ParseTypeAndValue(Value * & V,LocTy & Loc,PerFunctionState & PFS)357     bool ParseTypeAndValue(Value *&V, LocTy &Loc, PerFunctionState &PFS) {
358       Loc = Lex.getLoc();
359       return ParseTypeAndValue(V, PFS);
360     }
361     bool ParseTypeAndBasicBlock(BasicBlock *&BB, LocTy &Loc,
362                                 PerFunctionState &PFS);
ParseTypeAndBasicBlock(BasicBlock * & BB,PerFunctionState & PFS)363     bool ParseTypeAndBasicBlock(BasicBlock *&BB, PerFunctionState &PFS) {
364       LocTy Loc;
365       return ParseTypeAndBasicBlock(BB, Loc, PFS);
366     }
367 
368 
369     struct ParamInfo {
370       LocTy Loc;
371       Value *V;
372       AttributeSet Attrs;
ParamInfoParamInfo373       ParamInfo(LocTy loc, Value *v, AttributeSet attrs)
374         : Loc(loc), V(v), Attrs(attrs) {}
375     };
376     bool ParseParameterList(SmallVectorImpl<ParamInfo> &ArgList,
377                             PerFunctionState &PFS,
378                             bool IsMustTailCall = false,
379                             bool InVarArgsFunc = false);
380 
381     // Constant Parsing.
382     bool ParseValID(ValID &ID, PerFunctionState *PFS = nullptr);
383     bool ParseGlobalValue(Type *Ty, Constant *&V);
384     bool ParseGlobalTypeAndValue(Constant *&V);
385     bool ParseGlobalValueVector(SmallVectorImpl<Constant *> &Elts);
386     bool parseOptionalComdat(StringRef GlobalName, Comdat *&C);
387     bool ParseMetadataAsValue(Value *&V, PerFunctionState &PFS);
388     bool ParseValueAsMetadata(Metadata *&MD, const Twine &TypeMsg,
389                               PerFunctionState *PFS);
390     bool ParseMetadata(Metadata *&MD, PerFunctionState *PFS);
391     bool ParseMDTuple(MDNode *&MD, bool IsDistinct = false);
392     bool ParseMDNode(MDNode *&MD);
393     bool ParseMDNodeTail(MDNode *&MD);
394     bool ParseMDNodeVector(SmallVectorImpl<Metadata *> &MDs);
395     bool ParseInstructionMetadata(Instruction *Inst, PerFunctionState *PFS);
396 
397     template <class FieldTy>
398     bool ParseMDField(LocTy Loc, StringRef Name, FieldTy &Result);
399     template <class FieldTy> bool ParseMDField(StringRef Name, FieldTy &Result);
400     template <class ParserTy>
401     bool ParseMDFieldsImplBody(ParserTy parseField);
402     template <class ParserTy>
403     bool ParseMDFieldsImpl(ParserTy parseField, LocTy &ClosingLoc);
404     bool ParseSpecializedMDNode(MDNode *&N, bool IsDistinct = false);
405 
406 #define HANDLE_SPECIALIZED_MDNODE_LEAF(CLASS)                                  \
407   bool Parse##CLASS(MDNode *&Result, bool IsDistinct);
408 #include "llvm/IR/Metadata.def"
409 
410     // Function Parsing.
411     struct ArgInfo {
412       LocTy Loc;
413       Type *Ty;
414       AttributeSet Attrs;
415       std::string Name;
ArgInfoArgInfo416       ArgInfo(LocTy L, Type *ty, AttributeSet Attr, const std::string &N)
417         : Loc(L), Ty(ty), Attrs(Attr), Name(N) {}
418     };
419     bool ParseArgumentList(SmallVectorImpl<ArgInfo> &ArgList, bool &isVarArg);
420     bool ParseFunctionHeader(Function *&Fn, bool isDefine);
421     bool ParseFunctionBody(Function &Fn);
422     bool ParseBasicBlock(PerFunctionState &PFS);
423 
424     enum TailCallType { TCT_None, TCT_Tail, TCT_MustTail };
425 
426     // Instruction Parsing.  Each instruction parsing routine can return with a
427     // normal result, an error result, or return having eaten an extra comma.
428     enum InstResult { InstNormal = 0, InstError = 1, InstExtraComma = 2 };
429     int ParseInstruction(Instruction *&Inst, BasicBlock *BB,
430                          PerFunctionState &PFS);
431     bool ParseCmpPredicate(unsigned &Pred, unsigned Opc);
432 
433     bool ParseRet(Instruction *&Inst, BasicBlock *BB, PerFunctionState &PFS);
434     bool ParseBr(Instruction *&Inst, PerFunctionState &PFS);
435     bool ParseSwitch(Instruction *&Inst, PerFunctionState &PFS);
436     bool ParseIndirectBr(Instruction *&Inst, PerFunctionState &PFS);
437     bool ParseInvoke(Instruction *&Inst, PerFunctionState &PFS);
438     bool ParseResume(Instruction *&Inst, PerFunctionState &PFS);
439 
440     bool ParseArithmetic(Instruction *&I, PerFunctionState &PFS, unsigned Opc,
441                          unsigned OperandType);
442     bool ParseLogical(Instruction *&I, PerFunctionState &PFS, unsigned Opc);
443     bool ParseCompare(Instruction *&I, PerFunctionState &PFS, unsigned Opc);
444     bool ParseCast(Instruction *&I, PerFunctionState &PFS, unsigned Opc);
445     bool ParseSelect(Instruction *&I, PerFunctionState &PFS);
446     bool ParseVA_Arg(Instruction *&I, PerFunctionState &PFS);
447     bool ParseExtractElement(Instruction *&I, PerFunctionState &PFS);
448     bool ParseInsertElement(Instruction *&I, PerFunctionState &PFS);
449     bool ParseShuffleVector(Instruction *&I, PerFunctionState &PFS);
450     int ParsePHI(Instruction *&I, PerFunctionState &PFS);
451     bool ParseLandingPad(Instruction *&I, PerFunctionState &PFS);
452     bool ParseCall(Instruction *&I, PerFunctionState &PFS,
453                    CallInst::TailCallKind IsTail);
454     int ParseAlloc(Instruction *&I, PerFunctionState &PFS);
455     int ParseLoad(Instruction *&I, PerFunctionState &PFS);
456     int ParseStore(Instruction *&I, PerFunctionState &PFS);
457     int ParseCmpXchg(Instruction *&I, PerFunctionState &PFS);
458     int ParseAtomicRMW(Instruction *&I, PerFunctionState &PFS);
459     int ParseFence(Instruction *&I, PerFunctionState &PFS);
460     int ParseGetElementPtr(Instruction *&I, PerFunctionState &PFS);
461     int ParseExtractValue(Instruction *&I, PerFunctionState &PFS);
462     int ParseInsertValue(Instruction *&I, PerFunctionState &PFS);
463 
464     // Use-list order directives.
465     bool ParseUseListOrder(PerFunctionState *PFS = nullptr);
466     bool ParseUseListOrderBB();
467     bool ParseUseListOrderIndexes(SmallVectorImpl<unsigned> &Indexes);
468     bool sortUseListOrder(Value *V, ArrayRef<unsigned> Indexes, SMLoc Loc);
469   };
470 } // End llvm namespace
471 
472 #endif
473