1 //===-- llvmAsmParser.y - Parser for llvm assembly files ---------*- C++ -*--=//
3 // This file implements the bison parser for LLVM assembly languages files.
5 //===------------------------------------------------------------------------=//
8 // TODO: Parse comments and add them to an internal node... so that they may
9 // be saved in the bytecode format as well as everything else. Very important
10 // for a general IR format.
14 #include "ParserInternals.h"
15 #include "llvm/Assembly/Parser.h"
16 #include "llvm/SymbolTable.h"
17 #include "llvm/Module.h"
18 #include "llvm/GlobalVariable.h"
19 #include "llvm/Method.h"
20 #include "llvm/BasicBlock.h"
21 #include "llvm/DerivedTypes.h"
22 #include "llvm/iTerminators.h"
23 #include "llvm/iMemory.h"
24 #include "llvm/Support/STLExtras.h"
25 #include "llvm/Support/DepthFirstIterator.h"
27 #include <utility> // Get definition of pair class
29 #include <stdio.h> // This embarasment is due to our flex lexer...
31 int yyerror(const char *ErrorMsg); // Forward declarations to prevent "implicit
32 int yylex(); // declaration" of xxx warnings.
35 static Module *ParserResult;
38 // DEBUG_UPREFS - Define this symbol if you want to enable debugging output
39 // relating to upreferences in the input stream.
41 //#define DEBUG_UPREFS 1
43 #define UR_OUT(X) cerr << X
48 // This contains info used when building the body of a method. It is destroyed
49 // when the method is completed.
51 typedef vector<Value *> ValueList; // Numbered defs
52 static void ResolveDefinitions(vector<ValueList> &LateResolvers);
53 static void ResolveTypes (vector<PATypeHolder<Type> > &LateResolveTypes);
55 static struct PerModuleInfo {
56 Module *CurrentModule;
57 vector<ValueList> Values; // Module level numbered definitions
58 vector<ValueList> LateResolveValues;
59 vector<PATypeHolder<Type> > Types, LateResolveTypes;
62 // If we could not resolve some methods at method compilation time (calls to
63 // methods before they are defined), resolve them now... Types are resolved
64 // when the constant pool has been completely parsed.
66 ResolveDefinitions(LateResolveValues);
68 Values.clear(); // Clear out method local definitions
74 static struct PerMethodInfo {
75 Method *CurrentMethod; // Pointer to current method being created
77 vector<ValueList> Values; // Keep track of numbered definitions
78 vector<ValueList> LateResolveValues;
79 vector<PATypeHolder<Type> > Types, LateResolveTypes;
80 bool isDeclare; // Is this method a forward declararation?
82 inline PerMethodInfo() {
87 inline ~PerMethodInfo() {}
89 inline void MethodStart(Method *M) {
94 // If we could not resolve some blocks at parsing time (forward branches)
95 // resolve the branches now...
96 ResolveDefinitions(LateResolveValues);
98 Values.clear(); // Clear out method local definitions
103 } CurMeth; // Info for the current method...
106 //===----------------------------------------------------------------------===//
107 // Code to handle definitions of all the types
108 //===----------------------------------------------------------------------===//
110 static void InsertValue(Value *D, vector<ValueList> &ValueTab = CurMeth.Values){
111 if (!D->hasName()) { // Is this a numbered definition?
112 unsigned type = D->getType()->getUniqueID();
113 if (ValueTab.size() <= type)
114 ValueTab.resize(type+1, ValueList());
115 //printf("Values[%d][%d] = %d\n", type, ValueTab[type].size(), D);
116 ValueTab[type].push_back(D);
120 // TODO: FIXME when Type are not const
121 static void InsertType(const Type *Ty, vector<PATypeHolder<Type> > &Types) {
125 static const Type *getTypeVal(const ValID &D, bool DoNotImprovise = false) {
127 case 0: { // Is it a numbered definition?
128 unsigned Num = (unsigned)D.Num;
130 // Module constants occupy the lowest numbered slots...
131 if (Num < CurModule.Types.size())
132 return CurModule.Types[Num];
134 Num -= CurModule.Types.size();
136 // Check that the number is within bounds...
137 if (Num <= CurMeth.Types.size())
138 return CurMeth.Types[Num];
140 case 1: { // Is it a named definition?
142 SymbolTable *SymTab = 0;
143 if (CurMeth.CurrentMethod)
144 SymTab = CurMeth.CurrentMethod->getSymbolTable();
145 Value *N = SymTab ? SymTab->lookup(Type::TypeTy, Name) : 0;
148 // Symbol table doesn't automatically chain yet... because the method
149 // hasn't been added to the module...
151 SymTab = CurModule.CurrentModule->getSymbolTable();
153 N = SymTab->lookup(Type::TypeTy, Name);
157 D.destroy(); // Free old strdup'd memory...
158 return cast<const Type>(N);
161 ThrowException("Invalid symbol type reference!");
164 // If we reached here, we referenced either a symbol that we don't know about
165 // or an id number that hasn't been read yet. We may be referencing something
166 // forward, so just create an entry to be resolved later and get to it...
168 if (DoNotImprovise) return 0; // Do we just want a null to be returned?
170 vector<PATypeHolder<Type> > *LateResolver = CurMeth.CurrentMethod ?
171 &CurMeth.LateResolveTypes : &CurModule.LateResolveTypes;
173 Type *Typ = new TypePlaceHolder(Type::TypeTy, D);
174 InsertType(Typ, *LateResolver);
178 static Value *lookupInSymbolTable(const Type *Ty, const string &Name) {
179 SymbolTable *SymTab =
180 CurMeth.CurrentMethod ? CurMeth.CurrentMethod->getSymbolTable() : 0;
181 Value *N = SymTab ? SymTab->lookup(Ty, Name) : 0;
184 // Symbol table doesn't automatically chain yet... because the method
185 // hasn't been added to the module...
187 SymTab = CurModule.CurrentModule->getSymbolTable();
189 N = SymTab->lookup(Ty, Name);
195 static Value *getVal(const Type *Ty, const ValID &D,
196 bool DoNotImprovise = false) {
197 assert(Ty != Type::TypeTy && "Should use getTypeVal for types!");
200 case ValID::NumberVal: { // Is it a numbered definition?
201 unsigned type = Ty->getUniqueID();
202 unsigned Num = (unsigned)D.Num;
204 // Module constants occupy the lowest numbered slots...
205 if (type < CurModule.Values.size()) {
206 if (Num < CurModule.Values[type].size())
207 return CurModule.Values[type][Num];
209 Num -= CurModule.Values[type].size();
212 // Make sure that our type is within bounds
213 if (CurMeth.Values.size() <= type)
216 // Check that the number is within bounds...
217 if (CurMeth.Values[type].size() <= Num)
220 return CurMeth.Values[type][Num];
222 case ValID::NameVal: { // Is it a named definition?
224 Value *N = lookupInSymbolTable(Ty, Name);
227 D.destroy(); // Free old strdup'd memory...
231 case ValID::ConstSIntVal: // Is it a constant pool reference??
232 case ValID::ConstUIntVal: // Is it an unsigned const pool reference?
233 case ValID::ConstStringVal: // Is it a string const pool reference?
234 case ValID::ConstFPVal: // Is it a floating point const pool reference?
235 case ValID::ConstNullVal: { // Is it a null value?
236 ConstPoolVal *CPV = 0;
238 // Check to make sure that "Ty" is an integral type, and that our
239 // value will fit into the specified type...
241 case ValID::ConstSIntVal:
242 if (Ty == Type::BoolTy) { // Special handling for boolean data
243 CPV = ConstPoolBool::get(D.ConstPool64 != 0);
245 if (!ConstPoolSInt::isValueValidForType(Ty, D.ConstPool64))
246 ThrowException("Symbolic constant pool value '" +
247 itostr(D.ConstPool64) + "' is invalid for type '" +
248 Ty->getName() + "'!");
249 CPV = ConstPoolSInt::get(Ty, D.ConstPool64);
252 case ValID::ConstUIntVal:
253 if (!ConstPoolUInt::isValueValidForType(Ty, D.UConstPool64)) {
254 if (!ConstPoolSInt::isValueValidForType(Ty, D.ConstPool64)) {
255 ThrowException("Integral constant pool reference is invalid!");
256 } else { // This is really a signed reference. Transmogrify.
257 CPV = ConstPoolSInt::get(Ty, D.ConstPool64);
260 CPV = ConstPoolUInt::get(Ty, D.UConstPool64);
263 case ValID::ConstStringVal:
264 cerr << "FIXME: TODO: String constants [sbyte] not implemented yet!\n";
267 case ValID::ConstFPVal:
268 if (!ConstPoolFP::isValueValidForType(Ty, D.ConstPoolFP))
269 ThrowException("FP constant invalid for type!!");
270 CPV = ConstPoolFP::get(Ty, D.ConstPoolFP);
272 case ValID::ConstNullVal:
273 if (!Ty->isPointerType())
274 ThrowException("Cannot create a a non pointer null!");
275 CPV = ConstPoolPointer::getNullPointer(cast<PointerType>(Ty));
278 assert(0 && "Unhandled case!");
280 assert(CPV && "How did we escape creating a constant??");
282 } // End of case 2,3,4
284 assert(0 && "Unhandled case!");
288 // If we reached here, we referenced either a symbol that we don't know about
289 // or an id number that hasn't been read yet. We may be referencing something
290 // forward, so just create an entry to be resolved later and get to it...
292 if (DoNotImprovise) return 0; // Do we just want a null to be returned?
295 vector<ValueList> *LateResolver = (CurMeth.CurrentMethod) ?
296 &CurMeth.LateResolveValues : &CurModule.LateResolveValues;
298 if (const PointerType *PTy = dyn_cast<PointerType>(Ty))
299 if (const MethodType *MTy = dyn_cast<MethodType>(PTy->getValueType()))
300 Ty = MTy; // Convert pointer to method to method type
302 switch (Ty->getPrimitiveID()) {
303 case Type::LabelTyID: d = new BBPlaceHolder(Ty, D); break;
304 case Type::MethodTyID: d = new MethPlaceHolder(Ty, D);
305 LateResolver = &CurModule.LateResolveValues; break;
306 default: d = new ValuePlaceHolder(Ty, D); break;
309 assert(d != 0 && "How did we not make something?");
310 InsertValue(d, *LateResolver);
315 //===----------------------------------------------------------------------===//
316 // Code to handle forward references in instructions
317 //===----------------------------------------------------------------------===//
319 // This code handles the late binding needed with statements that reference
320 // values not defined yet... for example, a forward branch, or the PHI node for
323 // This keeps a table (CurMeth.LateResolveValues) of all such forward references
324 // and back patchs after we are done.
327 // ResolveDefinitions - If we could not resolve some defs at parsing
328 // time (forward branches, phi functions for loops, etc...) resolve the
331 static void ResolveDefinitions(vector<ValueList> &LateResolvers) {
332 // Loop over LateResolveDefs fixing up stuff that couldn't be resolved
333 for (unsigned ty = 0; ty < LateResolvers.size(); ty++) {
334 while (!LateResolvers[ty].empty()) {
335 Value *V = LateResolvers[ty].back();
336 LateResolvers[ty].pop_back();
337 ValID &DID = getValIDFromPlaceHolder(V);
339 Value *TheRealValue = getVal(Type::getUniqueIDType(ty), DID, true);
341 if (TheRealValue == 0) {
343 ThrowException("Reference to an invalid definition: '" +DID.getName()+
344 "' of type '" + V->getType()->getDescription() + "'",
345 getLineNumFromPlaceHolder(V));
347 ThrowException("Reference to an invalid definition: #" +
348 itostr(DID.Num) + " of type '" +
349 V->getType()->getDescription() + "'",
350 getLineNumFromPlaceHolder(V));
353 assert(!isa<Type>(V) && "Types should be in LateResolveTypes!");
355 V->replaceAllUsesWith(TheRealValue);
360 LateResolvers.clear();
363 // ResolveType - Take a specified unresolved type and resolve it. If there is
364 // nothing to resolve it to yet, return true. Otherwise resolve it and return
367 static bool ResolveType(PATypeHolder<Type> &T) {
369 ValID &DID = getValIDFromPlaceHolder(Ty);
371 const Type *TheRealType = getTypeVal(DID, true);
372 if (TheRealType == 0) return true;
374 // Refine the opaque type we had to the new type we are getting.
375 cast<DerivedType>(Ty)->refineAbstractTypeTo(TheRealType);
380 // ResolveTypes - This goes through the forward referenced type table and makes
381 // sure that all type references are complete. This code is executed after the
382 // constant pool of a method or module is completely parsed.
384 static void ResolveTypes(vector<PATypeHolder<Type> > &LateResolveTypes) {
385 while (!LateResolveTypes.empty()) {
386 if (ResolveType(LateResolveTypes.back())) {
387 const Type *Ty = LateResolveTypes.back();
388 ValID &DID = getValIDFromPlaceHolder(Ty);
390 if (DID.Type == ValID::NameVal)
391 ThrowException("Reference to an invalid type: '" +DID.getName(),
392 getLineNumFromPlaceHolder(Ty));
394 ThrowException("Reference to an invalid type: #" + itostr(DID.Num),
395 getLineNumFromPlaceHolder(Ty));
398 // No need to delete type, refine does that for us.
399 LateResolveTypes.pop_back();
404 // ResolveSomeTypes - This goes through the forward referenced type table and
405 // completes references that are now done. This is so that types are
406 // immediately resolved to be as concrete as possible. This does not cause
407 // thrown exceptions if not everything is resolved.
409 static void ResolveSomeTypes(vector<PATypeHolder<Type> > &LateResolveTypes) {
410 for (unsigned i = 0; i < LateResolveTypes.size(); ) {
411 if (ResolveType(LateResolveTypes[i]))
412 ++i; // Type didn't resolve
414 LateResolveTypes.erase(LateResolveTypes.begin()+i); // Type resolved!
419 // setValueName - Set the specified value to the name given. The name may be
420 // null potentially, in which case this is a noop. The string passed in is
421 // assumed to be a malloc'd string buffer, and is freed by this function.
423 static void setValueName(Value *V, char *NameStr) {
424 if (NameStr == 0) return;
425 string Name(NameStr); // Copy string
426 free(NameStr); // Free old string
428 SymbolTable *ST = CurMeth.CurrentMethod ?
429 CurMeth.CurrentMethod->getSymbolTableSure() :
430 CurModule.CurrentModule->getSymbolTableSure();
432 Value *Existing = ST->lookup(V->getType(), Name);
433 if (Existing) { // Inserting a name that is already defined???
434 // There is only one case where this is allowed: when we are refining an
435 // opaque type. In this case, Existing will be an opaque type.
436 if (const Type *Ty = dyn_cast<const Type>(Existing)) {
437 if (OpaqueType *OpTy = dyn_cast<OpaqueType>(Ty)) {
438 // We ARE replacing an opaque type!
439 OpTy->refineAbstractTypeTo(cast<Type>(V));
444 // Otherwise, we are a simple redefinition of a value, check to see if it
445 // is defined the same as the old one...
446 if (const Type *Ty = dyn_cast<const Type>(Existing)) {
447 if (Ty == cast<const Type>(V)) return; // Yes, it's equal.
448 cerr << "Type: " << Ty->getDescription() << " != "
449 << cast<const Type>(V)->getDescription() << "!\n";
453 ThrowException("Redefinition of value name '" + Name + "' in the '" +
454 V->getType()->getDescription() + "' type plane!");
457 V->setName(Name, ST);
461 //===----------------------------------------------------------------------===//
462 // Code for handling upreferences in type names...
465 // TypeContains - Returns true if Ty contains E in it.
467 static bool TypeContains(const Type *Ty, const Type *E) {
468 return find(df_begin(Ty), df_end(Ty), E) != df_end(Ty);
472 static vector<pair<unsigned, OpaqueType *> > UpRefs;
474 static PATypeHolder<Type> HandleUpRefs(const Type *ty) {
475 PATypeHolder<Type> Ty(ty);
476 UR_OUT(UpRefs.size() << " upreferences active!\n");
477 for (unsigned i = 0; i < UpRefs.size(); ) {
478 UR_OUT("TypeContains(" << Ty->getDescription() << ", "
479 << UpRefs[i].second->getDescription() << ") = "
480 << TypeContains(Ty, UpRefs[i].second) << endl);
481 if (TypeContains(Ty, UpRefs[i].second)) {
482 unsigned Level = --UpRefs[i].first; // Decrement level of upreference
483 UR_OUT("Uplevel Ref Level = " << Level << endl);
484 if (Level == 0) { // Upreference should be resolved!
485 UR_OUT("About to resolve upreference!\n";
486 string OldName = UpRefs[i].second->getDescription());
487 UpRefs[i].second->refineAbstractTypeTo(Ty);
488 UpRefs.erase(UpRefs.begin()+i); // Remove from upreference list...
489 UR_OUT("Type '" << OldName << "' refined upreference to: "
490 << (const void*)Ty << ", " << Ty->getDescription() << endl);
495 ++i; // Otherwise, no resolve, move on...
497 // FIXME: TODO: this should return the updated type
501 template <class TypeTy>
502 inline static void TypeDone(PATypeHolder<TypeTy> *Ty) {
504 ThrowException("Invalid upreference in type: " + (*Ty)->getDescription());
507 // newTH - Allocate a new type holder for the specified type
508 template <class TypeTy>
509 inline static PATypeHolder<TypeTy> *newTH(const TypeTy *Ty) {
510 return new PATypeHolder<TypeTy>(Ty);
512 template <class TypeTy>
513 inline static PATypeHolder<TypeTy> *newTH(const PATypeHolder<TypeTy> &TH) {
514 return new PATypeHolder<TypeTy>(TH);
518 //===----------------------------------------------------------------------===//
519 // RunVMAsmParser - Define an interface to this parser
520 //===----------------------------------------------------------------------===//
522 Module *RunVMAsmParser(const string &Filename, FILE *F) {
524 CurFilename = Filename;
525 llvmAsmlineno = 1; // Reset the current line number...
527 CurModule.CurrentModule = new Module(); // Allocate a new module to read
528 yyparse(); // Parse the file.
529 Module *Result = ParserResult;
530 llvmAsmin = stdin; // F is about to go away, don't use it anymore...
541 MethodArgument *MethArgVal;
542 BasicBlock *BasicBlockVal;
543 TerminatorInst *TermInstVal;
544 Instruction *InstVal;
545 ConstPoolVal *ConstVal;
547 const Type *PrimType;
548 PATypeHolder<Type> *TypeVal;
551 list<MethodArgument*> *MethodArgList;
552 list<Value*> *ValueList;
553 list<PATypeHolder<Type> > *TypeList;
554 list<pair<Value*, BasicBlock*> > *PHIList; // Represent the RHS of PHI node
555 list<pair<ConstPoolVal*, BasicBlock*> > *JumpTable;
556 vector<ConstPoolVal*> *ConstVector;
565 char *StrVal; // This memory is strdup'd!
566 ValID ValIDVal; // strdup'd memory maybe!
568 Instruction::UnaryOps UnaryOpVal;
569 Instruction::BinaryOps BinaryOpVal;
570 Instruction::TermOps TermOpVal;
571 Instruction::MemoryOps MemOpVal;
572 Instruction::OtherOps OtherOpVal;
575 %type <ModuleVal> Module MethodList
576 %type <MethodVal> Method MethodProto MethodHeader BasicBlockList
577 %type <BasicBlockVal> BasicBlock InstructionList
578 %type <TermInstVal> BBTerminatorInst
579 %type <InstVal> Inst InstVal MemoryInst
580 %type <ConstVal> ConstVal
581 %type <ConstVector> ConstVector UByteList
582 %type <MethodArgList> ArgList ArgListH
583 %type <MethArgVal> ArgVal
584 %type <PHIList> PHIList
585 %type <ValueList> ValueRefList ValueRefListE // For call param lists
586 %type <TypeList> TypeListI ArgTypeListI
587 %type <JumpTable> JumpTable
588 %type <BoolVal> GlobalType // GLOBAL or CONSTANT?
590 %type <ValIDVal> ValueRef ConstValueRef // Reference to a definition or BB
591 %type <ValueVal> ResolvedVal // <type> <valref> pair
592 // Tokens and types for handling constant integer values
594 // ESINT64VAL - A negative number within long long range
595 %token <SInt64Val> ESINT64VAL
597 // EUINT64VAL - A positive number within uns. long long range
598 %token <UInt64Val> EUINT64VAL
599 %type <SInt64Val> EINT64VAL
601 %token <SIntVal> SINTVAL // Signed 32 bit ints...
602 %token <UIntVal> UINTVAL // Unsigned 32 bit ints...
603 %type <SIntVal> INTVAL
604 %token <FPVal> FPVAL // Float or Double constant
607 %type <TypeVal> Types TypesV UpRTypes UpRTypesV
608 %type <PrimType> SIntType UIntType IntType FPType PrimType // Classifications
609 %token <TypeVal> OPAQUE
610 %token <PrimType> VOID BOOL SBYTE UBYTE SHORT USHORT INT UINT LONG ULONG
611 %token <PrimType> FLOAT DOUBLE TYPE LABEL
613 %token <StrVal> VAR_ID LABELSTR STRINGCONSTANT
614 %type <StrVal> OptVAR_ID OptAssign
617 %token IMPLEMENTATION TRUE FALSE BEGINTOK END DECLARE GLOBAL CONSTANT UNINIT
618 %token TO DOTDOTDOT STRING NULL_TOK CONST
620 // Basic Block Terminating Operators
621 %token <TermOpVal> RET BR SWITCH
624 %type <UnaryOpVal> UnaryOps // all the unary operators
625 %token <UnaryOpVal> NOT
628 %type <BinaryOpVal> BinaryOps // all the binary operators
629 %token <BinaryOpVal> ADD SUB MUL DIV REM
630 %token <BinaryOpVal> SETLE SETGE SETLT SETGT SETEQ SETNE // Binary Comarators
632 // Memory Instructions
633 %token <MemoryOpVal> MALLOC ALLOCA FREE LOAD STORE GETELEMENTPTR
636 %type <OtherOpVal> ShiftOps
637 %token <OtherOpVal> PHI CALL CAST SHL SHR
642 // Handle constant integer size restriction and conversion...
647 if ($1 > (uint32_t)INT32_MAX) // Outside of my range!
648 ThrowException("Value too large for type!");
653 EINT64VAL : ESINT64VAL // These have same type and can't cause problems...
654 EINT64VAL : EUINT64VAL {
655 if ($1 > (uint64_t)INT64_MAX) // Outside of my range!
656 ThrowException("Value too large for type!");
660 // Operations that are notably excluded from this list include:
661 // RET, BR, & SWITCH because they end basic blocks and are treated specially.
664 BinaryOps : ADD | SUB | MUL | DIV | REM
665 BinaryOps : SETLE | SETGE | SETLT | SETGT | SETEQ | SETNE
668 // These are some types that allow classification if we only want a particular
669 // thing... for example, only a signed, unsigned, or integral type.
670 SIntType : LONG | INT | SHORT | SBYTE
671 UIntType : ULONG | UINT | USHORT | UBYTE
672 IntType : SIntType | UIntType
673 FPType : FLOAT | DOUBLE
675 // OptAssign - Value producing statements have an optional assignment component
676 OptAssign : VAR_ID '=' {
684 //===----------------------------------------------------------------------===//
685 // Types includes all predefined types... except void, because it can only be
686 // used in specific contexts (method returning void for example). To have
687 // access to it, a user must explicitly use TypesV.
690 // TypesV includes all of 'Types', but it also includes the void type.
691 TypesV : Types | VOID { $$ = newTH($1); }
692 UpRTypesV : UpRTypes | VOID { $$ = newTH($1); }
699 // Derived types are added later...
701 PrimType : BOOL | SBYTE | UBYTE | SHORT | USHORT | INT | UINT
702 PrimType : LONG | ULONG | FLOAT | DOUBLE | TYPE | LABEL
703 UpRTypes : OPAQUE | PrimType { $$ = newTH($1); }
704 UpRTypes : ValueRef { // Named types are also simple types...
705 $$ = newTH(getTypeVal($1));
708 // Include derived types in the Types production.
710 UpRTypes : '\\' EUINT64VAL { // Type UpReference
711 if ($2 > (uint64_t)INT64_MAX) ThrowException("Value out of range!");
712 OpaqueType *OT = OpaqueType::get(); // Use temporary placeholder
713 UpRefs.push_back(make_pair((unsigned)$2, OT)); // Add to vector...
714 $$ = newTH<Type>(OT);
715 UR_OUT("New Upreference!\n");
717 | UpRTypesV '(' ArgTypeListI ')' { // Method derived type?
718 vector<const Type*> Params;
719 mapto($3->begin(), $3->end(), back_inserter(Params),
720 mem_fun_ref(&PATypeHandle<Type>::get));
721 $$ = newTH(HandleUpRefs(MethodType::get(*$1, Params)));
722 delete $3; // Delete the argument list
723 delete $1; // Delete the old type handle
725 | '[' UpRTypesV ']' { // Unsized array type?
726 $$ = newTH<Type>(HandleUpRefs(ArrayType::get(*$2)));
729 | '[' EUINT64VAL 'x' UpRTypes ']' { // Sized array type?
730 $$ = newTH<Type>(HandleUpRefs(ArrayType::get(*$4, (int)$2)));
733 | '{' TypeListI '}' { // Structure type?
734 vector<const Type*> Elements;
735 mapto($2->begin(), $2->end(), back_inserter(Elements),
736 mem_fun_ref(&PATypeHandle<Type>::get));
738 $$ = newTH<Type>(HandleUpRefs(StructType::get(Elements)));
741 | '{' '}' { // Empty structure type?
742 $$ = newTH<Type>(StructType::get(vector<const Type*>()));
744 | UpRTypes '*' { // Pointer type?
745 $$ = newTH<Type>(HandleUpRefs(PointerType::get(*$1)));
749 // TypeList - Used for struct declarations and as a basis for method type
750 // declaration type lists
752 TypeListI : UpRTypes {
753 $$ = new list<PATypeHolder<Type> >();
754 $$->push_back(*$1); delete $1;
756 | TypeListI ',' UpRTypes {
757 ($$=$1)->push_back(*$3); delete $3;
760 // ArgTypeList - List of types for a method type declaration...
761 ArgTypeListI : TypeListI
762 | TypeListI ',' DOTDOTDOT {
763 ($$=$1)->push_back(Type::VoidTy);
766 ($$ = new list<PATypeHolder<Type> >())->push_back(Type::VoidTy);
769 $$ = new list<PATypeHolder<Type> >();
773 // ConstVal - The various declarations that go into the constant pool. This
774 // includes all forward declarations of types, constants, and functions.
776 ConstVal: Types '[' ConstVector ']' { // Nonempty unsized arr
777 const ArrayType *ATy = dyn_cast<const ArrayType>($1->get());
779 ThrowException("Cannot make array constant with type: '" +
780 (*$1)->getDescription() + "'!");
781 const Type *ETy = ATy->getElementType();
782 int NumElements = ATy->getNumElements();
784 // Verify that we have the correct size...
785 if (NumElements != -1 && NumElements != (int)$3->size())
786 ThrowException("Type mismatch: constant sized array initialized with " +
787 utostr($3->size()) + " arguments, but has size of " +
788 itostr(NumElements) + "!");
790 // Verify all elements are correct type!
791 for (unsigned i = 0; i < $3->size(); i++) {
792 if (ETy != (*$3)[i]->getType())
793 ThrowException("Element #" + utostr(i) + " is not of type '" +
794 ETy->getName() + "' as required!\nIt is of type '" +
795 (*$3)[i]->getType()->getName() + "'.");
798 $$ = ConstPoolArray::get(ATy, *$3);
799 delete $1; delete $3;
802 const ArrayType *ATy = dyn_cast<const ArrayType>($1->get());
804 ThrowException("Cannot make array constant with type: '" +
805 (*$1)->getDescription() + "'!");
807 int NumElements = ATy->getNumElements();
808 if (NumElements != -1 && NumElements != 0)
809 ThrowException("Type mismatch: constant sized array initialized with 0"
810 " arguments, but has size of " + itostr(NumElements) +"!");
811 $$ = ConstPoolArray::get(ATy, vector<ConstPoolVal*>());
814 | Types 'c' STRINGCONSTANT {
815 const ArrayType *ATy = dyn_cast<const ArrayType>($1->get());
817 ThrowException("Cannot make array constant with type: '" +
818 (*$1)->getDescription() + "'!");
820 int NumElements = ATy->getNumElements();
821 const Type *ETy = ATy->getElementType();
822 char *EndStr = UnEscapeLexed($3, true);
823 if (NumElements != -1 && NumElements != (EndStr-$3))
824 ThrowException("Can't build string constant of size " +
825 itostr((int)(EndStr-$3)) +
826 " when array has size " + itostr(NumElements) + "!");
827 vector<ConstPoolVal*> Vals;
828 if (ETy == Type::SByteTy) {
829 for (char *C = $3; C != EndStr; ++C)
830 Vals.push_back(ConstPoolSInt::get(ETy, *C));
831 } else if (ETy == Type::UByteTy) {
832 for (char *C = $3; C != EndStr; ++C)
833 Vals.push_back(ConstPoolUInt::get(ETy, *C));
836 ThrowException("Cannot build string arrays of non byte sized elements!");
839 $$ = ConstPoolArray::get(ATy, Vals);
842 | Types '{' ConstVector '}' {
843 const StructType *STy = dyn_cast<const StructType>($1->get());
845 ThrowException("Cannot make struct constant with type: '" +
846 (*$1)->getDescription() + "'!");
847 // FIXME: TODO: Check to see that the constants are compatible with the type
849 $$ = ConstPoolStruct::get(STy, *$3);
850 delete $1; delete $3;
853 const PointerType *PTy = dyn_cast<const PointerType>($1->get());
855 ThrowException("Cannot make null pointer constant with type: '" +
856 (*$1)->getDescription() + "'!");
858 $$ = ConstPoolPointer::getNullPointer(PTy);
862 string Name($2); free($2); // Change to a responsible mem manager
863 const PointerType *Ty = dyn_cast<const PointerType>($1->get());
865 ThrowException("Global const reference must be a pointer type!");
867 Value *N = lookupInSymbolTable(Ty, Name);
869 ThrowException("Global pointer reference '%" + Name +
870 "' must be defined before use!");
872 // TODO FIXME: This should also allow methods... when common baseclass
874 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(N)) {
875 $$ = ConstPoolPointerReference::get(GV);
877 ThrowException("'%" + Name + "' is not a global value reference!");
884 ConstVal : SIntType EINT64VAL { // integral constants
885 if (!ConstPoolSInt::isValueValidForType($1, $2))
886 ThrowException("Constant value doesn't fit in type!");
887 $$ = ConstPoolSInt::get($1, $2);
889 | UIntType EUINT64VAL { // integral constants
890 if (!ConstPoolUInt::isValueValidForType($1, $2))
891 ThrowException("Constant value doesn't fit in type!");
892 $$ = ConstPoolUInt::get($1, $2);
894 | BOOL TRUE { // Boolean constants
895 $$ = ConstPoolBool::True;
897 | BOOL FALSE { // Boolean constants
898 $$ = ConstPoolBool::False;
900 | FPType FPVAL { // Float & Double constants
901 $$ = ConstPoolFP::get($1, $2);
904 // ConstVector - A list of comma seperated constants.
905 ConstVector : ConstVector ',' ConstVal {
906 ($$ = $1)->push_back($3);
909 $$ = new vector<ConstPoolVal*>();
914 // GlobalType - Match either GLOBAL or CONSTANT for global declarations...
915 GlobalType : GLOBAL { $$ = false; } | CONSTANT { $$ = true; }
918 // ConstPool - Constants with optional names assigned to them.
919 ConstPool : ConstPool OptAssign CONST ConstVal {
920 setValueName($4, $2);
923 | ConstPool OptAssign TYPE TypesV { // Types can be defined in the const pool
924 // TODO: FIXME when Type are not const
925 setValueName(const_cast<Type*>($4->get()), $2);
928 InsertType($4->get(),
929 CurMeth.CurrentMethod ? CurMeth.Types : CurModule.Types);
933 ResolveSomeTypes(CurMeth.CurrentMethod ? CurMeth.LateResolveTypes :
934 CurModule.LateResolveTypes);
936 | ConstPool MethodProto { // Method prototypes can be in const pool
938 | ConstPool OptAssign GlobalType ConstVal {
939 const Type *Ty = $4->getType();
940 // Global declarations appear in Constant Pool
941 ConstPoolVal *Initializer = $4;
942 if (Initializer == 0)
943 ThrowException("Global value initializer is not a constant!");
945 GlobalVariable *GV = new GlobalVariable(Ty, $3, Initializer);
946 setValueName(GV, $2);
948 CurModule.CurrentModule->getGlobalList().push_back(GV);
949 InsertValue(GV, CurModule.Values);
951 | ConstPool OptAssign UNINIT GlobalType Types {
952 const Type *Ty = *$5;
953 // Global declarations appear in Constant Pool
954 if (isa<ArrayType>(Ty) && cast<ArrayType>(Ty)->isUnsized()) {
955 ThrowException("Type '" + Ty->getDescription() +
956 "' is not a sized type!");
959 GlobalVariable *GV = new GlobalVariable(Ty, $4);
960 setValueName(GV, $2);
962 CurModule.CurrentModule->getGlobalList().push_back(GV);
963 InsertValue(GV, CurModule.Values);
965 | /* empty: end of list */ {
969 //===----------------------------------------------------------------------===//
970 // Rules to match Modules
971 //===----------------------------------------------------------------------===//
973 // Module rule: Capture the result of parsing the whole file into a result
976 Module : MethodList {
977 $$ = ParserResult = $1;
978 CurModule.ModuleDone();
981 // MethodList - A list of methods, preceeded by a constant pool.
983 MethodList : MethodList Method {
985 if (!$2->getParent())
986 $1->getMethodList().push_back($2);
987 CurMeth.MethodDone();
989 | MethodList MethodProto {
992 | ConstPool IMPLEMENTATION {
993 $$ = CurModule.CurrentModule;
994 // Resolve circular types before we parse the body of the module
995 ResolveTypes(CurModule.LateResolveTypes);
999 //===----------------------------------------------------------------------===//
1000 // Rules to match Method Headers
1001 //===----------------------------------------------------------------------===//
1003 OptVAR_ID : VAR_ID | /*empty*/ { $$ = 0; }
1005 ArgVal : Types OptVAR_ID {
1006 $$ = new MethodArgument(*$1); delete $1;
1007 setValueName($$, $2);
1010 ArgListH : ArgVal ',' ArgListH {
1015 $$ = new list<MethodArgument*>();
1019 $$ = new list<MethodArgument*>();
1020 $$->push_back(new MethodArgument(Type::VoidTy));
1023 ArgList : ArgListH {
1030 MethodHeaderH : TypesV STRINGCONSTANT '(' ArgList ')' {
1032 vector<const Type*> ParamTypeList;
1034 for (list<MethodArgument*>::iterator I = $4->begin(); I != $4->end(); ++I)
1035 ParamTypeList.push_back((*I)->getType());
1037 const MethodType *MT = MethodType::get(*$1, ParamTypeList);
1038 const PointerType *PMT = PointerType::get(MT);
1042 if (SymbolTable *ST = CurModule.CurrentModule->getSymbolTable()) {
1043 if (Value *V = ST->lookup(PMT, $2)) { // Method already in symtab?
1044 M = cast<Method>(V);
1046 // Yes it is. If this is the case, either we need to be a forward decl,
1047 // or it needs to be.
1048 if (!CurMeth.isDeclare && !M->isExternal())
1049 ThrowException("Redefinition of method '" + string($2) + "'!");
1053 if (M == 0) { // Not already defined?
1054 M = new Method(MT, $2);
1055 InsertValue(M, CurModule.Values);
1058 free($2); // Free strdup'd memory!
1060 CurMeth.MethodStart(M);
1062 // Add all of the arguments we parsed to the method...
1063 if ($4 && !CurMeth.isDeclare) { // Is null if empty...
1064 Method::ArgumentListType &ArgList = M->getArgumentList();
1066 for (list<MethodArgument*>::iterator I = $4->begin(); I != $4->end(); ++I) {
1068 ArgList.push_back(*I);
1070 delete $4; // We're now done with the argument list
1074 MethodHeader : MethodHeaderH ConstPool BEGINTOK {
1075 $$ = CurMeth.CurrentMethod;
1077 // Resolve circular types before we parse the body of the method.
1078 ResolveTypes(CurMeth.LateResolveTypes);
1081 Method : BasicBlockList END {
1085 MethodProto : DECLARE { CurMeth.isDeclare = true; } MethodHeaderH {
1086 $$ = CurMeth.CurrentMethod;
1087 if (!$$->getParent())
1088 CurModule.CurrentModule->getMethodList().push_back($$);
1089 CurMeth.MethodDone();
1092 //===----------------------------------------------------------------------===//
1093 // Rules to match Basic Blocks
1094 //===----------------------------------------------------------------------===//
1096 ConstValueRef : ESINT64VAL { // A reference to a direct constant
1097 $$ = ValID::create($1);
1100 $$ = ValID::create($1);
1102 | FPVAL { // Perhaps it's an FP constant?
1103 $$ = ValID::create($1);
1106 $$ = ValID::create((int64_t)1);
1109 $$ = ValID::create((int64_t)0);
1112 $$ = ValID::createNull();
1116 | STRINGCONSTANT { // Quoted strings work too... especially for methods
1117 $$ = ValID::create_conststr($1);
1121 // ValueRef - A reference to a definition...
1122 ValueRef : INTVAL { // Is it an integer reference...?
1123 $$ = ValID::create($1);
1125 | VAR_ID { // Is it a named reference...?
1126 $$ = ValID::create($1);
1132 // ResolvedVal - a <type> <value> pair. This is used only in cases where the
1133 // type immediately preceeds the value reference, and allows complex constant
1134 // pool references (for things like: 'ret [2 x int] [ int 12, int 42]')
1135 ResolvedVal : Types ValueRef {
1136 $$ = getVal(*$1, $2); delete $1;
1140 BasicBlockList : BasicBlockList BasicBlock {
1141 $1->getBasicBlocks().push_back($2);
1144 | MethodHeader BasicBlock { // Do not allow methods with 0 basic blocks
1145 $$ = $1; // in them...
1146 $1->getBasicBlocks().push_back($2);
1150 // Basic blocks are terminated by branching instructions:
1151 // br, br/cc, switch, ret
1153 BasicBlock : InstructionList BBTerminatorInst {
1154 $1->getInstList().push_back($2);
1158 | LABELSTR InstructionList BBTerminatorInst {
1159 $2->getInstList().push_back($3);
1160 setValueName($2, $1);
1166 InstructionList : InstructionList Inst {
1167 $1->getInstList().push_back($2);
1171 $$ = new BasicBlock();
1174 BBTerminatorInst : RET ResolvedVal { // Return with a result...
1175 $$ = new ReturnInst($2);
1177 | RET VOID { // Return with no result...
1178 $$ = new ReturnInst();
1180 | BR LABEL ValueRef { // Unconditional Branch...
1181 $$ = new BranchInst(cast<BasicBlock>(getVal(Type::LabelTy, $3)));
1182 } // Conditional Branch...
1183 | BR BOOL ValueRef ',' LABEL ValueRef ',' LABEL ValueRef {
1184 $$ = new BranchInst(cast<BasicBlock>(getVal(Type::LabelTy, $6)),
1185 cast<BasicBlock>(getVal(Type::LabelTy, $9)),
1186 getVal(Type::BoolTy, $3));
1188 | SWITCH IntType ValueRef ',' LABEL ValueRef '[' JumpTable ']' {
1189 SwitchInst *S = new SwitchInst(getVal($2, $3),
1190 cast<BasicBlock>(getVal(Type::LabelTy, $6)));
1193 list<pair<ConstPoolVal*, BasicBlock*> >::iterator I = $8->begin(),
1195 for (; I != end; ++I)
1196 S->dest_push_back(I->first, I->second);
1199 JumpTable : JumpTable IntType ConstValueRef ',' LABEL ValueRef {
1201 ConstPoolVal *V = cast<ConstPoolVal>(getVal($2, $3, true));
1203 ThrowException("May only switch on a constant pool value!");
1205 $$->push_back(make_pair(V, cast<BasicBlock>(getVal($5, $6))));
1207 | IntType ConstValueRef ',' LABEL ValueRef {
1208 $$ = new list<pair<ConstPoolVal*, BasicBlock*> >();
1209 ConstPoolVal *V = cast<ConstPoolVal>(getVal($1, $2, true));
1212 ThrowException("May only switch on a constant pool value!");
1214 $$->push_back(make_pair(V, cast<BasicBlock>(getVal($4, $5))));
1217 Inst : OptAssign InstVal {
1218 setValueName($2, $1); // Is this definition named?? if so, assign the name...
1224 PHIList : Types '[' ValueRef ',' ValueRef ']' { // Used for PHI nodes
1225 $$ = new list<pair<Value*, BasicBlock*> >();
1226 $$->push_back(make_pair(getVal(*$1, $3),
1227 cast<BasicBlock>(getVal(Type::LabelTy, $5))));
1230 | PHIList ',' '[' ValueRef ',' ValueRef ']' {
1232 $1->push_back(make_pair(getVal($1->front().first->getType(), $4),
1233 cast<BasicBlock>(getVal(Type::LabelTy, $6))));
1237 ValueRefList : ResolvedVal { // Used for call statements, and memory insts...
1238 $$ = new list<Value*>();
1241 | ValueRefList ',' ResolvedVal {
1246 // ValueRefListE - Just like ValueRefList, except that it may also be empty!
1247 ValueRefListE : ValueRefList | /*empty*/ { $$ = 0; }
1249 InstVal : BinaryOps Types ValueRef ',' ValueRef {
1250 $$ = BinaryOperator::create($1, getVal(*$2, $3), getVal(*$2, $5));
1252 ThrowException("binary operator returned null!");
1255 | UnaryOps ResolvedVal {
1256 $$ = UnaryOperator::create($1, $2);
1258 ThrowException("unary operator returned null!");
1260 | ShiftOps ResolvedVal ',' ResolvedVal {
1261 if ($4->getType() != Type::UByteTy)
1262 ThrowException("Shift amount must be ubyte!");
1263 $$ = new ShiftInst($1, $2, $4);
1265 | CAST ResolvedVal TO Types {
1266 $$ = new CastInst($2, *$4);
1270 const Type *Ty = $2->front().first->getType();
1271 $$ = new PHINode(Ty);
1272 while ($2->begin() != $2->end()) {
1273 if ($2->front().first->getType() != Ty)
1274 ThrowException("All elements of a PHI node must be of the same type!");
1275 cast<PHINode>($$)->addIncoming($2->front().first, $2->front().second);
1278 delete $2; // Free the list...
1280 | CALL TypesV ValueRef '(' ValueRefListE ')' {
1281 const PointerType *PMTy;
1282 const MethodType *Ty;
1284 if (!(PMTy = dyn_cast<PointerType>($2->get())) ||
1285 !(Ty = dyn_cast<MethodType>(PMTy->getValueType()))) {
1286 // Pull out the types of all of the arguments...
1287 vector<const Type*> ParamTypes;
1289 for (list<Value*>::iterator I = $5->begin(), E = $5->end(); I != E; ++I)
1290 ParamTypes.push_back((*I)->getType());
1292 Ty = MethodType::get($2->get(), ParamTypes);
1293 PMTy = PointerType::get(Ty);
1297 Value *V = getVal(PMTy, $3); // Get the method we're calling...
1299 // Create the call node...
1300 if (!$5) { // Has no arguments?
1301 $$ = new CallInst(cast<Method>(V), vector<Value*>());
1302 } else { // Has arguments?
1303 // Loop through MethodType's arguments and ensure they are specified
1306 MethodType::ParamTypes::const_iterator I = Ty->getParamTypes().begin();
1307 MethodType::ParamTypes::const_iterator E = Ty->getParamTypes().end();
1308 list<Value*>::iterator ArgI = $5->begin(), ArgE = $5->end();
1310 for (; ArgI != ArgE && I != E; ++ArgI, ++I)
1311 if ((*ArgI)->getType() != *I)
1312 ThrowException("Parameter " +(*ArgI)->getName()+ " is not of type '" +
1313 (*I)->getName() + "'!");
1315 if (I != E || (ArgI != ArgE && !Ty->isVarArg()))
1316 ThrowException("Invalid number of parameters detected!");
1318 $$ = new CallInst(cast<Method>(V),
1319 vector<Value*>($5->begin(), $5->end()));
1327 // UByteList - List of ubyte values for load and store instructions
1328 UByteList : ',' ConstVector {
1331 $$ = new vector<ConstPoolVal*>();
1334 MemoryInst : MALLOC Types {
1335 $$ = new MallocInst(PointerType::get(*$2));
1338 | MALLOC Types ',' UINT ValueRef {
1339 if (!(*$2)->isArrayType() || cast<const ArrayType>($2->get())->isSized())
1340 ThrowException("Trying to allocate " + (*$2)->getName() +
1341 " as unsized array!");
1342 const Type *Ty = PointerType::get(*$2);
1343 $$ = new MallocInst(Ty, getVal($4, $5));
1347 $$ = new AllocaInst(PointerType::get(*$2));
1350 | ALLOCA Types ',' UINT ValueRef {
1351 if (!(*$2)->isArrayType() || cast<const ArrayType>($2->get())->isSized())
1352 ThrowException("Trying to allocate " + (*$2)->getName() +
1353 " as unsized array!");
1354 const Type *Ty = PointerType::get(*$2);
1355 Value *ArrSize = getVal($4, $5);
1356 $$ = new AllocaInst(Ty, ArrSize);
1359 | FREE ResolvedVal {
1360 if (!$2->getType()->isPointerType())
1361 ThrowException("Trying to free nonpointer type " +
1362 $2->getType()->getName() + "!");
1363 $$ = new FreeInst($2);
1366 | LOAD Types ValueRef UByteList {
1367 if (!(*$2)->isPointerType())
1368 ThrowException("Can't load from nonpointer type: " + (*$2)->getName());
1369 if (LoadInst::getIndexedType(*$2, *$4) == 0)
1370 ThrowException("Invalid indices for load instruction!");
1372 $$ = new LoadInst(getVal(*$2, $3), *$4);
1373 delete $4; // Free the vector...
1376 | STORE ResolvedVal ',' Types ValueRef UByteList {
1377 if (!(*$4)->isPointerType())
1378 ThrowException("Can't store to a nonpointer type: " + (*$4)->getName());
1379 const Type *ElTy = StoreInst::getIndexedType(*$4, *$6);
1381 ThrowException("Can't store into that field list!");
1382 if (ElTy != $2->getType())
1383 ThrowException("Can't store '" + $2->getType()->getName() +
1384 "' into space of type '" + ElTy->getName() + "'!");
1385 $$ = new StoreInst($2, getVal(*$4, $5), *$6);
1386 delete $4; delete $6;
1388 | GETELEMENTPTR Types ValueRef UByteList {
1389 if (!(*$2)->isPointerType())
1390 ThrowException("getelementptr insn requires pointer operand!");
1391 if (!GetElementPtrInst::getIndexedType(*$2, *$4, true))
1392 ThrowException("Can't get element ptr '" + (*$2)->getName() + "'!");
1393 $$ = new GetElementPtrInst(getVal(*$2, $3), *$4);
1394 delete $2; delete $4;
1398 int yyerror(const char *ErrorMsg) {
1399 ThrowException(string("Parse error: ") + ErrorMsg);