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 switch (Ty->getPrimitiveID()) {
299 case Type::LabelTyID: d = new BBPlaceHolder(Ty, D); break;
300 case Type::MethodTyID: d = new MethPlaceHolder(Ty, D);
301 LateResolver = &CurModule.LateResolveValues; break;
302 default: d = new ValuePlaceHolder(Ty, D); break;
305 assert(d != 0 && "How did we not make something?");
306 InsertValue(d, *LateResolver);
311 //===----------------------------------------------------------------------===//
312 // Code to handle forward references in instructions
313 //===----------------------------------------------------------------------===//
315 // This code handles the late binding needed with statements that reference
316 // values not defined yet... for example, a forward branch, or the PHI node for
319 // This keeps a table (CurMeth.LateResolveValues) of all such forward references
320 // and back patchs after we are done.
323 // ResolveDefinitions - If we could not resolve some defs at parsing
324 // time (forward branches, phi functions for loops, etc...) resolve the
327 static void ResolveDefinitions(vector<ValueList> &LateResolvers) {
328 // Loop over LateResolveDefs fixing up stuff that couldn't be resolved
329 for (unsigned ty = 0; ty < LateResolvers.size(); ty++) {
330 while (!LateResolvers[ty].empty()) {
331 Value *V = LateResolvers[ty].back();
332 LateResolvers[ty].pop_back();
333 ValID &DID = getValIDFromPlaceHolder(V);
335 Value *TheRealValue = getVal(Type::getUniqueIDType(ty), DID, true);
337 if (TheRealValue == 0) {
339 ThrowException("Reference to an invalid definition: '" +DID.getName()+
340 "' of type '" + V->getType()->getDescription() + "'",
341 getLineNumFromPlaceHolder(V));
343 ThrowException("Reference to an invalid definition: #" +
344 itostr(DID.Num) + " of type '" +
345 V->getType()->getDescription() + "'",
346 getLineNumFromPlaceHolder(V));
349 assert(!isa<Type>(V) && "Types should be in LateResolveTypes!");
351 V->replaceAllUsesWith(TheRealValue);
356 LateResolvers.clear();
359 // ResolveType - Take a specified unresolved type and resolve it. If there is
360 // nothing to resolve it to yet, return true. Otherwise resolve it and return
363 static bool ResolveType(PATypeHolder<Type> &T) {
365 ValID &DID = getValIDFromPlaceHolder(Ty);
367 const Type *TheRealType = getTypeVal(DID, true);
368 if (TheRealType == 0) return true;
370 // Refine the opaque type we had to the new type we are getting.
371 cast<DerivedType>(Ty)->refineAbstractTypeTo(TheRealType);
376 // ResolveTypes - This goes through the forward referenced type table and makes
377 // sure that all type references are complete. This code is executed after the
378 // constant pool of a method or module is completely parsed.
380 static void ResolveTypes(vector<PATypeHolder<Type> > &LateResolveTypes) {
381 while (!LateResolveTypes.empty()) {
382 if (ResolveType(LateResolveTypes.back())) {
383 const Type *Ty = LateResolveTypes.back();
384 ValID &DID = getValIDFromPlaceHolder(Ty);
386 if (DID.Type == ValID::NameVal)
387 ThrowException("Reference to an invalid type: '" +DID.getName(),
388 getLineNumFromPlaceHolder(Ty));
390 ThrowException("Reference to an invalid type: #" + itostr(DID.Num),
391 getLineNumFromPlaceHolder(Ty));
394 // No need to delete type, refine does that for us.
395 LateResolveTypes.pop_back();
400 // ResolveSomeTypes - This goes through the forward referenced type table and
401 // completes references that are now done. This is so that types are
402 // immediately resolved to be as concrete as possible. This does not cause
403 // thrown exceptions if not everything is resolved.
405 static void ResolveSomeTypes(vector<PATypeHolder<Type> > &LateResolveTypes) {
406 for (unsigned i = 0; i < LateResolveTypes.size(); ) {
407 if (ResolveType(LateResolveTypes[i]))
408 ++i; // Type didn't resolve
410 LateResolveTypes.erase(LateResolveTypes.begin()+i); // Type resolved!
415 // setValueName - Set the specified value to the name given. The name may be
416 // null potentially, in which case this is a noop. The string passed in is
417 // assumed to be a malloc'd string buffer, and is freed by this function.
419 static void setValueName(Value *V, char *NameStr) {
420 if (NameStr == 0) return;
421 string Name(NameStr); // Copy string
422 free(NameStr); // Free old string
424 SymbolTable *ST = CurMeth.CurrentMethod ?
425 CurMeth.CurrentMethod->getSymbolTableSure() :
426 CurModule.CurrentModule->getSymbolTableSure();
428 Value *Existing = ST->lookup(V->getType(), Name);
429 if (Existing) { // Inserting a name that is already defined???
430 // There is only one case where this is allowed: when we are refining an
431 // opaque type. In this case, Existing will be an opaque type.
432 if (const Type *Ty = dyn_cast<const Type>(Existing)) {
433 if (OpaqueType *OpTy = dyn_cast<OpaqueType>(Ty)) {
434 // We ARE replacing an opaque type!
435 OpTy->refineAbstractTypeTo(cast<Type>(V));
440 // Otherwise, we are a simple redefinition of a value, check to see if it
441 // is defined the same as the old one...
442 if (const Type *Ty = dyn_cast<const Type>(Existing)) {
443 if (Ty == cast<const Type>(V)) return; // Yes, it's equal.
444 cerr << "Type: " << Ty->getDescription() << " != "
445 << cast<const Type>(V)->getDescription() << "!\n";
449 ThrowException("Redefinition of value name '" + Name + "' in the '" +
450 V->getType()->getDescription() + "' type plane!");
453 V->setName(Name, ST);
457 //===----------------------------------------------------------------------===//
458 // Code for handling upreferences in type names...
461 // TypeContains - Returns true if Ty contains E in it.
463 static bool TypeContains(const Type *Ty, const Type *E) {
464 return find(df_begin(Ty), df_end(Ty), E) != df_end(Ty);
468 static vector<pair<unsigned, OpaqueType *> > UpRefs;
470 static PATypeHolder<Type> HandleUpRefs(const Type *ty) {
471 PATypeHolder<Type> Ty(ty);
472 UR_OUT(UpRefs.size() << " upreferences active!\n");
473 for (unsigned i = 0; i < UpRefs.size(); ) {
474 UR_OUT("TypeContains(" << Ty->getDescription() << ", "
475 << UpRefs[i].second->getDescription() << ") = "
476 << TypeContains(Ty, UpRefs[i].second) << endl);
477 if (TypeContains(Ty, UpRefs[i].second)) {
478 unsigned Level = --UpRefs[i].first; // Decrement level of upreference
479 UR_OUT("Uplevel Ref Level = " << Level << endl);
480 if (Level == 0) { // Upreference should be resolved!
481 UR_OUT("About to resolve upreference!\n";
482 string OldName = UpRefs[i].second->getDescription());
483 UpRefs[i].second->refineAbstractTypeTo(Ty);
484 UpRefs.erase(UpRefs.begin()+i); // Remove from upreference list...
485 UR_OUT("Type '" << OldName << "' refined upreference to: "
486 << (const void*)Ty << ", " << Ty->getDescription() << endl);
491 ++i; // Otherwise, no resolve, move on...
493 // FIXME: TODO: this should return the updated type
497 template <class TypeTy>
498 inline static void TypeDone(PATypeHolder<TypeTy> *Ty) {
500 ThrowException("Invalid upreference in type: " + (*Ty)->getDescription());
503 // newTH - Allocate a new type holder for the specified type
504 template <class TypeTy>
505 inline static PATypeHolder<TypeTy> *newTH(const TypeTy *Ty) {
506 return new PATypeHolder<TypeTy>(Ty);
508 template <class TypeTy>
509 inline static PATypeHolder<TypeTy> *newTH(const PATypeHolder<TypeTy> &TH) {
510 return new PATypeHolder<TypeTy>(TH);
514 //===----------------------------------------------------------------------===//
515 // RunVMAsmParser - Define an interface to this parser
516 //===----------------------------------------------------------------------===//
518 Module *RunVMAsmParser(const string &Filename, FILE *F) {
520 CurFilename = Filename;
521 llvmAsmlineno = 1; // Reset the current line number...
523 CurModule.CurrentModule = new Module(); // Allocate a new module to read
524 yyparse(); // Parse the file.
525 Module *Result = ParserResult;
526 llvmAsmin = stdin; // F is about to go away, don't use it anymore...
537 MethodArgument *MethArgVal;
538 BasicBlock *BasicBlockVal;
539 TerminatorInst *TermInstVal;
540 Instruction *InstVal;
541 ConstPoolVal *ConstVal;
543 const Type *PrimType;
544 PATypeHolder<Type> *TypeVal;
547 list<MethodArgument*> *MethodArgList;
548 list<Value*> *ValueList;
549 list<PATypeHolder<Type> > *TypeList;
550 list<pair<Value*, BasicBlock*> > *PHIList; // Represent the RHS of PHI node
551 list<pair<ConstPoolVal*, BasicBlock*> > *JumpTable;
552 vector<ConstPoolVal*> *ConstVector;
561 char *StrVal; // This memory is strdup'd!
562 ValID ValIDVal; // strdup'd memory maybe!
564 Instruction::UnaryOps UnaryOpVal;
565 Instruction::BinaryOps BinaryOpVal;
566 Instruction::TermOps TermOpVal;
567 Instruction::MemoryOps MemOpVal;
568 Instruction::OtherOps OtherOpVal;
571 %type <ModuleVal> Module MethodList
572 %type <MethodVal> Method MethodProto MethodHeader BasicBlockList
573 %type <BasicBlockVal> BasicBlock InstructionList
574 %type <TermInstVal> BBTerminatorInst
575 %type <InstVal> Inst InstVal MemoryInst
576 %type <ConstVal> ConstVal
577 %type <ConstVector> ConstVector UByteList
578 %type <MethodArgList> ArgList ArgListH
579 %type <MethArgVal> ArgVal
580 %type <PHIList> PHIList
581 %type <ValueList> ValueRefList ValueRefListE // For call param lists
582 %type <TypeList> TypeListI ArgTypeListI
583 %type <JumpTable> JumpTable
584 %type <BoolVal> GlobalType // GLOBAL or CONSTANT?
586 %type <ValIDVal> ValueRef ConstValueRef // Reference to a definition or BB
587 %type <ValueVal> ResolvedVal // <type> <valref> pair
588 // Tokens and types for handling constant integer values
590 // ESINT64VAL - A negative number within long long range
591 %token <SInt64Val> ESINT64VAL
593 // EUINT64VAL - A positive number within uns. long long range
594 %token <UInt64Val> EUINT64VAL
595 %type <SInt64Val> EINT64VAL
597 %token <SIntVal> SINTVAL // Signed 32 bit ints...
598 %token <UIntVal> UINTVAL // Unsigned 32 bit ints...
599 %type <SIntVal> INTVAL
600 %token <FPVal> FPVAL // Float or Double constant
603 %type <TypeVal> Types TypesV UpRTypes UpRTypesV
604 %type <PrimType> SIntType UIntType IntType FPType PrimType // Classifications
605 %token <TypeVal> OPAQUE
606 %token <PrimType> VOID BOOL SBYTE UBYTE SHORT USHORT INT UINT LONG ULONG
607 %token <PrimType> FLOAT DOUBLE TYPE LABEL
609 %token <StrVal> VAR_ID LABELSTR STRINGCONSTANT
610 %type <StrVal> OptVAR_ID OptAssign
613 %token IMPLEMENTATION TRUE FALSE BEGINTOK END DECLARE GLOBAL CONSTANT UNINIT
614 %token TO DOTDOTDOT STRING NULL_TOK CONST
616 // Basic Block Terminating Operators
617 %token <TermOpVal> RET BR SWITCH
620 %type <UnaryOpVal> UnaryOps // all the unary operators
621 %token <UnaryOpVal> NOT
624 %type <BinaryOpVal> BinaryOps // all the binary operators
625 %token <BinaryOpVal> ADD SUB MUL DIV REM
626 %token <BinaryOpVal> SETLE SETGE SETLT SETGT SETEQ SETNE // Binary Comarators
628 // Memory Instructions
629 %token <MemoryOpVal> MALLOC ALLOCA FREE LOAD STORE GETELEMENTPTR
632 %type <OtherOpVal> ShiftOps
633 %token <OtherOpVal> PHI CALL CAST SHL SHR
638 // Handle constant integer size restriction and conversion...
643 if ($1 > (uint32_t)INT32_MAX) // Outside of my range!
644 ThrowException("Value too large for type!");
649 EINT64VAL : ESINT64VAL // These have same type and can't cause problems...
650 EINT64VAL : EUINT64VAL {
651 if ($1 > (uint64_t)INT64_MAX) // Outside of my range!
652 ThrowException("Value too large for type!");
656 // Operations that are notably excluded from this list include:
657 // RET, BR, & SWITCH because they end basic blocks and are treated specially.
660 BinaryOps : ADD | SUB | MUL | DIV | REM
661 BinaryOps : SETLE | SETGE | SETLT | SETGT | SETEQ | SETNE
664 // These are some types that allow classification if we only want a particular
665 // thing... for example, only a signed, unsigned, or integral type.
666 SIntType : LONG | INT | SHORT | SBYTE
667 UIntType : ULONG | UINT | USHORT | UBYTE
668 IntType : SIntType | UIntType
669 FPType : FLOAT | DOUBLE
671 // OptAssign - Value producing statements have an optional assignment component
672 OptAssign : VAR_ID '=' {
680 //===----------------------------------------------------------------------===//
681 // Types includes all predefined types... except void, because it can only be
682 // used in specific contexts (method returning void for example). To have
683 // access to it, a user must explicitly use TypesV.
686 // TypesV includes all of 'Types', but it also includes the void type.
687 TypesV : Types | VOID { $$ = newTH($1); }
688 UpRTypesV : UpRTypes | VOID { $$ = newTH($1); }
695 // Derived types are added later...
697 PrimType : BOOL | SBYTE | UBYTE | SHORT | USHORT | INT | UINT
698 PrimType : LONG | ULONG | FLOAT | DOUBLE | TYPE | LABEL
699 UpRTypes : OPAQUE | PrimType { $$ = newTH($1); }
700 UpRTypes : ValueRef { // Named types are also simple types...
701 $$ = newTH(getTypeVal($1));
704 // Include derived types in the Types production.
706 UpRTypes : '\\' EUINT64VAL { // Type UpReference
707 if ($2 > (uint64_t)INT64_MAX) ThrowException("Value out of range!");
708 OpaqueType *OT = OpaqueType::get(); // Use temporary placeholder
709 UpRefs.push_back(make_pair((unsigned)$2, OT)); // Add to vector...
710 $$ = newTH<Type>(OT);
711 UR_OUT("New Upreference!\n");
713 | UpRTypesV '(' ArgTypeListI ')' { // Method derived type?
714 vector<const Type*> Params;
715 mapto($3->begin(), $3->end(), back_inserter(Params),
716 mem_fun_ref(&PATypeHandle<Type>::get));
717 $$ = newTH(HandleUpRefs(MethodType::get(*$1, Params)));
718 delete $3; // Delete the argument list
719 delete $1; // Delete the old type handle
721 | '[' UpRTypesV ']' { // Unsized array type?
722 $$ = newTH<Type>(HandleUpRefs(ArrayType::get(*$2)));
725 | '[' EUINT64VAL 'x' UpRTypes ']' { // Sized array type?
726 $$ = newTH<Type>(HandleUpRefs(ArrayType::get(*$4, (int)$2)));
729 | '{' TypeListI '}' { // Structure type?
730 vector<const Type*> Elements;
731 mapto($2->begin(), $2->end(), back_inserter(Elements),
732 mem_fun_ref(&PATypeHandle<Type>::get));
734 $$ = newTH<Type>(HandleUpRefs(StructType::get(Elements)));
737 | '{' '}' { // Empty structure type?
738 $$ = newTH<Type>(StructType::get(vector<const Type*>()));
740 | UpRTypes '*' { // Pointer type?
741 $$ = newTH<Type>(HandleUpRefs(PointerType::get(*$1)));
745 // TypeList - Used for struct declarations and as a basis for method type
746 // declaration type lists
748 TypeListI : UpRTypes {
749 $$ = new list<PATypeHolder<Type> >();
750 $$->push_back(*$1); delete $1;
752 | TypeListI ',' UpRTypes {
753 ($$=$1)->push_back(*$3); delete $3;
756 // ArgTypeList - List of types for a method type declaration...
757 ArgTypeListI : TypeListI
758 | TypeListI ',' DOTDOTDOT {
759 ($$=$1)->push_back(Type::VoidTy);
762 ($$ = new list<PATypeHolder<Type> >())->push_back(Type::VoidTy);
765 $$ = new list<PATypeHolder<Type> >();
769 // ConstVal - The various declarations that go into the constant pool. This
770 // includes all forward declarations of types, constants, and functions.
772 ConstVal: Types '[' ConstVector ']' { // Nonempty unsized arr
773 const ArrayType *ATy = dyn_cast<const ArrayType>($1->get());
775 ThrowException("Cannot make array constant with type: '" +
776 (*$1)->getDescription() + "'!");
777 const Type *ETy = ATy->getElementType();
778 int NumElements = ATy->getNumElements();
780 // Verify that we have the correct size...
781 if (NumElements != -1 && NumElements != (int)$3->size())
782 ThrowException("Type mismatch: constant sized array initialized with " +
783 utostr($3->size()) + " arguments, but has size of " +
784 itostr(NumElements) + "!");
786 // Verify all elements are correct type!
787 for (unsigned i = 0; i < $3->size(); i++) {
788 if (ETy != (*$3)[i]->getType())
789 ThrowException("Element #" + utostr(i) + " is not of type '" +
790 ETy->getName() + "' as required!\nIt is of type '" +
791 (*$3)[i]->getType()->getName() + "'.");
794 $$ = ConstPoolArray::get(ATy, *$3);
795 delete $1; delete $3;
798 const ArrayType *ATy = dyn_cast<const ArrayType>($1->get());
800 ThrowException("Cannot make array constant with type: '" +
801 (*$1)->getDescription() + "'!");
803 int NumElements = ATy->getNumElements();
804 if (NumElements != -1 && NumElements != 0)
805 ThrowException("Type mismatch: constant sized array initialized with 0"
806 " arguments, but has size of " + itostr(NumElements) +"!");
807 $$ = ConstPoolArray::get(ATy, vector<ConstPoolVal*>());
810 | Types 'c' STRINGCONSTANT {
811 const ArrayType *ATy = dyn_cast<const ArrayType>($1->get());
813 ThrowException("Cannot make array constant with type: '" +
814 (*$1)->getDescription() + "'!");
816 int NumElements = ATy->getNumElements();
817 const Type *ETy = ATy->getElementType();
818 char *EndStr = UnEscapeLexed($3, true);
819 if (NumElements != -1 && NumElements != (EndStr-$3))
820 ThrowException("Can't build string constant of size " +
821 itostr((int)(EndStr-$3)) +
822 " when array has size " + itostr(NumElements) + "!");
823 vector<ConstPoolVal*> Vals;
824 if (ETy == Type::SByteTy) {
825 for (char *C = $3; C != EndStr; ++C)
826 Vals.push_back(ConstPoolSInt::get(ETy, *C));
827 } else if (ETy == Type::UByteTy) {
828 for (char *C = $3; C != EndStr; ++C)
829 Vals.push_back(ConstPoolUInt::get(ETy, *C));
832 ThrowException("Cannot build string arrays of non byte sized elements!");
835 $$ = ConstPoolArray::get(ATy, Vals);
838 | Types '{' ConstVector '}' {
839 const StructType *STy = dyn_cast<const StructType>($1->get());
841 ThrowException("Cannot make struct constant with type: '" +
842 (*$1)->getDescription() + "'!");
843 // FIXME: TODO: Check to see that the constants are compatible with the type
845 $$ = ConstPoolStruct::get(STy, *$3);
846 delete $1; delete $3;
849 const PointerType *PTy = dyn_cast<const PointerType>($1->get());
851 ThrowException("Cannot make null pointer constant with type: '" +
852 (*$1)->getDescription() + "'!");
854 $$ = ConstPoolPointer::getNullPointer(PTy);
858 string Name($2); free($2); // Change to a responsible mem manager
859 const PointerType *Ty = dyn_cast<const PointerType>($1->get());
861 ThrowException("Global const reference must be a pointer type!");
863 Value *N = lookupInSymbolTable(Ty, Name);
865 ThrowException("Global pointer reference '%" + Name +
866 "' must be defined before use!");
868 // TODO FIXME: This should also allow methods... when common baseclass
870 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(N)) {
871 $$ = ConstPoolPointerReference::get(GV);
873 ThrowException("'%" + Name + "' is not a global value reference!");
880 ConstVal : SIntType EINT64VAL { // integral constants
881 if (!ConstPoolSInt::isValueValidForType($1, $2))
882 ThrowException("Constant value doesn't fit in type!");
883 $$ = ConstPoolSInt::get($1, $2);
885 | UIntType EUINT64VAL { // integral constants
886 if (!ConstPoolUInt::isValueValidForType($1, $2))
887 ThrowException("Constant value doesn't fit in type!");
888 $$ = ConstPoolUInt::get($1, $2);
890 | BOOL TRUE { // Boolean constants
891 $$ = ConstPoolBool::True;
893 | BOOL FALSE { // Boolean constants
894 $$ = ConstPoolBool::False;
896 | FPType FPVAL { // Float & Double constants
897 $$ = ConstPoolFP::get($1, $2);
900 // ConstVector - A list of comma seperated constants.
901 ConstVector : ConstVector ',' ConstVal {
902 ($$ = $1)->push_back($3);
905 $$ = new vector<ConstPoolVal*>();
910 // GlobalType - Match either GLOBAL or CONSTANT for global declarations...
911 GlobalType : GLOBAL { $$ = false; } | CONSTANT { $$ = true; }
914 // ConstPool - Constants with optional names assigned to them.
915 ConstPool : ConstPool OptAssign CONST ConstVal {
916 setValueName($4, $2);
919 | ConstPool OptAssign TYPE TypesV { // Types can be defined in the const pool
920 // TODO: FIXME when Type are not const
921 setValueName(const_cast<Type*>($4->get()), $2);
924 InsertType($4->get(),
925 CurMeth.CurrentMethod ? CurMeth.Types : CurModule.Types);
929 ResolveSomeTypes(CurMeth.CurrentMethod ? CurMeth.LateResolveTypes :
930 CurModule.LateResolveTypes);
932 | ConstPool MethodProto { // Method prototypes can be in const pool
934 | ConstPool OptAssign GlobalType ConstVal {
935 const Type *Ty = $4->getType();
936 // Global declarations appear in Constant Pool
937 ConstPoolVal *Initializer = $4;
938 if (Initializer == 0)
939 ThrowException("Global value initializer is not a constant!");
941 GlobalVariable *GV = new GlobalVariable(PointerType::get(Ty), $3,
943 setValueName(GV, $2);
945 CurModule.CurrentModule->getGlobalList().push_back(GV);
946 InsertValue(GV, CurModule.Values);
948 | ConstPool OptAssign UNINIT GlobalType Types {
949 const Type *Ty = *$5;
950 // Global declarations appear in Constant Pool
951 if (isa<ArrayType>(Ty) && cast<ArrayType>(Ty)->isUnsized()) {
952 ThrowException("Type '" + Ty->getDescription() +
953 "' is not a sized type!");
956 GlobalVariable *GV = new GlobalVariable(PointerType::get(Ty), $4);
957 setValueName(GV, $2);
959 CurModule.CurrentModule->getGlobalList().push_back(GV);
960 InsertValue(GV, CurModule.Values);
962 | /* empty: end of list */ {
966 //===----------------------------------------------------------------------===//
967 // Rules to match Modules
968 //===----------------------------------------------------------------------===//
970 // Module rule: Capture the result of parsing the whole file into a result
973 Module : MethodList {
974 $$ = ParserResult = $1;
975 CurModule.ModuleDone();
978 // MethodList - A list of methods, preceeded by a constant pool.
980 MethodList : MethodList Method {
982 if (!$2->getParent())
983 $1->getMethodList().push_back($2);
984 CurMeth.MethodDone();
986 | MethodList MethodProto {
989 | ConstPool IMPLEMENTATION {
990 $$ = CurModule.CurrentModule;
991 // Resolve circular types before we parse the body of the module
992 ResolveTypes(CurModule.LateResolveTypes);
996 //===----------------------------------------------------------------------===//
997 // Rules to match Method Headers
998 //===----------------------------------------------------------------------===//
1000 OptVAR_ID : VAR_ID | /*empty*/ { $$ = 0; }
1002 ArgVal : Types OptVAR_ID {
1003 $$ = new MethodArgument(*$1); delete $1;
1004 setValueName($$, $2);
1007 ArgListH : ArgVal ',' ArgListH {
1012 $$ = new list<MethodArgument*>();
1016 $$ = new list<MethodArgument*>();
1017 $$->push_back(new MethodArgument(Type::VoidTy));
1020 ArgList : ArgListH {
1027 MethodHeaderH : TypesV STRINGCONSTANT '(' ArgList ')' {
1029 vector<const Type*> ParamTypeList;
1031 for (list<MethodArgument*>::iterator I = $4->begin(); I != $4->end(); ++I)
1032 ParamTypeList.push_back((*I)->getType());
1034 const MethodType *MT = MethodType::get(*$1, ParamTypeList);
1038 if (SymbolTable *ST = CurModule.CurrentModule->getSymbolTable()) {
1039 if (Value *V = ST->lookup(MT, $2)) { // Method already in symtab?
1040 M = cast<Method>(V);
1042 // Yes it is. If this is the case, either we need to be a forward decl,
1043 // or it needs to be.
1044 if (!CurMeth.isDeclare && !M->isExternal())
1045 ThrowException("Redefinition of method '" + string($2) + "'!");
1049 if (M == 0) { // Not already defined?
1050 M = new Method(MT, $2);
1051 InsertValue(M, CurModule.Values);
1054 free($2); // Free strdup'd memory!
1056 CurMeth.MethodStart(M);
1058 // Add all of the arguments we parsed to the method...
1059 if ($4 && !CurMeth.isDeclare) { // Is null if empty...
1060 Method::ArgumentListType &ArgList = M->getArgumentList();
1062 for (list<MethodArgument*>::iterator I = $4->begin(); I != $4->end(); ++I) {
1064 ArgList.push_back(*I);
1066 delete $4; // We're now done with the argument list
1070 MethodHeader : MethodHeaderH ConstPool BEGINTOK {
1071 $$ = CurMeth.CurrentMethod;
1073 // Resolve circular types before we parse the body of the method.
1074 ResolveTypes(CurMeth.LateResolveTypes);
1077 Method : BasicBlockList END {
1081 MethodProto : DECLARE { CurMeth.isDeclare = true; } MethodHeaderH {
1082 $$ = CurMeth.CurrentMethod;
1083 if (!$$->getParent())
1084 CurModule.CurrentModule->getMethodList().push_back($$);
1085 CurMeth.MethodDone();
1088 //===----------------------------------------------------------------------===//
1089 // Rules to match Basic Blocks
1090 //===----------------------------------------------------------------------===//
1092 ConstValueRef : ESINT64VAL { // A reference to a direct constant
1093 $$ = ValID::create($1);
1096 $$ = ValID::create($1);
1098 | FPVAL { // Perhaps it's an FP constant?
1099 $$ = ValID::create($1);
1102 $$ = ValID::create((int64_t)1);
1105 $$ = ValID::create((int64_t)0);
1108 $$ = ValID::createNull();
1112 | STRINGCONSTANT { // Quoted strings work too... especially for methods
1113 $$ = ValID::create_conststr($1);
1117 // ValueRef - A reference to a definition...
1118 ValueRef : INTVAL { // Is it an integer reference...?
1119 $$ = ValID::create($1);
1121 | VAR_ID { // Is it a named reference...?
1122 $$ = ValID::create($1);
1128 // ResolvedVal - a <type> <value> pair. This is used only in cases where the
1129 // type immediately preceeds the value reference, and allows complex constant
1130 // pool references (for things like: 'ret [2 x int] [ int 12, int 42]')
1131 ResolvedVal : Types ValueRef {
1132 $$ = getVal(*$1, $2); delete $1;
1136 BasicBlockList : BasicBlockList BasicBlock {
1137 $1->getBasicBlocks().push_back($2);
1140 | MethodHeader BasicBlock { // Do not allow methods with 0 basic blocks
1141 $$ = $1; // in them...
1142 $1->getBasicBlocks().push_back($2);
1146 // Basic blocks are terminated by branching instructions:
1147 // br, br/cc, switch, ret
1149 BasicBlock : InstructionList BBTerminatorInst {
1150 $1->getInstList().push_back($2);
1154 | LABELSTR InstructionList BBTerminatorInst {
1155 $2->getInstList().push_back($3);
1156 setValueName($2, $1);
1162 InstructionList : InstructionList Inst {
1163 $1->getInstList().push_back($2);
1167 $$ = new BasicBlock();
1170 BBTerminatorInst : RET ResolvedVal { // Return with a result...
1171 $$ = new ReturnInst($2);
1173 | RET VOID { // Return with no result...
1174 $$ = new ReturnInst();
1176 | BR LABEL ValueRef { // Unconditional Branch...
1177 $$ = new BranchInst(cast<BasicBlock>(getVal(Type::LabelTy, $3)));
1178 } // Conditional Branch...
1179 | BR BOOL ValueRef ',' LABEL ValueRef ',' LABEL ValueRef {
1180 $$ = new BranchInst(cast<BasicBlock>(getVal(Type::LabelTy, $6)),
1181 cast<BasicBlock>(getVal(Type::LabelTy, $9)),
1182 getVal(Type::BoolTy, $3));
1184 | SWITCH IntType ValueRef ',' LABEL ValueRef '[' JumpTable ']' {
1185 SwitchInst *S = new SwitchInst(getVal($2, $3),
1186 cast<BasicBlock>(getVal(Type::LabelTy, $6)));
1189 list<pair<ConstPoolVal*, BasicBlock*> >::iterator I = $8->begin(),
1191 for (; I != end; ++I)
1192 S->dest_push_back(I->first, I->second);
1195 JumpTable : JumpTable IntType ConstValueRef ',' LABEL ValueRef {
1197 ConstPoolVal *V = cast<ConstPoolVal>(getVal($2, $3, true));
1199 ThrowException("May only switch on a constant pool value!");
1201 $$->push_back(make_pair(V, cast<BasicBlock>(getVal($5, $6))));
1203 | IntType ConstValueRef ',' LABEL ValueRef {
1204 $$ = new list<pair<ConstPoolVal*, BasicBlock*> >();
1205 ConstPoolVal *V = cast<ConstPoolVal>(getVal($1, $2, true));
1208 ThrowException("May only switch on a constant pool value!");
1210 $$->push_back(make_pair(V, cast<BasicBlock>(getVal($4, $5))));
1213 Inst : OptAssign InstVal {
1214 setValueName($2, $1); // Is this definition named?? if so, assign the name...
1220 PHIList : Types '[' ValueRef ',' ValueRef ']' { // Used for PHI nodes
1221 $$ = new list<pair<Value*, BasicBlock*> >();
1222 $$->push_back(make_pair(getVal(*$1, $3),
1223 cast<BasicBlock>(getVal(Type::LabelTy, $5))));
1226 | PHIList ',' '[' ValueRef ',' ValueRef ']' {
1228 $1->push_back(make_pair(getVal($1->front().first->getType(), $4),
1229 cast<BasicBlock>(getVal(Type::LabelTy, $6))));
1233 ValueRefList : ResolvedVal { // Used for call statements, and memory insts...
1234 $$ = new list<Value*>();
1237 | ValueRefList ',' ResolvedVal {
1242 // ValueRefListE - Just like ValueRefList, except that it may also be empty!
1243 ValueRefListE : ValueRefList | /*empty*/ { $$ = 0; }
1245 InstVal : BinaryOps Types ValueRef ',' ValueRef {
1246 $$ = BinaryOperator::create($1, getVal(*$2, $3), getVal(*$2, $5));
1248 ThrowException("binary operator returned null!");
1251 | UnaryOps ResolvedVal {
1252 $$ = UnaryOperator::create($1, $2);
1254 ThrowException("unary operator returned null!");
1256 | ShiftOps ResolvedVal ',' ResolvedVal {
1257 if ($4->getType() != Type::UByteTy)
1258 ThrowException("Shift amount must be ubyte!");
1259 $$ = new ShiftInst($1, $2, $4);
1261 | CAST ResolvedVal TO Types {
1262 $$ = new CastInst($2, *$4);
1266 const Type *Ty = $2->front().first->getType();
1267 $$ = new PHINode(Ty);
1268 while ($2->begin() != $2->end()) {
1269 if ($2->front().first->getType() != Ty)
1270 ThrowException("All elements of a PHI node must be of the same type!");
1271 cast<PHINode>($$)->addIncoming($2->front().first, $2->front().second);
1274 delete $2; // Free the list...
1276 | CALL TypesV ValueRef '(' ValueRefListE ')' {
1277 const MethodType *Ty;
1279 if (!(Ty = dyn_cast<MethodType>($2->get()))) {
1280 // Pull out the types of all of the arguments...
1281 vector<const Type*> ParamTypes;
1282 for (list<Value*>::iterator I = $5->begin(), E = $5->end(); I != E; ++I)
1283 ParamTypes.push_back((*I)->getType());
1284 Ty = MethodType::get(*$2, ParamTypes);
1288 Value *V = getVal(Ty, $3); // Get the method we're calling...
1290 // Create the call node...
1291 if (!$5) { // Has no arguments?
1292 $$ = new CallInst(cast<Method>(V), vector<Value*>());
1293 } else { // Has arguments?
1294 // Loop through MethodType's arguments and ensure they are specified
1297 MethodType::ParamTypes::const_iterator I = Ty->getParamTypes().begin();
1298 MethodType::ParamTypes::const_iterator E = Ty->getParamTypes().end();
1299 list<Value*>::iterator ArgI = $5->begin(), ArgE = $5->end();
1301 for (; ArgI != ArgE && I != E; ++ArgI, ++I)
1302 if ((*ArgI)->getType() != *I)
1303 ThrowException("Parameter " +(*ArgI)->getName()+ " is not of type '" +
1304 (*I)->getName() + "'!");
1306 if (I != E || (ArgI != ArgE && !Ty->isVarArg()))
1307 ThrowException("Invalid number of parameters detected!");
1309 $$ = new CallInst(cast<Method>(V),
1310 vector<Value*>($5->begin(), $5->end()));
1318 // UByteList - List of ubyte values for load and store instructions
1319 UByteList : ',' ConstVector {
1322 $$ = new vector<ConstPoolVal*>();
1325 MemoryInst : MALLOC Types {
1326 $$ = new MallocInst(PointerType::get(*$2));
1329 | MALLOC Types ',' UINT ValueRef {
1330 if (!(*$2)->isArrayType() || cast<const ArrayType>($2->get())->isSized())
1331 ThrowException("Trying to allocate " + (*$2)->getName() +
1332 " as unsized array!");
1333 const Type *Ty = PointerType::get(*$2);
1334 $$ = new MallocInst(Ty, getVal($4, $5));
1338 $$ = new AllocaInst(PointerType::get(*$2));
1341 | ALLOCA Types ',' UINT ValueRef {
1342 if (!(*$2)->isArrayType() || cast<const ArrayType>($2->get())->isSized())
1343 ThrowException("Trying to allocate " + (*$2)->getName() +
1344 " as unsized array!");
1345 const Type *Ty = PointerType::get(*$2);
1346 Value *ArrSize = getVal($4, $5);
1347 $$ = new AllocaInst(Ty, ArrSize);
1350 | FREE ResolvedVal {
1351 if (!$2->getType()->isPointerType())
1352 ThrowException("Trying to free nonpointer type " +
1353 $2->getType()->getName() + "!");
1354 $$ = new FreeInst($2);
1357 | LOAD Types ValueRef UByteList {
1358 if (!(*$2)->isPointerType())
1359 ThrowException("Can't load from nonpointer type: " + (*$2)->getName());
1360 if (LoadInst::getIndexedType(*$2, *$4) == 0)
1361 ThrowException("Invalid indices for load instruction!");
1363 $$ = new LoadInst(getVal(*$2, $3), *$4);
1364 delete $4; // Free the vector...
1367 | STORE ResolvedVal ',' Types ValueRef UByteList {
1368 if (!(*$4)->isPointerType())
1369 ThrowException("Can't store to a nonpointer type: " + (*$4)->getName());
1370 const Type *ElTy = StoreInst::getIndexedType(*$4, *$6);
1372 ThrowException("Can't store into that field list!");
1373 if (ElTy != $2->getType())
1374 ThrowException("Can't store '" + $2->getType()->getName() +
1375 "' into space of type '" + ElTy->getName() + "'!");
1376 $$ = new StoreInst($2, getVal(*$4, $5), *$6);
1377 delete $4; delete $6;
1379 | GETELEMENTPTR Types ValueRef UByteList {
1380 if (!(*$2)->isPointerType())
1381 ThrowException("getelementptr insn requires pointer operand!");
1382 if (!GetElementPtrInst::getIndexedType(*$2, *$4, true))
1383 ThrowException("Can't get element ptr '" + (*$2)->getName() + "'!");
1384 $$ = new GetElementPtrInst(getVal(*$2, $3), *$4);
1385 delete $2; delete $4;
1389 int yyerror(const char *ErrorMsg) {
1390 ThrowException(string("Parse error: ") + ErrorMsg);