1 //===-- Writer.cpp - Library for converting LLVM code to C ----------------===//
3 // This library converts LLVM code to C code, compilable by GCC.
5 //===----------------------------------------------------------------------===//
7 #include "llvm/Assembly/CWriter.h"
8 #include "llvm/Constants.h"
9 #include "llvm/DerivedTypes.h"
10 #include "llvm/Module.h"
11 #include "llvm/Instructions.h"
12 #include "llvm/Pass.h"
13 #include "llvm/SymbolTable.h"
14 #include "llvm/Intrinsics.h"
15 #include "llvm/Analysis/FindUsedTypes.h"
16 #include "llvm/Analysis/ConstantsScanner.h"
17 #include "llvm/Support/InstVisitor.h"
18 #include "llvm/Support/InstIterator.h"
19 #include "llvm/Support/CallSite.h"
20 #include "llvm/Support/Mangler.h"
21 #include "Support/StringExtras.h"
22 #include "Support/STLExtras.h"
27 class CWriter : public Pass, public InstVisitor<CWriter> {
30 const Module *TheModule;
31 std::map<const Type *, std::string> TypeNames;
32 std::set<const Value*> MangledGlobals;
33 bool needsMalloc, emittedInvoke;
35 std::map<const ConstantFP *, unsigned> FPConstantMap;
37 CWriter(std::ostream &o) : Out(o) {}
39 void getAnalysisUsage(AnalysisUsage &AU) const {
41 AU.addRequired<FindUsedTypes>();
44 virtual bool run(Module &M) {
48 // Ensure that all structure types have names...
49 bool Changed = nameAllUsedStructureTypes(M);
50 Mang = new Mangler(M);
58 MangledGlobals.clear();
62 std::ostream &printType(std::ostream &Out, const Type *Ty,
63 const std::string &VariableName = "",
64 bool IgnoreName = false, bool namedContext = true);
66 void writeOperand(Value *Operand);
67 void writeOperandInternal(Value *Operand);
70 bool nameAllUsedStructureTypes(Module &M);
71 void printModule(Module *M);
72 void printSymbolTable(const SymbolTable &ST);
73 void printContainedStructs(const Type *Ty, std::set<const StructType *> &);
74 void printFunctionSignature(const Function *F, bool Prototype);
76 void printFunction(Function *);
78 void printConstant(Constant *CPV);
79 void printConstantArray(ConstantArray *CPA);
81 // isInlinableInst - Attempt to inline instructions into their uses to build
82 // trees as much as possible. To do this, we have to consistently decide
83 // what is acceptable to inline, so that variable declarations don't get
84 // printed and an extra copy of the expr is not emitted.
86 static bool isInlinableInst(const Instruction &I) {
87 // Must be an expression, must be used exactly once. If it is dead, we
88 // emit it inline where it would go.
89 if (I.getType() == Type::VoidTy || I.use_size() != 1 ||
90 isa<TerminatorInst>(I) || isa<CallInst>(I) || isa<PHINode>(I) ||
91 isa<LoadInst>(I) || isa<VarArgInst>(I))
92 // Don't inline a load across a store or other bad things!
95 // Only inline instruction it it's use is in the same BB as the inst.
96 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
99 // isDirectAlloca - Define fixed sized allocas in the entry block as direct
100 // variables which are accessed with the & operator. This causes GCC to
101 // generate significantly better code than to emit alloca calls directly.
103 static const AllocaInst *isDirectAlloca(const Value *V) {
104 const AllocaInst *AI = dyn_cast<AllocaInst>(V);
105 if (!AI) return false;
106 if (AI->isArrayAllocation())
107 return 0; // FIXME: we can also inline fixed size array allocas!
108 if (AI->getParent() != &AI->getParent()->getParent()->getEntryNode())
113 // Instruction visitation functions
114 friend class InstVisitor<CWriter>;
116 void visitReturnInst(ReturnInst &I);
117 void visitBranchInst(BranchInst &I);
118 void visitSwitchInst(SwitchInst &I);
119 void visitInvokeInst(InvokeInst &I);
121 void visitPHINode(PHINode &I);
122 void visitBinaryOperator(Instruction &I);
124 void visitCastInst (CastInst &I);
125 void visitCallInst (CallInst &I);
126 void visitCallSite (CallSite CS);
127 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
129 void visitMallocInst(MallocInst &I);
130 void visitAllocaInst(AllocaInst &I);
131 void visitFreeInst (FreeInst &I);
132 void visitLoadInst (LoadInst &I);
133 void visitStoreInst (StoreInst &I);
134 void visitGetElementPtrInst(GetElementPtrInst &I);
135 void visitVarArgInst(VarArgInst &I);
137 void visitInstruction(Instruction &I) {
138 std::cerr << "C Writer does not know about " << I;
142 void outputLValue(Instruction *I) {
143 Out << " " << Mang->getValueName(I) << " = ";
145 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
147 void printIndexingExpression(Value *Ptr, User::op_iterator I,
148 User::op_iterator E);
152 // A pointer type should not use parens around *'s alone, e.g., (**)
153 inline bool ptrTypeNameNeedsParens(const std::string &NameSoFar) {
154 return (NameSoFar.find_last_not_of('*') != std::string::npos);
157 // Pass the Type* and the variable name and this prints out the variable
160 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
161 const std::string &NameSoFar,
162 bool IgnoreName, bool namedContext) {
163 if (Ty->isPrimitiveType())
164 switch (Ty->getPrimitiveID()) {
165 case Type::VoidTyID: return Out << "void " << NameSoFar;
166 case Type::BoolTyID: return Out << "bool " << NameSoFar;
167 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
168 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
169 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
170 case Type::ShortTyID: return Out << "short " << NameSoFar;
171 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
172 case Type::IntTyID: return Out << "int " << NameSoFar;
173 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
174 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
175 case Type::FloatTyID: return Out << "float " << NameSoFar;
176 case Type::DoubleTyID: return Out << "double " << NameSoFar;
178 std::cerr << "Unknown primitive type: " << Ty << "\n";
182 // Check to see if the type is named.
183 if (!IgnoreName || isa<OpaqueType>(Ty)) {
184 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
185 if (I != TypeNames.end()) {
186 return Out << I->second << " " << NameSoFar;
190 switch (Ty->getPrimitiveID()) {
191 case Type::FunctionTyID: {
192 const FunctionType *MTy = cast<FunctionType>(Ty);
193 std::stringstream FunctionInnards;
194 FunctionInnards << " (" << NameSoFar << ") (";
195 for (FunctionType::ParamTypes::const_iterator
196 I = MTy->getParamTypes().begin(),
197 E = MTy->getParamTypes().end(); I != E; ++I) {
198 if (I != MTy->getParamTypes().begin())
199 FunctionInnards << ", ";
200 printType(FunctionInnards, *I, "");
202 if (MTy->isVarArg()) {
203 if (!MTy->getParamTypes().empty())
204 FunctionInnards << ", ...";
205 } else if (MTy->getParamTypes().empty()) {
206 FunctionInnards << "void";
208 FunctionInnards << ")";
209 std::string tstr = FunctionInnards.str();
210 printType(Out, MTy->getReturnType(), tstr);
213 case Type::StructTyID: {
214 const StructType *STy = cast<StructType>(Ty);
215 Out << NameSoFar + " {\n";
217 for (StructType::ElementTypes::const_iterator
218 I = STy->getElementTypes().begin(),
219 E = STy->getElementTypes().end(); I != E; ++I) {
221 printType(Out, *I, "field" + utostr(Idx++));
227 case Type::PointerTyID: {
228 const PointerType *PTy = cast<PointerType>(Ty);
229 std::string ptrName = "*" + NameSoFar;
231 // Do not need parens around "* NameSoFar" if NameSoFar consists only
232 // of zero or more '*' chars *and* this is not an unnamed pointer type
233 // such as the result type in a cast statement. Otherwise, enclose in ( ).
234 if (ptrTypeNameNeedsParens(NameSoFar) || !namedContext ||
235 PTy->getElementType()->getPrimitiveID() == Type::ArrayTyID)
236 ptrName = "(" + ptrName + ")"; //
238 return printType(Out, PTy->getElementType(), ptrName);
241 case Type::ArrayTyID: {
242 const ArrayType *ATy = cast<ArrayType>(Ty);
243 unsigned NumElements = ATy->getNumElements();
244 return printType(Out, ATy->getElementType(),
245 NameSoFar + "[" + utostr(NumElements) + "]");
248 case Type::OpaqueTyID: {
249 static int Count = 0;
250 std::string TyName = "struct opaque_" + itostr(Count++);
251 assert(TypeNames.find(Ty) == TypeNames.end());
252 TypeNames[Ty] = TyName;
253 return Out << TyName << " " << NameSoFar;
256 assert(0 && "Unhandled case in getTypeProps!");
263 void CWriter::printConstantArray(ConstantArray *CPA) {
265 // As a special case, print the array as a string if it is an array of
266 // ubytes or an array of sbytes with positive values.
268 const Type *ETy = CPA->getType()->getElementType();
269 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
271 // Make sure the last character is a null char, as automatically added by C
272 if (isString && (CPA->getNumOperands() == 0 ||
273 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
278 // Keep track of whether the last number was a hexadecimal escape
279 bool LastWasHex = false;
281 // Do not include the last character, which we know is null
282 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
283 unsigned char C = cast<ConstantInt>(CPA->getOperand(i))->getRawValue();
285 // Print it out literally if it is a printable character. The only thing
286 // to be careful about is when the last letter output was a hex escape
287 // code, in which case we have to be careful not to print out hex digits
288 // explicitly (the C compiler thinks it is a continuation of the previous
289 // character, sheesh...)
291 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
293 if (C == '"' || C == '\\')
300 case '\n': Out << "\\n"; break;
301 case '\t': Out << "\\t"; break;
302 case '\r': Out << "\\r"; break;
303 case '\v': Out << "\\v"; break;
304 case '\a': Out << "\\a"; break;
305 case '\"': Out << "\\\""; break;
306 case '\'': Out << "\\\'"; break;
309 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
310 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
319 if (CPA->getNumOperands()) {
321 printConstant(cast<Constant>(CPA->getOperand(0)));
322 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
324 printConstant(cast<Constant>(CPA->getOperand(i)));
331 /// FPCSafeToPrint - Returns true if we may assume that CFP may be
332 /// written out textually as a double (rather than as a reference to a
333 /// stack-allocated variable). We decide this by converting CFP to a
334 /// string and back into a double, and then checking whether the
335 /// conversion results in a bit-equal double to the original value of
336 /// CFP. This depends on us and the target C compiler agreeing on the
337 /// conversion process (which is pretty likely since we only deal in
338 /// IEEE FP.) This is adapted from similar code in
339 /// lib/VMCore/AsmWriter.cpp:WriteConstantInt().
340 static bool FPCSafeToPrint (const ConstantFP *CFP) {
341 std::string StrVal = ftostr(CFP->getValue());
342 // Check to make sure that the stringized number is not some string like
343 // "Inf" or NaN, that atof will accept, but the lexer will not. Check that
344 // the string matches the "[-+]?[0-9]" regex.
345 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
346 ((StrVal[0] == '-' || StrVal[0] == '+') &&
347 (StrVal[1] >= '0' && StrVal[1] <= '9')))
348 // Reparse stringized version!
349 return (atof(StrVal.c_str()) == CFP->getValue());
353 // printConstant - The LLVM Constant to C Constant converter.
354 void CWriter::printConstant(Constant *CPV) {
355 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
356 switch (CE->getOpcode()) {
357 case Instruction::Cast:
359 printType(Out, CPV->getType());
361 printConstant(CE->getOperand(0));
365 case Instruction::GetElementPtr:
367 printIndexingExpression(CE->getOperand(0),
368 CPV->op_begin()+1, CPV->op_end());
371 case Instruction::Add:
372 case Instruction::Sub:
373 case Instruction::Mul:
374 case Instruction::Div:
375 case Instruction::Rem:
376 case Instruction::SetEQ:
377 case Instruction::SetNE:
378 case Instruction::SetLT:
379 case Instruction::SetLE:
380 case Instruction::SetGT:
381 case Instruction::SetGE:
383 printConstant(CE->getOperand(0));
384 switch (CE->getOpcode()) {
385 case Instruction::Add: Out << " + "; break;
386 case Instruction::Sub: Out << " - "; break;
387 case Instruction::Mul: Out << " * "; break;
388 case Instruction::Div: Out << " / "; break;
389 case Instruction::Rem: Out << " % "; break;
390 case Instruction::SetEQ: Out << " == "; break;
391 case Instruction::SetNE: Out << " != "; break;
392 case Instruction::SetLT: Out << " < "; break;
393 case Instruction::SetLE: Out << " <= "; break;
394 case Instruction::SetGT: Out << " > "; break;
395 case Instruction::SetGE: Out << " >= "; break;
396 default: assert(0 && "Illegal opcode here!");
398 printConstant(CE->getOperand(1));
403 std::cerr << "CWriter Error: Unhandled constant expression: "
409 switch (CPV->getType()->getPrimitiveID()) {
411 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
412 case Type::SByteTyID:
413 case Type::ShortTyID:
414 Out << cast<ConstantSInt>(CPV)->getValue(); break;
416 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
417 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
419 Out << cast<ConstantSInt>(CPV)->getValue();
423 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
425 case Type::UByteTyID:
426 case Type::UShortTyID:
427 Out << cast<ConstantUInt>(CPV)->getValue(); break;
429 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
430 case Type::ULongTyID:
431 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
433 case Type::FloatTyID:
434 case Type::DoubleTyID: {
435 ConstantFP *FPC = cast<ConstantFP>(CPV);
436 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
437 if (I != FPConstantMap.end()) {
438 // Because of FP precision problems we must load from a stack allocated
439 // value that holds the value in hex.
440 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
441 << "*)&FloatConstant" << I->second << ")";
443 if (FPCSafeToPrint (FPC)) {
444 Out << ftostr (FPC->getValue ());
446 Out << FPC->getValue(); // Who knows? Give it our best shot...
452 case Type::ArrayTyID:
453 printConstantArray(cast<ConstantArray>(CPV));
456 case Type::StructTyID: {
458 if (CPV->getNumOperands()) {
460 printConstant(cast<Constant>(CPV->getOperand(0)));
461 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
463 printConstant(cast<Constant>(CPV->getOperand(i)));
470 case Type::PointerTyID:
471 if (isa<ConstantPointerNull>(CPV)) {
473 printType(Out, CPV->getType());
474 Out << ")/*NULL*/0)";
476 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
477 writeOperand(CPR->getValue());
482 std::cerr << "Unknown constant type: " << CPV << "\n";
487 void CWriter::writeOperandInternal(Value *Operand) {
488 if (Instruction *I = dyn_cast<Instruction>(Operand))
489 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
490 // Should we inline this instruction to build a tree?
497 if (Constant *CPV = dyn_cast<Constant>(Operand)) {
500 Out << Mang->getValueName(Operand);
504 void CWriter::writeOperand(Value *Operand) {
505 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
506 Out << "(&"; // Global variables are references as their addresses by llvm
508 writeOperandInternal(Operand);
510 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
514 // nameAllUsedStructureTypes - If there are structure types in the module that
515 // are used but do not have names assigned to them in the symbol table yet then
516 // we assign them names now.
518 bool CWriter::nameAllUsedStructureTypes(Module &M) {
519 // Get a set of types that are used by the program...
520 std::set<const Type *> UT = getAnalysis<FindUsedTypes>().getTypes();
522 // Loop over the module symbol table, removing types from UT that are already
525 SymbolTable &MST = M.getSymbolTable();
526 if (MST.find(Type::TypeTy) != MST.end())
527 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
528 E = MST.type_end(Type::TypeTy); I != E; ++I)
529 UT.erase(cast<Type>(I->second));
531 // UT now contains types that are not named. Loop over it, naming structure
534 bool Changed = false;
535 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
537 if (const StructType *ST = dyn_cast<StructType>(*I)) {
538 ((Value*)ST)->setName("unnamed", &MST);
544 // generateCompilerSpecificCode - This is where we add conditional compilation
545 // directives to cater to specific compilers as need be.
547 static void generateCompilerSpecificCode(std::ostream& Out) {
548 // Alloca is hard to get, and we don't want to include stdlib.h here...
549 Out << "/* get a declaration for alloca */\n"
551 << "extern void *__builtin_alloca(unsigned long);\n"
552 << "#define alloca(x) __builtin_alloca(x)\n"
554 << "#ifndef __FreeBSD__\n"
555 << "#include <alloca.h>\n"
559 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
560 // If we aren't being compiled with GCC, just drop these attributes.
561 Out << "#ifndef __GNUC__ /* Can only support \"linkonce\" vars with GCC */\n"
562 << "#define __attribute__(X)\n"
566 void CWriter::printModule(Module *M) {
567 // Calculate which global values have names that will collide when we throw
568 // away type information.
569 { // Scope to delete the FoundNames set when we are done with it...
570 std::set<std::string> FoundNames;
571 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
572 if (I->hasName()) // If the global has a name...
573 if (FoundNames.count(I->getName())) // And the name is already used
574 MangledGlobals.insert(I); // Mangle the name
576 FoundNames.insert(I->getName()); // Otherwise, keep track of name
578 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
579 if (I->hasName()) // If the global has a name...
580 if (FoundNames.count(I->getName())) // And the name is already used
581 MangledGlobals.insert(I); // Mangle the name
583 FoundNames.insert(I->getName()); // Otherwise, keep track of name
586 // get declaration for alloca
587 Out << "/* Provide Declarations */\n";
588 Out << "#include <stdarg.h>\n";
589 Out << "#include <setjmp.h>\n";
590 generateCompilerSpecificCode(Out);
592 // Provide a definition for `bool' if not compiling with a C++ compiler.
594 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
596 << "\n\n/* Support for floating point constants */\n"
597 << "typedef unsigned long long ConstantDoubleTy;\n"
598 << "typedef unsigned int ConstantFloatTy;\n"
600 << "\n\n/* Support for the invoke instruction */\n"
601 << "extern struct __llvm_jmpbuf_list_t {\n"
602 << " jmp_buf buf; struct __llvm_jmpbuf_list_t *next;\n"
603 << "} *__llvm_jmpbuf_list;\n"
605 << "\n\n/* Global Declarations */\n";
607 // First output all the declarations for the program, because C requires
608 // Functions & globals to be declared before they are used.
611 // Loop over the symbol table, emitting all named constants...
612 printSymbolTable(M->getSymbolTable());
614 // Global variable declarations...
616 Out << "\n/* External Global Variable Declarations */\n";
617 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
618 if (I->hasExternalLinkage()) {
620 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
626 // Function declarations
628 Out << "\n/* Function Declarations */\n";
630 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
631 // If the function is external and the name collides don't print it.
632 // Sometimes the bytecode likes to have multiple "declarations" for
633 // external functions
634 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
635 !I->getIntrinsicID()) {
636 printFunctionSignature(I, true);
642 // Print Malloc prototype if needed
644 Out << "\n/* Malloc to make sun happy */\n";
645 Out << "extern void * malloc();\n\n";
648 // Output the global variable declarations
650 Out << "\n\n/* Global Variable Declarations */\n";
651 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
652 if (!I->isExternal()) {
654 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
660 // Output the global variable definitions and contents...
662 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
663 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
664 if (!I->isExternal()) {
665 if (I->hasInternalLinkage())
667 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
668 if (I->hasLinkOnceLinkage())
669 Out << " __attribute__((common))";
670 if (!I->getInitializer()->isNullValue()) {
672 writeOperand(I->getInitializer());
678 // Output all of the functions...
679 emittedInvoke = false;
681 Out << "\n\n/* Function Bodies */\n";
682 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
686 // If the program included an invoke instruction, we need to output the
687 // support code for it here!
689 Out << "\n/* More support for the invoke instruction */\n"
690 << "struct __llvm_jmpbuf_list_t *__llvm_jmpbuf_list "
691 << "__attribute__((common)) = 0;\n";
696 /// printSymbolTable - Run through symbol table looking for type names. If a
697 /// type name is found, emit it's declaration...
699 void CWriter::printSymbolTable(const SymbolTable &ST) {
700 // If there are no type names, exit early.
701 if (ST.find(Type::TypeTy) == ST.end())
704 // We are only interested in the type plane of the symbol table...
705 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
706 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
708 // Print out forward declarations for structure types before anything else!
709 Out << "/* Structure forward decls */\n";
710 for (; I != End; ++I)
711 if (const Type *STy = dyn_cast<StructType>(I->second)) {
712 std::string Name = "struct l_" + Mangler::makeNameProper(I->first);
713 Out << Name << ";\n";
714 TypeNames.insert(std::make_pair(STy, Name));
719 // Now we can print out typedefs...
720 Out << "/* Typedefs */\n";
721 for (I = ST.type_begin(Type::TypeTy); I != End; ++I) {
722 const Type *Ty = cast<Type>(I->second);
723 std::string Name = "l_" + Mangler::makeNameProper(I->first);
725 printType(Out, Ty, Name);
731 // Keep track of which structures have been printed so far...
732 std::set<const StructType *> StructPrinted;
734 // Loop over all structures then push them into the stack so they are
735 // printed in the correct order.
737 Out << "/* Structure contents */\n";
738 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
739 if (const StructType *STy = dyn_cast<StructType>(I->second))
740 printContainedStructs(STy, StructPrinted);
743 // Push the struct onto the stack and recursively push all structs
744 // this one depends on.
745 void CWriter::printContainedStructs(const Type *Ty,
746 std::set<const StructType*> &StructPrinted){
747 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
748 //Check to see if we have already printed this struct
749 if (StructPrinted.count(STy) == 0) {
750 // Print all contained types first...
751 for (StructType::ElementTypes::const_iterator
752 I = STy->getElementTypes().begin(),
753 E = STy->getElementTypes().end(); I != E; ++I) {
754 const Type *Ty1 = I->get();
755 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
756 printContainedStructs(*I, StructPrinted);
759 //Print structure type out..
760 StructPrinted.insert(STy);
761 std::string Name = TypeNames[STy];
762 printType(Out, STy, Name, true);
766 // If it is an array, check contained types and continue
767 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
768 const Type *Ty1 = ATy->getElementType();
769 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
770 printContainedStructs(Ty1, StructPrinted);
775 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
776 // If the program provides its own malloc prototype we don't need
777 // to include the general one.
778 if (Mang->getValueName(F) == "malloc")
781 if (F->hasInternalLinkage()) Out << "static ";
782 if (F->hasLinkOnceLinkage()) Out << "inline ";
784 // Loop over the arguments, printing them...
785 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
787 std::stringstream FunctionInnards;
789 // Print out the name...
790 FunctionInnards << Mang->getValueName(F) << "(";
792 if (!F->isExternal()) {
795 if (F->abegin()->hasName() || !Prototype)
796 ArgName = Mang->getValueName(F->abegin());
797 printType(FunctionInnards, F->afront().getType(), ArgName);
798 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
800 FunctionInnards << ", ";
801 if (I->hasName() || !Prototype)
802 ArgName = Mang->getValueName(I);
805 printType(FunctionInnards, I->getType(), ArgName);
809 // Loop over the arguments, printing them...
810 for (FunctionType::ParamTypes::const_iterator I =
811 FT->getParamTypes().begin(),
812 E = FT->getParamTypes().end(); I != E; ++I) {
813 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
814 printType(FunctionInnards, *I);
818 // Finish printing arguments... if this is a vararg function, print the ...,
819 // unless there are no known types, in which case, we just emit ().
821 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
822 if (FT->getParamTypes().size()) FunctionInnards << ", ";
823 FunctionInnards << "..."; // Output varargs portion of signature!
825 FunctionInnards << ")";
826 // Print out the return type and the entire signature for that matter
827 printType(Out, F->getReturnType(), FunctionInnards.str());
830 void CWriter::printFunction(Function *F) {
831 if (F->isExternal()) return;
833 printFunctionSignature(F, false);
836 // print local variable information for the function
837 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
838 if (const AllocaInst *AI = isDirectAlloca(*I)) {
840 printType(Out, AI->getAllocatedType(), Mang->getValueName(AI));
841 Out << "; /* Address exposed local */\n";
842 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
844 printType(Out, (*I)->getType(), Mang->getValueName(*I));
847 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
849 printType(Out, (*I)->getType(), Mang->getValueName(*I)+"__PHI_TEMPORARY");
856 // Scan the function for floating point constants. If any FP constant is used
857 // in the function, we want to redirect it here so that we do not depend on
858 // the precision of the printed form, unless the printed form preserves
861 unsigned FPCounter = 0;
862 for (constant_iterator I = constant_begin(F), E = constant_end(F); I != E;++I)
863 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
864 if ((!FPCSafeToPrint(FPC)) // Do not put in FPConstantMap if safe.
865 && (FPConstantMap.find(FPC) == FPConstantMap.end())) {
866 double Val = FPC->getValue();
868 FPConstantMap[FPC] = FPCounter; // Number the FP constants
870 if (FPC->getType() == Type::DoubleTy)
871 Out << " const ConstantDoubleTy FloatConstant" << FPCounter++
872 << " = 0x" << std::hex << *(unsigned long long*)&Val << std::dec
873 << "; /* " << Val << " */\n";
874 else if (FPC->getType() == Type::FloatTy) {
876 Out << " const ConstantFloatTy FloatConstant" << FPCounter++
877 << " = 0x" << std::hex << *(unsigned*)&fVal << std::dec
878 << "; /* " << Val << " */\n";
880 assert(0 && "Unknown float type!");
885 // print the basic blocks
886 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
887 BasicBlock *Prev = BB->getPrev();
889 // Don't print the label for the basic block if there are no uses, or if the
890 // only terminator use is the precessor basic block's terminator. We have
891 // to scan the use list because PHI nodes use basic blocks too but do not
892 // require a label to be generated.
894 bool NeedsLabel = false;
895 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
897 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
898 if (TI != Prev->getTerminator() ||
899 isa<SwitchInst>(Prev->getTerminator())) {
904 if (NeedsLabel) Out << Mang->getValueName(BB) << ":\n";
906 // Output all of the instructions in the basic block...
907 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
908 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
909 if (II->getType() != Type::VoidTy)
918 // Don't emit prefix or suffix for the terminator...
919 visit(*BB->getTerminator());
923 FPConstantMap.clear();
926 // Specific Instruction type classes... note that all of the casts are
927 // necessary because we use the instruction classes as opaque types...
929 void CWriter::visitReturnInst(ReturnInst &I) {
930 // Don't output a void return if this is the last basic block in the function
931 if (I.getNumOperands() == 0 &&
932 &*--I.getParent()->getParent()->end() == I.getParent() &&
933 !I.getParent()->size() == 1) {
938 if (I.getNumOperands()) {
940 writeOperand(I.getOperand(0));
945 void CWriter::visitSwitchInst(SwitchInst &SI) {
947 writeOperand(SI.getOperand(0));
948 Out << ") {\n default:\n";
949 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
951 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
953 writeOperand(SI.getOperand(i));
955 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
956 printBranchToBlock(SI.getParent(), Succ, 2);
957 if (Succ == SI.getParent()->getNext())
963 void CWriter::visitInvokeInst(InvokeInst &II) {
965 << " struct __llvm_jmpbuf_list_t Entry;\n"
966 << " Entry.next = __llvm_jmpbuf_list;\n"
967 << " if (setjmp(Entry.buf)) {\n"
968 << " __llvm_jmpbuf_list = Entry.next;\n";
969 printBranchToBlock(II.getParent(), II.getExceptionalDest(), 4);
971 << " __llvm_jmpbuf_list = &Entry;\n"
975 << " __llvm_jmpbuf_list = Entry.next;\n"
977 printBranchToBlock(II.getParent(), II.getNormalDest(), 0);
978 emittedInvoke = true;
982 static bool isGotoCodeNeccessary(BasicBlock *From, BasicBlock *To) {
983 // If PHI nodes need copies, we need the copy code...
984 if (isa<PHINode>(To->front()) ||
985 From->getNext() != To) // Not directly successor, need goto
988 // Otherwise we don't need the code.
992 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
994 for (BasicBlock::iterator I = Succ->begin();
995 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
996 // now we have to do the printing
997 Out << std::string(Indent, ' ');
998 Out << " " << Mang->getValueName(I) << "__PHI_TEMPORARY = ";
999 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1000 Out << "; /* for PHI node */\n";
1003 if (CurBB->getNext() != Succ) {
1004 Out << std::string(Indent, ' ') << " goto ";
1010 // Brach instruction printing - Avoid printing out a brach to a basic block that
1011 // immediately succeeds the current one.
1013 void CWriter::visitBranchInst(BranchInst &I) {
1014 if (I.isConditional()) {
1015 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(0))) {
1017 writeOperand(I.getCondition());
1020 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1022 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(1))) {
1023 Out << " } else {\n";
1024 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1027 // First goto not necessary, assume second one is...
1029 writeOperand(I.getCondition());
1032 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1037 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1042 // PHI nodes get copied into temporary values at the end of predecessor basic
1043 // blocks. We now need to copy these temporary values into the REAL value for
1045 void CWriter::visitPHINode(PHINode &I) {
1047 Out << "__PHI_TEMPORARY";
1051 void CWriter::visitBinaryOperator(Instruction &I) {
1052 // binary instructions, shift instructions, setCond instructions.
1053 assert(!isa<PointerType>(I.getType()));
1055 // We must cast the results of binary operations which might be promoted.
1056 bool needsCast = false;
1057 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1058 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1059 || (I.getType() == Type::FloatTy)) {
1062 printType(Out, I.getType(), "", false, false);
1066 writeOperand(I.getOperand(0));
1068 switch (I.getOpcode()) {
1069 case Instruction::Add: Out << " + "; break;
1070 case Instruction::Sub: Out << " - "; break;
1071 case Instruction::Mul: Out << "*"; break;
1072 case Instruction::Div: Out << "/"; break;
1073 case Instruction::Rem: Out << "%"; break;
1074 case Instruction::And: Out << " & "; break;
1075 case Instruction::Or: Out << " | "; break;
1076 case Instruction::Xor: Out << " ^ "; break;
1077 case Instruction::SetEQ: Out << " == "; break;
1078 case Instruction::SetNE: Out << " != "; break;
1079 case Instruction::SetLE: Out << " <= "; break;
1080 case Instruction::SetGE: Out << " >= "; break;
1081 case Instruction::SetLT: Out << " < "; break;
1082 case Instruction::SetGT: Out << " > "; break;
1083 case Instruction::Shl : Out << " << "; break;
1084 case Instruction::Shr : Out << " >> "; break;
1085 default: std::cerr << "Invalid operator type!" << I; abort();
1088 writeOperand(I.getOperand(1));
1095 void CWriter::visitCastInst(CastInst &I) {
1096 if (I.getType() == Type::BoolTy) {
1098 writeOperand(I.getOperand(0));
1103 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1105 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1106 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1107 // Avoid "cast to pointer from integer of different size" warnings
1111 writeOperand(I.getOperand(0));
1114 void CWriter::visitCallInst(CallInst &I) {
1115 // Handle intrinsic function calls first...
1116 if (Function *F = I.getCalledFunction())
1117 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID()) {
1119 default: assert(0 && "Unknown LLVM intrinsic!");
1120 case LLVMIntrinsic::va_start:
1121 Out << "va_start((va_list)*";
1122 writeOperand(I.getOperand(1));
1124 // Output the last argument to the enclosing function...
1125 writeOperand(&I.getParent()->getParent()->aback());
1128 case LLVMIntrinsic::va_end:
1129 Out << "va_end((va_list)*";
1130 writeOperand(I.getOperand(1));
1133 case LLVMIntrinsic::va_copy:
1134 Out << "va_copy((va_list)*";
1135 writeOperand(I.getOperand(1));
1136 Out << ", (va_list)";
1137 writeOperand(I.getOperand(2));
1141 case LLVMIntrinsic::unwind:
1142 // The unwind intrinsic calls a control flow transfer out of the current
1143 // function, unwinding the stack until a caller who used the invoke
1144 // instruction is found. In this context, we code generated the invoke
1145 // instruction to add an entry to the top of the jmpbuf_list. Thus,
1146 // here we just have to longjmp to the specified handler.
1147 Out << "if (__llvm_jmpbuf_list == 0) { /* llvm.unwind */\n"
1148 << " printf(\"throw found with no handler!\\n\"); abort();\n"
1150 << " longjmp(__llvm_jmpbuf_list->buf, 1)";
1154 case LLVMIntrinsic::setjmp:
1155 case LLVMIntrinsic::sigsetjmp:
1156 // This instrinsic should never exist in the program, but until we get
1157 // setjmp/longjmp transformations going on, we should codegen it to
1158 // something reasonable. This will allow code that never calls longjmp
1162 case LLVMIntrinsic::longjmp:
1163 case LLVMIntrinsic::siglongjmp:
1164 // Longjmp is not implemented, and never will be. It would cause an
1173 void CWriter::visitCallSite(CallSite CS) {
1174 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1175 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1176 const Type *RetTy = FTy->getReturnType();
1178 writeOperand(CS.getCalledValue());
1181 if (CS.arg_begin() != CS.arg_end()) {
1182 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1185 for (++AI; AI != AE; ++AI) {
1193 void CWriter::visitMallocInst(MallocInst &I) {
1195 printType(Out, I.getType());
1196 Out << ")malloc(sizeof(";
1197 printType(Out, I.getType()->getElementType());
1200 if (I.isArrayAllocation()) {
1202 writeOperand(I.getOperand(0));
1207 void CWriter::visitAllocaInst(AllocaInst &I) {
1209 printType(Out, I.getType());
1210 Out << ") alloca(sizeof(";
1211 printType(Out, I.getType()->getElementType());
1213 if (I.isArrayAllocation()) {
1215 writeOperand(I.getOperand(0));
1220 void CWriter::visitFreeInst(FreeInst &I) {
1221 Out << "free((char*)";
1222 writeOperand(I.getOperand(0));
1226 void CWriter::printIndexingExpression(Value *Ptr, User::op_iterator I,
1227 User::op_iterator E) {
1228 bool HasImplicitAddress = false;
1229 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1230 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1231 HasImplicitAddress = true;
1232 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1233 HasImplicitAddress = true;
1234 Ptr = CPR->getValue(); // Get to the global...
1235 } else if (isDirectAlloca(Ptr)) {
1236 HasImplicitAddress = true;
1240 if (!HasImplicitAddress)
1241 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1243 writeOperandInternal(Ptr);
1247 const Constant *CI = dyn_cast<Constant>(I);
1248 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1251 writeOperandInternal(Ptr);
1253 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1255 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1258 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1259 "Can only have implicit address with direct accessing");
1261 if (HasImplicitAddress) {
1263 } else if (CI && CI->isNullValue() && I+1 != E) {
1264 // Print out the -> operator if possible...
1265 if ((*(I+1))->getType() == Type::UByteTy) {
1266 Out << (HasImplicitAddress ? "." : "->");
1267 Out << "field" << cast<ConstantUInt>(*(I+1))->getValue();
1273 if ((*I)->getType() == Type::LongTy) {
1278 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
1282 void CWriter::visitLoadInst(LoadInst &I) {
1284 writeOperand(I.getOperand(0));
1287 void CWriter::visitStoreInst(StoreInst &I) {
1289 writeOperand(I.getPointerOperand());
1291 writeOperand(I.getOperand(0));
1294 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1296 printIndexingExpression(I.getPointerOperand(), I.idx_begin(), I.idx_end());
1299 void CWriter::visitVarArgInst(VarArgInst &I) {
1300 Out << "va_arg((va_list)*";
1301 writeOperand(I.getOperand(0));
1303 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1308 //===----------------------------------------------------------------------===//
1309 // External Interface declaration
1310 //===----------------------------------------------------------------------===//
1312 Pass *createWriteToCPass(std::ostream &o) { return new CWriter(o); }