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/SlotCalculator.h"
16 #include "llvm/Analysis/FindUsedTypes.h"
17 #include "llvm/Analysis/ConstantsScanner.h"
18 #include "llvm/Support/InstVisitor.h"
19 #include "llvm/Support/InstIterator.h"
20 #include "Support/StringExtras.h"
21 #include "Support/STLExtras.h"
27 class CWriter : public Pass, public InstVisitor<CWriter> {
29 SlotCalculator *Table;
30 const Module *TheModule;
31 std::map<const Type *, std::string> TypeNames;
32 std::set<const Value*> MangledGlobals;
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) {
46 Table = new SlotCalculator(&M, false);
49 // Ensure that all structure types have names...
50 bool Changed = nameAllUsedStructureTypes(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);
69 std::string getValueName(const Value *V);
72 bool nameAllUsedStructureTypes(Module &M);
73 void printModule(Module *M);
74 void printSymbolTable(const SymbolTable &ST);
75 void printContainedStructs(const Type *Ty, std::set<const StructType *> &);
76 void printFunctionSignature(const Function *F, bool Prototype);
78 void printFunction(Function *);
80 void printConstant(Constant *CPV);
81 void printConstantArray(ConstantArray *CPA);
83 // isInlinableInst - Attempt to inline instructions into their uses to build
84 // trees as much as possible. To do this, we have to consistently decide
85 // what is acceptable to inline, so that variable declarations don't get
86 // printed and an extra copy of the expr is not emitted.
88 static bool isInlinableInst(const Instruction &I) {
89 // Must be an expression, must be used exactly once. If it is dead, we
90 // emit it inline where it would go.
91 if (I.getType() == Type::VoidTy || I.use_size() != 1 ||
92 isa<TerminatorInst>(I) || isa<CallInst>(I) || isa<PHINode>(I) ||
93 isa<LoadInst>(I)) // Don't inline a load across a store!
96 // Only inline instruction it it's use is in the same BB as the inst.
97 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
100 // isDirectAlloca - Define fixed sized allocas in the entry block as direct
101 // variables which are accessed with the & operator. This causes GCC to
102 // generate significantly better code than to emit alloca calls directly.
104 static const AllocaInst *isDirectAlloca(const Value *V) {
105 const AllocaInst *AI = dyn_cast<AllocaInst>(V);
106 if (!AI) return false;
107 if (AI->isArrayAllocation())
108 return 0; // FIXME: we can also inline fixed size array allocas!
109 if (AI->getParent() != &AI->getParent()->getParent()->getEntryNode())
114 // Instruction visitation functions
115 friend class InstVisitor<CWriter>;
117 void visitReturnInst(ReturnInst &I);
118 void visitBranchInst(BranchInst &I);
119 void visitSwitchInst(SwitchInst &I);
121 void visitPHINode(PHINode &I);
122 void visitBinaryOperator(Instruction &I);
124 void visitCastInst (CastInst &I);
125 void visitCallInst (CallInst &I);
126 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
128 void visitMallocInst(MallocInst &I);
129 void visitAllocaInst(AllocaInst &I);
130 void visitFreeInst (FreeInst &I);
131 void visitLoadInst (LoadInst &I);
132 void visitStoreInst (StoreInst &I);
133 void visitGetElementPtrInst(GetElementPtrInst &I);
134 void visitVarArgInst(VarArgInst &I);
136 void visitInstruction(Instruction &I) {
137 std::cerr << "C Writer does not know about " << I;
141 void outputLValue(Instruction *I) {
142 Out << " " << getValueName(I) << " = ";
144 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
146 void printIndexingExpression(Value *Ptr, User::op_iterator I,
147 User::op_iterator E);
151 // We dont want identifier names with ., space, - in them.
152 // So we replace them with _
153 static std::string makeNameProper(std::string x) {
155 for (std::string::iterator sI = x.begin(), sEnd = x.end(); sI != sEnd; sI++)
157 case '.': tmp += "d_"; break;
158 case ' ': tmp += "s_"; break;
159 case '-': tmp += "D_"; break;
166 std::string CWriter::getValueName(const Value *V) {
167 if (V->hasName()) { // Print out the label if it exists...
169 // Name mangling occurs as follows:
170 // - If V is not a global, mangling always occurs.
171 // - Otherwise, mangling occurs when any of the following are true:
172 // 1) V has internal linkage
173 // 2) V's name would collide if it is not mangled.
176 if(const GlobalValue* gv = dyn_cast<GlobalValue>(V)) {
177 if(!gv->hasInternalLinkage() && !MangledGlobals.count(gv)) {
178 // No internal linkage, name will not collide -> no mangling.
179 return makeNameProper(gv->getName());
183 // Non-global, or global with internal linkage / colliding name -> mangle.
184 return "l" + utostr(V->getType()->getUniqueID()) + "_" +
185 makeNameProper(V->getName());
188 int Slot = Table->getValSlot(V);
189 assert(Slot >= 0 && "Invalid value!");
190 return "ltmp_" + itostr(Slot) + "_" + utostr(V->getType()->getUniqueID());
193 // A pointer type should not use parens around *'s alone, e.g., (**)
194 inline bool ptrTypeNameNeedsParens(const std::string &NameSoFar) {
195 return (NameSoFar.find_last_not_of('*') != std::string::npos);
198 // Pass the Type* and the variable name and this prints out the variable
201 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
202 const std::string &NameSoFar,
203 bool IgnoreName, bool namedContext) {
204 if (Ty->isPrimitiveType())
205 switch (Ty->getPrimitiveID()) {
206 case Type::VoidTyID: return Out << "void " << NameSoFar;
207 case Type::BoolTyID: return Out << "bool " << NameSoFar;
208 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
209 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
210 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
211 case Type::ShortTyID: return Out << "short " << NameSoFar;
212 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
213 case Type::IntTyID: return Out << "int " << NameSoFar;
214 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
215 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
216 case Type::FloatTyID: return Out << "float " << NameSoFar;
217 case Type::DoubleTyID: return Out << "double " << NameSoFar;
219 std::cerr << "Unknown primitive type: " << Ty << "\n";
223 // Check to see if the type is named.
224 if (!IgnoreName || isa<OpaqueType>(Ty)) {
225 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
226 if (I != TypeNames.end()) {
227 return Out << I->second << " " << NameSoFar;
231 switch (Ty->getPrimitiveID()) {
232 case Type::FunctionTyID: {
233 const FunctionType *MTy = cast<FunctionType>(Ty);
234 std::stringstream FunctionInnards;
235 FunctionInnards << " (" << NameSoFar << ") (";
236 for (FunctionType::ParamTypes::const_iterator
237 I = MTy->getParamTypes().begin(),
238 E = MTy->getParamTypes().end(); I != E; ++I) {
239 if (I != MTy->getParamTypes().begin())
240 FunctionInnards << ", ";
241 printType(FunctionInnards, *I, "");
243 if (MTy->isVarArg()) {
244 if (!MTy->getParamTypes().empty())
245 FunctionInnards << ", ...";
246 } else if (MTy->getParamTypes().empty()) {
247 FunctionInnards << "void";
249 FunctionInnards << ")";
250 std::string tstr = FunctionInnards.str();
251 printType(Out, MTy->getReturnType(), tstr);
254 case Type::StructTyID: {
255 const StructType *STy = cast<StructType>(Ty);
256 Out << NameSoFar + " {\n";
258 for (StructType::ElementTypes::const_iterator
259 I = STy->getElementTypes().begin(),
260 E = STy->getElementTypes().end(); I != E; ++I) {
262 printType(Out, *I, "field" + utostr(Idx++));
268 case Type::PointerTyID: {
269 const PointerType *PTy = cast<PointerType>(Ty);
270 std::string ptrName = "*" + NameSoFar;
272 // Do not need parens around "* NameSoFar" if NameSoFar consists only
273 // of zero or more '*' chars *and* this is not an unnamed pointer type
274 // such as the result type in a cast statement. Otherwise, enclose in ( ).
275 if (ptrTypeNameNeedsParens(NameSoFar) || !namedContext ||
276 PTy->getElementType()->getPrimitiveID() == Type::ArrayTyID)
277 ptrName = "(" + ptrName + ")"; //
279 return printType(Out, PTy->getElementType(), ptrName);
282 case Type::ArrayTyID: {
283 const ArrayType *ATy = cast<ArrayType>(Ty);
284 unsigned NumElements = ATy->getNumElements();
285 return printType(Out, ATy->getElementType(),
286 NameSoFar + "[" + utostr(NumElements) + "]");
289 case Type::OpaqueTyID: {
290 static int Count = 0;
291 std::string TyName = "struct opaque_" + itostr(Count++);
292 assert(TypeNames.find(Ty) == TypeNames.end());
293 TypeNames[Ty] = TyName;
294 return Out << TyName << " " << NameSoFar;
297 assert(0 && "Unhandled case in getTypeProps!");
304 void CWriter::printConstantArray(ConstantArray *CPA) {
306 // As a special case, print the array as a string if it is an array of
307 // ubytes or an array of sbytes with positive values.
309 const Type *ETy = CPA->getType()->getElementType();
310 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
312 // Make sure the last character is a null char, as automatically added by C
313 if (isString && (CPA->getNumOperands() == 0 ||
314 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
319 // Keep track of whether the last number was a hexadecimal escape
320 bool LastWasHex = false;
322 // Do not include the last character, which we know is null
323 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
324 unsigned char C = (ETy == Type::SByteTy) ?
325 (unsigned char)cast<ConstantSInt>(CPA->getOperand(i))->getValue() :
326 (unsigned char)cast<ConstantUInt>(CPA->getOperand(i))->getValue();
328 // Print it out literally if it is a printable character. The only thing
329 // to be careful about is when the last letter output was a hex escape
330 // code, in which case we have to be careful not to print out hex digits
331 // explicitly (the C compiler thinks it is a continuation of the previous
332 // character, sheesh...)
334 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
336 if (C == '"' || C == '\\')
343 case '\n': Out << "\\n"; break;
344 case '\t': Out << "\\t"; break;
345 case '\r': Out << "\\r"; break;
346 case '\v': Out << "\\v"; break;
347 case '\a': Out << "\\a"; break;
348 case '\"': Out << "\\\""; break;
349 case '\'': Out << "\\\'"; break;
352 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
353 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
362 if (CPA->getNumOperands()) {
364 printConstant(cast<Constant>(CPA->getOperand(0)));
365 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
367 printConstant(cast<Constant>(CPA->getOperand(i)));
374 /// FPCSafeToPrint - Returns true if we may assume that CFP may be
375 /// written out textually as a double (rather than as a reference to a
376 /// stack-allocated variable). We decide this by converting CFP to a
377 /// string and back into a double, and then checking whether the
378 /// conversion results in a bit-equal double to the original value of
379 /// CFP. This depends on us and the target C compiler agreeing on the
380 /// conversion process (which is pretty likely since we only deal in
381 /// IEEE FP.) This is adapted from similar code in
382 /// lib/VMCore/AsmWriter.cpp:WriteConstantInt().
383 static bool FPCSafeToPrint (const ConstantFP *CFP) {
384 std::string StrVal = ftostr(CFP->getValue());
385 // Check to make sure that the stringized number is not some string like
386 // "Inf" or NaN, that atof will accept, but the lexer will not. Check that
387 // the string matches the "[-+]?[0-9]" regex.
388 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
389 ((StrVal[0] == '-' || StrVal[0] == '+') &&
390 (StrVal[1] >= '0' && StrVal[1] <= '9')))
391 // Reparse stringized version!
392 return (atof(StrVal.c_str()) == CFP->getValue());
396 // printConstant - The LLVM Constant to C Constant converter.
397 void CWriter::printConstant(Constant *CPV) {
398 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
399 switch (CE->getOpcode()) {
400 case Instruction::Cast:
402 printType(Out, CPV->getType());
404 printConstant(CE->getOperand(0));
408 case Instruction::GetElementPtr:
410 printIndexingExpression(CE->getOperand(0),
411 CPV->op_begin()+1, CPV->op_end());
414 case Instruction::Add:
416 printConstant(CE->getOperand(0));
418 printConstant(CE->getOperand(1));
421 case Instruction::Sub:
423 printConstant(CE->getOperand(0));
425 printConstant(CE->getOperand(1));
430 std::cerr << "CWriter Error: Unhandled constant expression: "
436 switch (CPV->getType()->getPrimitiveID()) {
438 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
439 case Type::SByteTyID:
440 case Type::ShortTyID:
441 Out << cast<ConstantSInt>(CPV)->getValue(); break;
443 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
444 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
446 Out << cast<ConstantSInt>(CPV)->getValue();
450 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
452 case Type::UByteTyID:
453 case Type::UShortTyID:
454 Out << cast<ConstantUInt>(CPV)->getValue(); break;
456 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
457 case Type::ULongTyID:
458 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
460 case Type::FloatTyID:
461 case Type::DoubleTyID: {
462 ConstantFP *FPC = cast<ConstantFP>(CPV);
463 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
464 if (I != FPConstantMap.end()) {
465 // Because of FP precision problems we must load from a stack allocated
466 // value that holds the value in hex.
467 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
468 << "*)&FloatConstant" << I->second << ")";
470 if (FPCSafeToPrint (FPC)) {
471 Out << ftostr (FPC->getValue ());
473 Out << FPC->getValue(); // Who knows? Give it our best shot...
479 case Type::ArrayTyID:
480 printConstantArray(cast<ConstantArray>(CPV));
483 case Type::StructTyID: {
485 if (CPV->getNumOperands()) {
487 printConstant(cast<Constant>(CPV->getOperand(0)));
488 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
490 printConstant(cast<Constant>(CPV->getOperand(i)));
497 case Type::PointerTyID:
498 if (isa<ConstantPointerNull>(CPV)) {
500 printType(Out, CPV->getType());
501 Out << ")/*NULL*/0)";
503 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
504 writeOperand(CPR->getValue());
509 std::cerr << "Unknown constant type: " << CPV << "\n";
514 void CWriter::writeOperandInternal(Value *Operand) {
515 if (Instruction *I = dyn_cast<Instruction>(Operand))
516 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
517 // Should we inline this instruction to build a tree?
524 if (Operand->hasName()) {
525 Out << getValueName(Operand);
526 } else if (Constant *CPV = dyn_cast<Constant>(Operand)) {
529 int Slot = Table->getValSlot(Operand);
530 assert(Slot >= 0 && "Malformed LLVM!");
531 Out << "ltmp_" << Slot << "_" << Operand->getType()->getUniqueID();
535 void CWriter::writeOperand(Value *Operand) {
536 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
537 Out << "(&"; // Global variables are references as their addresses by llvm
539 writeOperandInternal(Operand);
541 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
545 // nameAllUsedStructureTypes - If there are structure types in the module that
546 // are used but do not have names assigned to them in the symbol table yet then
547 // we assign them names now.
549 bool CWriter::nameAllUsedStructureTypes(Module &M) {
550 // Get a set of types that are used by the program...
551 std::set<const Type *> UT = getAnalysis<FindUsedTypes>().getTypes();
553 // Loop over the module symbol table, removing types from UT that are already
556 SymbolTable &MST = M.getSymbolTable();
557 if (MST.find(Type::TypeTy) != MST.end())
558 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
559 E = MST.type_end(Type::TypeTy); I != E; ++I)
560 UT.erase(cast<Type>(I->second));
562 // UT now contains types that are not named. Loop over it, naming structure
565 bool Changed = false;
566 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
568 if (const StructType *ST = dyn_cast<StructType>(*I)) {
569 ((Value*)ST)->setName("unnamed", &MST);
575 static void generateAllocaDecl(std::ostream& Out) {
576 // On SunOS, we need to insert the alloca macro & proto for the builtin.
577 Out << "#ifdef sun\n"
578 << "extern void *__builtin_alloca(unsigned long);\n"
579 << "#define alloca(x) __builtin_alloca(x)\n"
581 << "#ifndef __FreeBSD__\n"
582 << "#include <alloca.h>\n"
587 void CWriter::printModule(Module *M) {
588 // Calculate which global values have names that will collide when we throw
589 // away type information.
590 { // Scope to delete the FoundNames set when we are done with it...
591 std::set<std::string> FoundNames;
592 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
593 if (I->hasName()) // If the global has a name...
594 if (FoundNames.count(I->getName())) // And the name is already used
595 MangledGlobals.insert(I); // Mangle the name
597 FoundNames.insert(I->getName()); // Otherwise, keep track of name
599 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
600 if (I->hasName()) // If the global has a name...
601 if (FoundNames.count(I->getName())) // And the name is already used
602 MangledGlobals.insert(I); // Mangle the name
604 FoundNames.insert(I->getName()); // Otherwise, keep track of name
607 // get declaration for alloca
608 Out << "/* Provide Declarations */\n";
609 generateAllocaDecl(Out);
610 Out << "#include <stdarg.h>\n";
611 Out << "#include <setjmp.h>\n";
613 // Provide a definition for `bool' if not compiling with a C++ compiler.
615 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
617 << "\n\n/* Support for floating point constants */\n"
618 << "typedef unsigned long long ConstantDoubleTy;\n"
619 << "typedef unsigned int ConstantFloatTy;\n"
621 << "\n\n/* Global Declarations */\n";
623 // First output all the declarations for the program, because C requires
624 // Functions & globals to be declared before they are used.
627 // Loop over the symbol table, emitting all named constants...
628 printSymbolTable(M->getSymbolTable());
630 // Global variable declarations...
632 Out << "\n/* External Global Variable Declarations */\n";
633 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
634 if (I->hasExternalLinkage()) {
636 printType(Out, I->getType()->getElementType(), getValueName(I));
642 // Function declarations
644 Out << "\n/* Function Declarations */\n";
646 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
647 // If the function is external and the name collides don't print it.
648 // Sometimes the bytecode likes to have multiple "declarations" for
649 // external functions
650 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
651 !I->getIntrinsicID()) {
652 printFunctionSignature(I, true);
658 // Print Malloc prototype if needed
660 Out << "\n/* Malloc to make sun happy */\n";
661 Out << "extern void * malloc(size_t);\n\n";
664 // Output the global variable declarations
666 Out << "\n\n/* Global Variable Declarations */\n";
667 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
668 if (!I->isExternal()) {
670 printType(Out, I->getType()->getElementType(), getValueName(I));
677 // Output the global variable definitions and contents...
679 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
680 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
681 if (!I->isExternal()) {
682 if (I->hasInternalLinkage())
684 printType(Out, I->getType()->getElementType(), getValueName(I));
685 if (!I->getInitializer()->isNullValue()) {
687 writeOperand(I->getInitializer());
693 // Output all of the functions...
695 Out << "\n\n/* Function Bodies */\n";
696 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
702 /// printSymbolTable - Run through symbol table looking for type names. If a
703 /// type name is found, emit it's declaration...
705 void CWriter::printSymbolTable(const SymbolTable &ST) {
706 // If there are no type names, exit early.
707 if (ST.find(Type::TypeTy) == ST.end())
710 // We are only interested in the type plane of the symbol table...
711 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
712 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
714 // Print out forward declarations for structure types before anything else!
715 Out << "/* Structure forward decls */\n";
716 for (; I != End; ++I)
717 if (const Type *STy = dyn_cast<StructType>(I->second)) {
718 std::string Name = "struct l_" + makeNameProper(I->first);
719 Out << Name << ";\n";
720 TypeNames.insert(std::make_pair(STy, Name));
725 // Now we can print out typedefs...
726 Out << "/* Typedefs */\n";
727 for (I = ST.type_begin(Type::TypeTy); I != End; ++I) {
728 const Type *Ty = cast<Type>(I->second);
729 std::string Name = "l_" + makeNameProper(I->first);
731 printType(Out, Ty, Name);
737 // Keep track of which structures have been printed so far...
738 std::set<const StructType *> StructPrinted;
740 // Loop over all structures then push them into the stack so they are
741 // printed in the correct order.
743 Out << "/* Structure contents */\n";
744 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
745 if (const StructType *STy = dyn_cast<StructType>(I->second))
746 printContainedStructs(STy, StructPrinted);
749 // Push the struct onto the stack and recursively push all structs
750 // this one depends on.
751 void CWriter::printContainedStructs(const Type *Ty,
752 std::set<const StructType*> &StructPrinted){
753 if (const StructType *STy = dyn_cast<StructType>(Ty)){
754 //Check to see if we have already printed this struct
755 if (StructPrinted.count(STy) == 0) {
756 // Print all contained types first...
757 for (StructType::ElementTypes::const_iterator
758 I = STy->getElementTypes().begin(),
759 E = STy->getElementTypes().end(); I != E; ++I) {
760 const Type *Ty1 = I->get();
761 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
762 printContainedStructs(*I, StructPrinted);
765 //Print structure type out..
766 StructPrinted.insert(STy);
767 std::string Name = TypeNames[STy];
768 printType(Out, STy, Name, true);
772 // If it is an array, check contained types and continue
773 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
774 const Type *Ty1 = ATy->getElementType();
775 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
776 printContainedStructs(Ty1, StructPrinted);
781 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
782 // If the program provides its own malloc prototype we don't need
783 // to include the general one.
784 if (getValueName(F) == "malloc")
787 if (F->hasInternalLinkage()) Out << "static ";
788 if (F->hasLinkOnceLinkage()) Out << "inline ";
790 // Loop over the arguments, printing them...
791 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
793 std::stringstream FunctionInnards;
795 // Print out the name...
796 FunctionInnards << getValueName(F) << "(";
798 if (!F->isExternal()) {
801 if (F->abegin()->hasName() || !Prototype)
802 ArgName = getValueName(F->abegin());
803 printType(FunctionInnards, F->afront().getType(), ArgName);
804 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
806 FunctionInnards << ", ";
807 if (I->hasName() || !Prototype)
808 ArgName = getValueName(I);
811 printType(FunctionInnards, I->getType(), ArgName);
815 // Loop over the arguments, printing them...
816 for (FunctionType::ParamTypes::const_iterator I =
817 FT->getParamTypes().begin(),
818 E = FT->getParamTypes().end(); I != E; ++I) {
819 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
820 printType(FunctionInnards, *I);
824 // Finish printing arguments... if this is a vararg function, print the ...,
825 // unless there are no known types, in which case, we just emit ().
827 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
828 if (FT->getParamTypes().size()) FunctionInnards << ", ";
829 FunctionInnards << "..."; // Output varargs portion of signature!
831 FunctionInnards << ")";
832 // Print out the return type and the entire signature for that matter
833 printType(Out, F->getReturnType(), FunctionInnards.str());
837 void CWriter::printFunction(Function *F) {
838 if (F->isExternal()) return;
840 Table->incorporateFunction(F);
842 printFunctionSignature(F, false);
845 // print local variable information for the function
846 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
847 if (const AllocaInst *AI = isDirectAlloca(*I)) {
849 printType(Out, AI->getAllocatedType(), getValueName(AI));
850 Out << "; /* Address exposed local */\n";
851 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
853 printType(Out, (*I)->getType(), getValueName(*I));
856 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
858 printType(Out, (*I)->getType(), getValueName(*I)+"__PHI_TEMPORARY");
865 // Scan the function for floating point constants. If any FP constant is used
866 // in the function, we want to redirect it here so that we do not depend on
867 // the precision of the printed form, unless the printed form preserves
870 unsigned FPCounter = 0;
871 for (constant_iterator I = constant_begin(F), E = constant_end(F); I != E;++I)
872 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
873 if ((!FPCSafeToPrint(FPC)) // Do not put in FPConstantMap if safe.
874 && (FPConstantMap.find(FPC) == FPConstantMap.end())) {
875 double Val = FPC->getValue();
877 FPConstantMap[FPC] = FPCounter; // Number the FP constants
879 if (FPC->getType() == Type::DoubleTy)
880 Out << " const ConstantDoubleTy FloatConstant" << FPCounter++
881 << " = 0x" << std::hex << *(unsigned long long*)&Val << std::dec
882 << "; /* " << Val << " */\n";
883 else if (FPC->getType() == Type::FloatTy) {
885 Out << " const ConstantFloatTy FloatConstant" << FPCounter++
886 << " = 0x" << std::hex << *(unsigned*)&fVal << std::dec
887 << "; /* " << Val << " */\n";
889 assert(0 && "Unknown float type!");
894 // print the basic blocks
895 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
896 BasicBlock *Prev = BB->getPrev();
898 // Don't print the label for the basic block if there are no uses, or if the
899 // only terminator use is the precessor basic block's terminator. We have
900 // to scan the use list because PHI nodes use basic blocks too but do not
901 // require a label to be generated.
903 bool NeedsLabel = false;
904 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
906 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
907 if (TI != Prev->getTerminator() ||
908 isa<SwitchInst>(Prev->getTerminator())) {
913 if (NeedsLabel) Out << getValueName(BB) << ":\n";
915 // Output all of the instructions in the basic block...
916 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
917 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
918 if (II->getType() != Type::VoidTy)
927 // Don't emit prefix or suffix for the terminator...
928 visit(*BB->getTerminator());
932 Table->purgeFunction();
933 FPConstantMap.clear();
936 // Specific Instruction type classes... note that all of the casts are
937 // neccesary because we use the instruction classes as opaque types...
939 void CWriter::visitReturnInst(ReturnInst &I) {
940 // Don't output a void return if this is the last basic block in the function
941 if (I.getNumOperands() == 0 &&
942 &*--I.getParent()->getParent()->end() == I.getParent() &&
943 !I.getParent()->size() == 1) {
948 if (I.getNumOperands()) {
950 writeOperand(I.getOperand(0));
955 void CWriter::visitSwitchInst(SwitchInst &SI) {
957 writeOperand(SI.getOperand(0));
958 Out << ") {\n default:\n";
959 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
961 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
963 writeOperand(SI.getOperand(i));
965 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
966 printBranchToBlock(SI.getParent(), Succ, 2);
967 if (Succ == SI.getParent()->getNext())
974 static bool isGotoCodeNeccessary(BasicBlock *From, BasicBlock *To) {
975 // If PHI nodes need copies, we need the copy code...
976 if (isa<PHINode>(To->front()) ||
977 From->getNext() != To) // Not directly successor, need goto
980 // Otherwise we don't need the code.
984 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
986 for (BasicBlock::iterator I = Succ->begin();
987 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
988 // now we have to do the printing
989 Out << std::string(Indent, ' ');
990 Out << " " << getValueName(I) << "__PHI_TEMPORARY = ";
991 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
992 Out << "; /* for PHI node */\n";
995 if (CurBB->getNext() != Succ) {
996 Out << std::string(Indent, ' ') << " goto ";
1002 // Brach instruction printing - Avoid printing out a brach to a basic block that
1003 // immediately succeeds the current one.
1005 void CWriter::visitBranchInst(BranchInst &I) {
1006 if (I.isConditional()) {
1007 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(0))) {
1009 writeOperand(I.getCondition());
1012 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1014 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(1))) {
1015 Out << " } else {\n";
1016 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1019 // First goto not neccesary, assume second one is...
1021 writeOperand(I.getCondition());
1024 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1029 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1034 // PHI nodes get copied into temporary values at the end of predecessor basic
1035 // blocks. We now need to copy these temporary values into the REAL value for
1037 void CWriter::visitPHINode(PHINode &I) {
1039 Out << "__PHI_TEMPORARY";
1043 void CWriter::visitBinaryOperator(Instruction &I) {
1044 // binary instructions, shift instructions, setCond instructions.
1045 assert(!isa<PointerType>(I.getType()));
1047 // We must cast the results of binary operations which might be promoted.
1048 bool needsCast = false;
1049 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1050 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1051 || (I.getType() == Type::FloatTy)) {
1054 printType(Out, I.getType(), "", false, false);
1058 writeOperand(I.getOperand(0));
1060 switch (I.getOpcode()) {
1061 case Instruction::Add: Out << " + "; break;
1062 case Instruction::Sub: Out << " - "; break;
1063 case Instruction::Mul: Out << "*"; break;
1064 case Instruction::Div: Out << "/"; break;
1065 case Instruction::Rem: Out << "%"; break;
1066 case Instruction::And: Out << " & "; break;
1067 case Instruction::Or: Out << " | "; break;
1068 case Instruction::Xor: Out << " ^ "; break;
1069 case Instruction::SetEQ: Out << " == "; break;
1070 case Instruction::SetNE: Out << " != "; break;
1071 case Instruction::SetLE: Out << " <= "; break;
1072 case Instruction::SetGE: Out << " >= "; break;
1073 case Instruction::SetLT: Out << " < "; break;
1074 case Instruction::SetGT: Out << " > "; break;
1075 case Instruction::Shl : Out << " << "; break;
1076 case Instruction::Shr : Out << " >> "; break;
1077 default: std::cerr << "Invalid operator type!" << I; abort();
1080 writeOperand(I.getOperand(1));
1087 void CWriter::visitCastInst(CastInst &I) {
1088 if (I.getType() == Type::BoolTy) {
1090 writeOperand(I.getOperand(0));
1095 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1097 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1098 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1099 // Avoid "cast to pointer from integer of different size" warnings
1103 writeOperand(I.getOperand(0));
1106 void CWriter::visitCallInst(CallInst &I) {
1107 // Handle intrinsic function calls first...
1108 if (Function *F = I.getCalledFunction())
1109 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID()) {
1111 default: assert(0 && "Unknown LLVM intrinsic!");
1112 case LLVMIntrinsic::va_start:
1113 Out << "va_start((va_list)*";
1114 writeOperand(I.getOperand(1));
1116 // Output the last argument to the enclosing function...
1117 writeOperand(&I.getParent()->getParent()->aback());
1120 case LLVMIntrinsic::va_end:
1121 Out << "va_end((va_list)*";
1122 writeOperand(I.getOperand(1));
1125 case LLVMIntrinsic::va_copy:
1126 Out << "va_copy((va_list)*";
1127 writeOperand(I.getOperand(1));
1128 Out << ", (va_list)";
1129 writeOperand(I.getOperand(2));
1133 case LLVMIntrinsic::setjmp:
1134 Out << "setjmp((jmp_buf)";
1135 writeOperand(I.getOperand(1));
1138 case LLVMIntrinsic::longjmp:
1139 Out << "longjmp((jmp_buf)";
1140 writeOperand(I.getOperand(1));
1142 writeOperand(I.getOperand(2));
1148 const PointerType *PTy = cast<PointerType>(I.getCalledValue()->getType());
1149 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1150 const Type *RetTy = FTy->getReturnType();
1152 writeOperand(I.getOperand(0));
1155 if (I.getNumOperands() > 1) {
1156 writeOperand(I.getOperand(1));
1158 for (unsigned op = 2, Eop = I.getNumOperands(); op != Eop; ++op) {
1160 writeOperand(I.getOperand(op));
1166 void CWriter::visitMallocInst(MallocInst &I) {
1168 printType(Out, I.getType());
1169 Out << ")malloc(sizeof(";
1170 printType(Out, I.getType()->getElementType());
1173 if (I.isArrayAllocation()) {
1175 writeOperand(I.getOperand(0));
1180 void CWriter::visitAllocaInst(AllocaInst &I) {
1182 printType(Out, I.getType());
1183 Out << ") alloca(sizeof(";
1184 printType(Out, I.getType()->getElementType());
1186 if (I.isArrayAllocation()) {
1188 writeOperand(I.getOperand(0));
1193 void CWriter::visitFreeInst(FreeInst &I) {
1195 writeOperand(I.getOperand(0));
1199 void CWriter::printIndexingExpression(Value *Ptr, User::op_iterator I,
1200 User::op_iterator E) {
1201 bool HasImplicitAddress = false;
1202 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1203 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1204 HasImplicitAddress = true;
1205 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1206 HasImplicitAddress = true;
1207 Ptr = CPR->getValue(); // Get to the global...
1208 } else if (isDirectAlloca(Ptr)) {
1209 HasImplicitAddress = true;
1213 if (!HasImplicitAddress)
1214 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1216 writeOperandInternal(Ptr);
1220 const Constant *CI = dyn_cast<Constant>(I);
1221 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1224 writeOperandInternal(Ptr);
1226 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1228 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1231 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1232 "Can only have implicit address with direct accessing");
1234 if (HasImplicitAddress) {
1236 } else if (CI && CI->isNullValue() && I+1 != E) {
1237 // Print out the -> operator if possible...
1238 if ((*(I+1))->getType() == Type::UByteTy) {
1239 Out << (HasImplicitAddress ? "." : "->");
1240 Out << "field" << cast<ConstantUInt>(*(I+1))->getValue();
1246 if ((*I)->getType() == Type::LongTy) {
1251 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
1255 void CWriter::visitLoadInst(LoadInst &I) {
1257 writeOperand(I.getOperand(0));
1260 void CWriter::visitStoreInst(StoreInst &I) {
1262 writeOperand(I.getPointerOperand());
1264 writeOperand(I.getOperand(0));
1267 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1269 printIndexingExpression(I.getPointerOperand(), I.idx_begin(), I.idx_end());
1272 void CWriter::visitVarArgInst(VarArgInst &I) {
1273 Out << "va_arg((va_list)*";
1274 writeOperand(I.getOperand(0));
1276 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1281 //===----------------------------------------------------------------------===//
1282 // External Interface declaration
1283 //===----------------------------------------------------------------------===//
1285 Pass *createWriteToCPass(std::ostream &o) { return new CWriter(o); }