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 // generateCompilerSpecificCode - This is where we add conditional compilation
576 // directives to cater to specific compilers as need be.
578 static void generateCompilerSpecificCode(std::ostream& Out) {
579 // Alloca is hard to get, and we don't want to include stdlib.h here...
580 Out << "/* get a declaration for alloca */\n"
582 << "extern void *__builtin_alloca(unsigned long);\n"
583 << "#define alloca(x) __builtin_alloca(x)\n"
585 << "#ifndef __FreeBSD__\n"
586 << "#include <alloca.h>\n"
590 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
591 // If we aren't being compiled with GCC, just drop these attributes.
592 Out << "#ifndef __GNUC__\n"
593 << "#define __attribute__(X)\n"
597 void CWriter::printModule(Module *M) {
598 // Calculate which global values have names that will collide when we throw
599 // away type information.
600 { // Scope to delete the FoundNames set when we are done with it...
601 std::set<std::string> FoundNames;
602 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
603 if (I->hasName()) // If the global has a name...
604 if (FoundNames.count(I->getName())) // And the name is already used
605 MangledGlobals.insert(I); // Mangle the name
607 FoundNames.insert(I->getName()); // Otherwise, keep track of name
609 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
610 if (I->hasName()) // If the global has a name...
611 if (FoundNames.count(I->getName())) // And the name is already used
612 MangledGlobals.insert(I); // Mangle the name
614 FoundNames.insert(I->getName()); // Otherwise, keep track of name
617 // get declaration for alloca
618 Out << "/* Provide Declarations */\n";
619 Out << "#include <stdarg.h>\n";
620 Out << "#include <setjmp.h>\n";
621 generateCompilerSpecificCode(Out);
623 // Provide a definition for `bool' if not compiling with a C++ compiler.
625 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
627 << "\n\n/* Support for floating point constants */\n"
628 << "typedef unsigned long long ConstantDoubleTy;\n"
629 << "typedef unsigned int ConstantFloatTy;\n"
631 << "\n\n/* Global Declarations */\n";
633 // First output all the declarations for the program, because C requires
634 // Functions & globals to be declared before they are used.
637 // Loop over the symbol table, emitting all named constants...
638 printSymbolTable(M->getSymbolTable());
640 // Global variable declarations...
642 Out << "\n/* External Global Variable Declarations */\n";
643 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
644 if (I->hasExternalLinkage()) {
646 printType(Out, I->getType()->getElementType(), getValueName(I));
652 // Function declarations
654 Out << "\n/* Function Declarations */\n";
656 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
657 // If the function is external and the name collides don't print it.
658 // Sometimes the bytecode likes to have multiple "declarations" for
659 // external functions
660 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
661 !I->getIntrinsicID()) {
662 printFunctionSignature(I, true);
668 // Print Malloc prototype if needed
670 Out << "\n/* Malloc to make sun happy */\n";
671 Out << "extern void * malloc(size_t);\n\n";
674 // Output the global variable declarations
676 Out << "\n\n/* Global Variable Declarations */\n";
677 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
678 if (!I->isExternal()) {
680 printType(Out, I->getType()->getElementType(), getValueName(I));
687 // Output the global variable definitions and contents...
689 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
690 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
691 if (!I->isExternal()) {
692 if (I->hasInternalLinkage())
694 printType(Out, I->getType()->getElementType(), getValueName(I));
695 if (I->hasLinkOnceLinkage())
696 Out << " __attribute__((common))";
697 if (!I->getInitializer()->isNullValue()) {
699 writeOperand(I->getInitializer());
705 // Output all of the functions...
707 Out << "\n\n/* Function Bodies */\n";
708 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
714 /// printSymbolTable - Run through symbol table looking for type names. If a
715 /// type name is found, emit it's declaration...
717 void CWriter::printSymbolTable(const SymbolTable &ST) {
718 // If there are no type names, exit early.
719 if (ST.find(Type::TypeTy) == ST.end())
722 // We are only interested in the type plane of the symbol table...
723 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
724 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
726 // Print out forward declarations for structure types before anything else!
727 Out << "/* Structure forward decls */\n";
728 for (; I != End; ++I)
729 if (const Type *STy = dyn_cast<StructType>(I->second)) {
730 std::string Name = "struct l_" + makeNameProper(I->first);
731 Out << Name << ";\n";
732 TypeNames.insert(std::make_pair(STy, Name));
737 // Now we can print out typedefs...
738 Out << "/* Typedefs */\n";
739 for (I = ST.type_begin(Type::TypeTy); I != End; ++I) {
740 const Type *Ty = cast<Type>(I->second);
741 std::string Name = "l_" + makeNameProper(I->first);
743 printType(Out, Ty, Name);
749 // Keep track of which structures have been printed so far...
750 std::set<const StructType *> StructPrinted;
752 // Loop over all structures then push them into the stack so they are
753 // printed in the correct order.
755 Out << "/* Structure contents */\n";
756 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
757 if (const StructType *STy = dyn_cast<StructType>(I->second))
758 printContainedStructs(STy, StructPrinted);
761 // Push the struct onto the stack and recursively push all structs
762 // this one depends on.
763 void CWriter::printContainedStructs(const Type *Ty,
764 std::set<const StructType*> &StructPrinted){
765 if (const StructType *STy = dyn_cast<StructType>(Ty)){
766 //Check to see if we have already printed this struct
767 if (StructPrinted.count(STy) == 0) {
768 // Print all contained types first...
769 for (StructType::ElementTypes::const_iterator
770 I = STy->getElementTypes().begin(),
771 E = STy->getElementTypes().end(); I != E; ++I) {
772 const Type *Ty1 = I->get();
773 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
774 printContainedStructs(*I, StructPrinted);
777 //Print structure type out..
778 StructPrinted.insert(STy);
779 std::string Name = TypeNames[STy];
780 printType(Out, STy, Name, true);
784 // If it is an array, check contained types and continue
785 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
786 const Type *Ty1 = ATy->getElementType();
787 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
788 printContainedStructs(Ty1, StructPrinted);
793 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
794 // If the program provides its own malloc prototype we don't need
795 // to include the general one.
796 if (getValueName(F) == "malloc")
799 if (F->hasInternalLinkage()) Out << "static ";
800 if (F->hasLinkOnceLinkage()) Out << "inline ";
802 // Loop over the arguments, printing them...
803 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
805 std::stringstream FunctionInnards;
807 // Print out the name...
808 FunctionInnards << getValueName(F) << "(";
810 if (!F->isExternal()) {
813 if (F->abegin()->hasName() || !Prototype)
814 ArgName = getValueName(F->abegin());
815 printType(FunctionInnards, F->afront().getType(), ArgName);
816 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
818 FunctionInnards << ", ";
819 if (I->hasName() || !Prototype)
820 ArgName = getValueName(I);
823 printType(FunctionInnards, I->getType(), ArgName);
827 // Loop over the arguments, printing them...
828 for (FunctionType::ParamTypes::const_iterator I =
829 FT->getParamTypes().begin(),
830 E = FT->getParamTypes().end(); I != E; ++I) {
831 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
832 printType(FunctionInnards, *I);
836 // Finish printing arguments... if this is a vararg function, print the ...,
837 // unless there are no known types, in which case, we just emit ().
839 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
840 if (FT->getParamTypes().size()) FunctionInnards << ", ";
841 FunctionInnards << "..."; // Output varargs portion of signature!
843 FunctionInnards << ")";
844 // Print out the return type and the entire signature for that matter
845 printType(Out, F->getReturnType(), FunctionInnards.str());
849 void CWriter::printFunction(Function *F) {
850 if (F->isExternal()) return;
852 Table->incorporateFunction(F);
854 printFunctionSignature(F, false);
857 // print local variable information for the function
858 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
859 if (const AllocaInst *AI = isDirectAlloca(*I)) {
861 printType(Out, AI->getAllocatedType(), getValueName(AI));
862 Out << "; /* Address exposed local */\n";
863 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
865 printType(Out, (*I)->getType(), getValueName(*I));
868 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
870 printType(Out, (*I)->getType(), getValueName(*I)+"__PHI_TEMPORARY");
877 // Scan the function for floating point constants. If any FP constant is used
878 // in the function, we want to redirect it here so that we do not depend on
879 // the precision of the printed form, unless the printed form preserves
882 unsigned FPCounter = 0;
883 for (constant_iterator I = constant_begin(F), E = constant_end(F); I != E;++I)
884 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
885 if ((!FPCSafeToPrint(FPC)) // Do not put in FPConstantMap if safe.
886 && (FPConstantMap.find(FPC) == FPConstantMap.end())) {
887 double Val = FPC->getValue();
889 FPConstantMap[FPC] = FPCounter; // Number the FP constants
891 if (FPC->getType() == Type::DoubleTy)
892 Out << " const ConstantDoubleTy FloatConstant" << FPCounter++
893 << " = 0x" << std::hex << *(unsigned long long*)&Val << std::dec
894 << "; /* " << Val << " */\n";
895 else if (FPC->getType() == Type::FloatTy) {
897 Out << " const ConstantFloatTy FloatConstant" << FPCounter++
898 << " = 0x" << std::hex << *(unsigned*)&fVal << std::dec
899 << "; /* " << Val << " */\n";
901 assert(0 && "Unknown float type!");
906 // print the basic blocks
907 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
908 BasicBlock *Prev = BB->getPrev();
910 // Don't print the label for the basic block if there are no uses, or if the
911 // only terminator use is the precessor basic block's terminator. We have
912 // to scan the use list because PHI nodes use basic blocks too but do not
913 // require a label to be generated.
915 bool NeedsLabel = false;
916 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
918 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
919 if (TI != Prev->getTerminator() ||
920 isa<SwitchInst>(Prev->getTerminator())) {
925 if (NeedsLabel) Out << getValueName(BB) << ":\n";
927 // Output all of the instructions in the basic block...
928 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
929 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
930 if (II->getType() != Type::VoidTy)
939 // Don't emit prefix or suffix for the terminator...
940 visit(*BB->getTerminator());
944 Table->purgeFunction();
945 FPConstantMap.clear();
948 // Specific Instruction type classes... note that all of the casts are
949 // neccesary because we use the instruction classes as opaque types...
951 void CWriter::visitReturnInst(ReturnInst &I) {
952 // Don't output a void return if this is the last basic block in the function
953 if (I.getNumOperands() == 0 &&
954 &*--I.getParent()->getParent()->end() == I.getParent() &&
955 !I.getParent()->size() == 1) {
960 if (I.getNumOperands()) {
962 writeOperand(I.getOperand(0));
967 void CWriter::visitSwitchInst(SwitchInst &SI) {
969 writeOperand(SI.getOperand(0));
970 Out << ") {\n default:\n";
971 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
973 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
975 writeOperand(SI.getOperand(i));
977 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
978 printBranchToBlock(SI.getParent(), Succ, 2);
979 if (Succ == SI.getParent()->getNext())
986 static bool isGotoCodeNeccessary(BasicBlock *From, BasicBlock *To) {
987 // If PHI nodes need copies, we need the copy code...
988 if (isa<PHINode>(To->front()) ||
989 From->getNext() != To) // Not directly successor, need goto
992 // Otherwise we don't need the code.
996 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
998 for (BasicBlock::iterator I = Succ->begin();
999 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1000 // now we have to do the printing
1001 Out << std::string(Indent, ' ');
1002 Out << " " << getValueName(I) << "__PHI_TEMPORARY = ";
1003 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1004 Out << "; /* for PHI node */\n";
1007 if (CurBB->getNext() != Succ) {
1008 Out << std::string(Indent, ' ') << " goto ";
1014 // Brach instruction printing - Avoid printing out a brach to a basic block that
1015 // immediately succeeds the current one.
1017 void CWriter::visitBranchInst(BranchInst &I) {
1018 if (I.isConditional()) {
1019 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(0))) {
1021 writeOperand(I.getCondition());
1024 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1026 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(1))) {
1027 Out << " } else {\n";
1028 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1031 // First goto not neccesary, assume second one is...
1033 writeOperand(I.getCondition());
1036 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1041 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1046 // PHI nodes get copied into temporary values at the end of predecessor basic
1047 // blocks. We now need to copy these temporary values into the REAL value for
1049 void CWriter::visitPHINode(PHINode &I) {
1051 Out << "__PHI_TEMPORARY";
1055 void CWriter::visitBinaryOperator(Instruction &I) {
1056 // binary instructions, shift instructions, setCond instructions.
1057 assert(!isa<PointerType>(I.getType()));
1059 // We must cast the results of binary operations which might be promoted.
1060 bool needsCast = false;
1061 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1062 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1063 || (I.getType() == Type::FloatTy)) {
1066 printType(Out, I.getType(), "", false, false);
1070 writeOperand(I.getOperand(0));
1072 switch (I.getOpcode()) {
1073 case Instruction::Add: Out << " + "; break;
1074 case Instruction::Sub: Out << " - "; break;
1075 case Instruction::Mul: Out << "*"; break;
1076 case Instruction::Div: Out << "/"; break;
1077 case Instruction::Rem: Out << "%"; break;
1078 case Instruction::And: Out << " & "; break;
1079 case Instruction::Or: Out << " | "; break;
1080 case Instruction::Xor: Out << " ^ "; break;
1081 case Instruction::SetEQ: Out << " == "; break;
1082 case Instruction::SetNE: Out << " != "; break;
1083 case Instruction::SetLE: Out << " <= "; break;
1084 case Instruction::SetGE: Out << " >= "; break;
1085 case Instruction::SetLT: Out << " < "; break;
1086 case Instruction::SetGT: Out << " > "; break;
1087 case Instruction::Shl : Out << " << "; break;
1088 case Instruction::Shr : Out << " >> "; break;
1089 default: std::cerr << "Invalid operator type!" << I; abort();
1092 writeOperand(I.getOperand(1));
1099 void CWriter::visitCastInst(CastInst &I) {
1100 if (I.getType() == Type::BoolTy) {
1102 writeOperand(I.getOperand(0));
1107 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1109 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1110 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1111 // Avoid "cast to pointer from integer of different size" warnings
1115 writeOperand(I.getOperand(0));
1118 void CWriter::visitCallInst(CallInst &I) {
1119 // Handle intrinsic function calls first...
1120 if (Function *F = I.getCalledFunction())
1121 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID()) {
1123 default: assert(0 && "Unknown LLVM intrinsic!");
1124 case LLVMIntrinsic::va_start:
1125 Out << "va_start((va_list)*";
1126 writeOperand(I.getOperand(1));
1128 // Output the last argument to the enclosing function...
1129 writeOperand(&I.getParent()->getParent()->aback());
1132 case LLVMIntrinsic::va_end:
1133 Out << "va_end((va_list)*";
1134 writeOperand(I.getOperand(1));
1137 case LLVMIntrinsic::va_copy:
1138 Out << "va_copy((va_list)*";
1139 writeOperand(I.getOperand(1));
1140 Out << ", (va_list)";
1141 writeOperand(I.getOperand(2));
1145 case LLVMIntrinsic::setjmp:
1146 Out << "setjmp((jmp_buf)";
1147 writeOperand(I.getOperand(1));
1150 case LLVMIntrinsic::longjmp:
1151 Out << "longjmp((jmp_buf)";
1152 writeOperand(I.getOperand(1));
1154 writeOperand(I.getOperand(2));
1160 const PointerType *PTy = cast<PointerType>(I.getCalledValue()->getType());
1161 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1162 const Type *RetTy = FTy->getReturnType();
1164 writeOperand(I.getOperand(0));
1167 if (I.getNumOperands() > 1) {
1168 writeOperand(I.getOperand(1));
1170 for (unsigned op = 2, Eop = I.getNumOperands(); op != Eop; ++op) {
1172 writeOperand(I.getOperand(op));
1178 void CWriter::visitMallocInst(MallocInst &I) {
1180 printType(Out, I.getType());
1181 Out << ")malloc(sizeof(";
1182 printType(Out, I.getType()->getElementType());
1185 if (I.isArrayAllocation()) {
1187 writeOperand(I.getOperand(0));
1192 void CWriter::visitAllocaInst(AllocaInst &I) {
1194 printType(Out, I.getType());
1195 Out << ") alloca(sizeof(";
1196 printType(Out, I.getType()->getElementType());
1198 if (I.isArrayAllocation()) {
1200 writeOperand(I.getOperand(0));
1205 void CWriter::visitFreeInst(FreeInst &I) {
1207 writeOperand(I.getOperand(0));
1211 void CWriter::printIndexingExpression(Value *Ptr, User::op_iterator I,
1212 User::op_iterator E) {
1213 bool HasImplicitAddress = false;
1214 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1215 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1216 HasImplicitAddress = true;
1217 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1218 HasImplicitAddress = true;
1219 Ptr = CPR->getValue(); // Get to the global...
1220 } else if (isDirectAlloca(Ptr)) {
1221 HasImplicitAddress = true;
1225 if (!HasImplicitAddress)
1226 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1228 writeOperandInternal(Ptr);
1232 const Constant *CI = dyn_cast<Constant>(I);
1233 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1236 writeOperandInternal(Ptr);
1238 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1240 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1243 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1244 "Can only have implicit address with direct accessing");
1246 if (HasImplicitAddress) {
1248 } else if (CI && CI->isNullValue() && I+1 != E) {
1249 // Print out the -> operator if possible...
1250 if ((*(I+1))->getType() == Type::UByteTy) {
1251 Out << (HasImplicitAddress ? "." : "->");
1252 Out << "field" << cast<ConstantUInt>(*(I+1))->getValue();
1258 if ((*I)->getType() == Type::LongTy) {
1263 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
1267 void CWriter::visitLoadInst(LoadInst &I) {
1269 writeOperand(I.getOperand(0));
1272 void CWriter::visitStoreInst(StoreInst &I) {
1274 writeOperand(I.getPointerOperand());
1276 writeOperand(I.getOperand(0));
1279 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1281 printIndexingExpression(I.getPointerOperand(), I.idx_begin(), I.idx_end());
1284 void CWriter::visitVarArgInst(VarArgInst &I) {
1285 Out << "va_arg((va_list)*";
1286 writeOperand(I.getOperand(0));
1288 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1293 //===----------------------------------------------------------------------===//
1294 // External Interface declaration
1295 //===----------------------------------------------------------------------===//
1297 Pass *createWriteToCPass(std::ostream &o) { return new CWriter(o); }