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:
373 printConstant(CE->getOperand(0));
375 printConstant(CE->getOperand(1));
378 case Instruction::Sub:
380 printConstant(CE->getOperand(0));
382 printConstant(CE->getOperand(1));
387 std::cerr << "CWriter Error: Unhandled constant expression: "
393 switch (CPV->getType()->getPrimitiveID()) {
395 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
396 case Type::SByteTyID:
397 case Type::ShortTyID:
398 Out << cast<ConstantSInt>(CPV)->getValue(); break;
400 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
401 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
403 Out << cast<ConstantSInt>(CPV)->getValue();
407 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
409 case Type::UByteTyID:
410 case Type::UShortTyID:
411 Out << cast<ConstantUInt>(CPV)->getValue(); break;
413 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
414 case Type::ULongTyID:
415 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
417 case Type::FloatTyID:
418 case Type::DoubleTyID: {
419 ConstantFP *FPC = cast<ConstantFP>(CPV);
420 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
421 if (I != FPConstantMap.end()) {
422 // Because of FP precision problems we must load from a stack allocated
423 // value that holds the value in hex.
424 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
425 << "*)&FloatConstant" << I->second << ")";
427 if (FPCSafeToPrint (FPC)) {
428 Out << ftostr (FPC->getValue ());
430 Out << FPC->getValue(); // Who knows? Give it our best shot...
436 case Type::ArrayTyID:
437 printConstantArray(cast<ConstantArray>(CPV));
440 case Type::StructTyID: {
442 if (CPV->getNumOperands()) {
444 printConstant(cast<Constant>(CPV->getOperand(0)));
445 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
447 printConstant(cast<Constant>(CPV->getOperand(i)));
454 case Type::PointerTyID:
455 if (isa<ConstantPointerNull>(CPV)) {
457 printType(Out, CPV->getType());
458 Out << ")/*NULL*/0)";
460 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
461 writeOperand(CPR->getValue());
466 std::cerr << "Unknown constant type: " << CPV << "\n";
471 void CWriter::writeOperandInternal(Value *Operand) {
472 if (Instruction *I = dyn_cast<Instruction>(Operand))
473 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
474 // Should we inline this instruction to build a tree?
481 if (Constant *CPV = dyn_cast<Constant>(Operand)) {
484 Out << Mang->getValueName(Operand);
488 void CWriter::writeOperand(Value *Operand) {
489 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
490 Out << "(&"; // Global variables are references as their addresses by llvm
492 writeOperandInternal(Operand);
494 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
498 // nameAllUsedStructureTypes - If there are structure types in the module that
499 // are used but do not have names assigned to them in the symbol table yet then
500 // we assign them names now.
502 bool CWriter::nameAllUsedStructureTypes(Module &M) {
503 // Get a set of types that are used by the program...
504 std::set<const Type *> UT = getAnalysis<FindUsedTypes>().getTypes();
506 // Loop over the module symbol table, removing types from UT that are already
509 SymbolTable &MST = M.getSymbolTable();
510 if (MST.find(Type::TypeTy) != MST.end())
511 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
512 E = MST.type_end(Type::TypeTy); I != E; ++I)
513 UT.erase(cast<Type>(I->second));
515 // UT now contains types that are not named. Loop over it, naming structure
518 bool Changed = false;
519 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
521 if (const StructType *ST = dyn_cast<StructType>(*I)) {
522 ((Value*)ST)->setName("unnamed", &MST);
528 // generateCompilerSpecificCode - This is where we add conditional compilation
529 // directives to cater to specific compilers as need be.
531 static void generateCompilerSpecificCode(std::ostream& Out) {
532 // Alloca is hard to get, and we don't want to include stdlib.h here...
533 Out << "/* get a declaration for alloca */\n"
535 << "extern void *__builtin_alloca(unsigned long);\n"
536 << "#define alloca(x) __builtin_alloca(x)\n"
538 << "#ifndef __FreeBSD__\n"
539 << "#include <alloca.h>\n"
543 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
544 // If we aren't being compiled with GCC, just drop these attributes.
545 Out << "#ifndef __GNUC__ /* Can only support \"linkonce\" vars with GCC */\n"
546 << "#define __attribute__(X)\n"
550 void CWriter::printModule(Module *M) {
551 // Calculate which global values have names that will collide when we throw
552 // away type information.
553 { // Scope to delete the FoundNames set when we are done with it...
554 std::set<std::string> FoundNames;
555 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
556 if (I->hasName()) // If the global has a name...
557 if (FoundNames.count(I->getName())) // And the name is already used
558 MangledGlobals.insert(I); // Mangle the name
560 FoundNames.insert(I->getName()); // Otherwise, keep track of name
562 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
563 if (I->hasName()) // If the global has a name...
564 if (FoundNames.count(I->getName())) // And the name is already used
565 MangledGlobals.insert(I); // Mangle the name
567 FoundNames.insert(I->getName()); // Otherwise, keep track of name
570 // get declaration for alloca
571 Out << "/* Provide Declarations */\n";
572 Out << "#include <stdarg.h>\n";
573 Out << "#include <setjmp.h>\n";
574 generateCompilerSpecificCode(Out);
576 // Provide a definition for `bool' if not compiling with a C++ compiler.
578 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
580 << "\n\n/* Support for floating point constants */\n"
581 << "typedef unsigned long long ConstantDoubleTy;\n"
582 << "typedef unsigned int ConstantFloatTy;\n"
584 << "\n\n/* Support for the invoke instruction */\n"
585 << "extern struct __llvm_jmpbuf_list_t {\n"
586 << " jmp_buf buf; struct __llvm_jmpbuf_list_t *next;\n"
587 << "} *__llvm_jmpbuf_list;\n"
589 << "\n\n/* Global Declarations */\n";
591 // First output all the declarations for the program, because C requires
592 // Functions & globals to be declared before they are used.
595 // Loop over the symbol table, emitting all named constants...
596 printSymbolTable(M->getSymbolTable());
598 // Global variable declarations...
600 Out << "\n/* External Global Variable Declarations */\n";
601 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
602 if (I->hasExternalLinkage()) {
604 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
610 // Function declarations
612 Out << "\n/* Function Declarations */\n";
614 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
615 // If the function is external and the name collides don't print it.
616 // Sometimes the bytecode likes to have multiple "declarations" for
617 // external functions
618 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
619 !I->getIntrinsicID()) {
620 printFunctionSignature(I, true);
626 // Print Malloc prototype if needed
628 Out << "\n/* Malloc to make sun happy */\n";
629 Out << "extern void * malloc();\n\n";
632 // Output the global variable declarations
634 Out << "\n\n/* Global Variable Declarations */\n";
635 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
636 if (!I->isExternal()) {
638 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
644 // Output the global variable definitions and contents...
646 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
647 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
648 if (!I->isExternal()) {
649 if (I->hasInternalLinkage())
651 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
652 if (I->hasLinkOnceLinkage())
653 Out << " __attribute__((common))";
654 if (!I->getInitializer()->isNullValue()) {
656 writeOperand(I->getInitializer());
662 // Output all of the functions...
663 emittedInvoke = false;
665 Out << "\n\n/* Function Bodies */\n";
666 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
670 // If the program included an invoke instruction, we need to output the
671 // support code for it here!
673 Out << "\n/* More support for the invoke instruction */\n"
674 << "struct __llvm_jmpbuf_list_t *__llvm_jmpbuf_list "
675 << "__attribute__((common)) = 0;\n";
680 /// printSymbolTable - Run through symbol table looking for type names. If a
681 /// type name is found, emit it's declaration...
683 void CWriter::printSymbolTable(const SymbolTable &ST) {
684 // If there are no type names, exit early.
685 if (ST.find(Type::TypeTy) == ST.end())
688 // We are only interested in the type plane of the symbol table...
689 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
690 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
692 // Print out forward declarations for structure types before anything else!
693 Out << "/* Structure forward decls */\n";
694 for (; I != End; ++I)
695 if (const Type *STy = dyn_cast<StructType>(I->second)) {
696 std::string Name = "struct l_" + Mangler::makeNameProper(I->first);
697 Out << Name << ";\n";
698 TypeNames.insert(std::make_pair(STy, Name));
703 // Now we can print out typedefs...
704 Out << "/* Typedefs */\n";
705 for (I = ST.type_begin(Type::TypeTy); I != End; ++I) {
706 const Type *Ty = cast<Type>(I->second);
707 std::string Name = "l_" + Mangler::makeNameProper(I->first);
709 printType(Out, Ty, Name);
715 // Keep track of which structures have been printed so far...
716 std::set<const StructType *> StructPrinted;
718 // Loop over all structures then push them into the stack so they are
719 // printed in the correct order.
721 Out << "/* Structure contents */\n";
722 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
723 if (const StructType *STy = dyn_cast<StructType>(I->second))
724 printContainedStructs(STy, StructPrinted);
727 // Push the struct onto the stack and recursively push all structs
728 // this one depends on.
729 void CWriter::printContainedStructs(const Type *Ty,
730 std::set<const StructType*> &StructPrinted){
731 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
732 //Check to see if we have already printed this struct
733 if (StructPrinted.count(STy) == 0) {
734 // Print all contained types first...
735 for (StructType::ElementTypes::const_iterator
736 I = STy->getElementTypes().begin(),
737 E = STy->getElementTypes().end(); I != E; ++I) {
738 const Type *Ty1 = I->get();
739 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
740 printContainedStructs(*I, StructPrinted);
743 //Print structure type out..
744 StructPrinted.insert(STy);
745 std::string Name = TypeNames[STy];
746 printType(Out, STy, Name, true);
750 // If it is an array, check contained types and continue
751 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
752 const Type *Ty1 = ATy->getElementType();
753 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
754 printContainedStructs(Ty1, StructPrinted);
759 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
760 // If the program provides its own malloc prototype we don't need
761 // to include the general one.
762 if (Mang->getValueName(F) == "malloc")
765 if (F->hasInternalLinkage()) Out << "static ";
766 if (F->hasLinkOnceLinkage()) Out << "inline ";
768 // Loop over the arguments, printing them...
769 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
771 std::stringstream FunctionInnards;
773 // Print out the name...
774 FunctionInnards << Mang->getValueName(F) << "(";
776 if (!F->isExternal()) {
779 if (F->abegin()->hasName() || !Prototype)
780 ArgName = Mang->getValueName(F->abegin());
781 printType(FunctionInnards, F->afront().getType(), ArgName);
782 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
784 FunctionInnards << ", ";
785 if (I->hasName() || !Prototype)
786 ArgName = Mang->getValueName(I);
789 printType(FunctionInnards, I->getType(), ArgName);
793 // Loop over the arguments, printing them...
794 for (FunctionType::ParamTypes::const_iterator I =
795 FT->getParamTypes().begin(),
796 E = FT->getParamTypes().end(); I != E; ++I) {
797 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
798 printType(FunctionInnards, *I);
802 // Finish printing arguments... if this is a vararg function, print the ...,
803 // unless there are no known types, in which case, we just emit ().
805 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
806 if (FT->getParamTypes().size()) FunctionInnards << ", ";
807 FunctionInnards << "..."; // Output varargs portion of signature!
809 FunctionInnards << ")";
810 // Print out the return type and the entire signature for that matter
811 printType(Out, F->getReturnType(), FunctionInnards.str());
814 void CWriter::printFunction(Function *F) {
815 if (F->isExternal()) return;
817 printFunctionSignature(F, false);
820 // print local variable information for the function
821 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
822 if (const AllocaInst *AI = isDirectAlloca(*I)) {
824 printType(Out, AI->getAllocatedType(), Mang->getValueName(AI));
825 Out << "; /* Address exposed local */\n";
826 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
828 printType(Out, (*I)->getType(), Mang->getValueName(*I));
831 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
833 printType(Out, (*I)->getType(), Mang->getValueName(*I)+"__PHI_TEMPORARY");
840 // Scan the function for floating point constants. If any FP constant is used
841 // in the function, we want to redirect it here so that we do not depend on
842 // the precision of the printed form, unless the printed form preserves
845 unsigned FPCounter = 0;
846 for (constant_iterator I = constant_begin(F), E = constant_end(F); I != E;++I)
847 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
848 if ((!FPCSafeToPrint(FPC)) // Do not put in FPConstantMap if safe.
849 && (FPConstantMap.find(FPC) == FPConstantMap.end())) {
850 double Val = FPC->getValue();
852 FPConstantMap[FPC] = FPCounter; // Number the FP constants
854 if (FPC->getType() == Type::DoubleTy)
855 Out << " const ConstantDoubleTy FloatConstant" << FPCounter++
856 << " = 0x" << std::hex << *(unsigned long long*)&Val << std::dec
857 << "; /* " << Val << " */\n";
858 else if (FPC->getType() == Type::FloatTy) {
860 Out << " const ConstantFloatTy FloatConstant" << FPCounter++
861 << " = 0x" << std::hex << *(unsigned*)&fVal << std::dec
862 << "; /* " << Val << " */\n";
864 assert(0 && "Unknown float type!");
869 // print the basic blocks
870 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
871 BasicBlock *Prev = BB->getPrev();
873 // Don't print the label for the basic block if there are no uses, or if the
874 // only terminator use is the precessor basic block's terminator. We have
875 // to scan the use list because PHI nodes use basic blocks too but do not
876 // require a label to be generated.
878 bool NeedsLabel = false;
879 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
881 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
882 if (TI != Prev->getTerminator() ||
883 isa<SwitchInst>(Prev->getTerminator())) {
888 if (NeedsLabel) Out << Mang->getValueName(BB) << ":\n";
890 // Output all of the instructions in the basic block...
891 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
892 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
893 if (II->getType() != Type::VoidTy)
902 // Don't emit prefix or suffix for the terminator...
903 visit(*BB->getTerminator());
907 FPConstantMap.clear();
910 // Specific Instruction type classes... note that all of the casts are
911 // neccesary because we use the instruction classes as opaque types...
913 void CWriter::visitReturnInst(ReturnInst &I) {
914 // Don't output a void return if this is the last basic block in the function
915 if (I.getNumOperands() == 0 &&
916 &*--I.getParent()->getParent()->end() == I.getParent() &&
917 !I.getParent()->size() == 1) {
922 if (I.getNumOperands()) {
924 writeOperand(I.getOperand(0));
929 void CWriter::visitSwitchInst(SwitchInst &SI) {
931 writeOperand(SI.getOperand(0));
932 Out << ") {\n default:\n";
933 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
935 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
937 writeOperand(SI.getOperand(i));
939 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
940 printBranchToBlock(SI.getParent(), Succ, 2);
941 if (Succ == SI.getParent()->getNext())
947 void CWriter::visitInvokeInst(InvokeInst &II) {
949 << " struct __llvm_jmpbuf_list_t Entry;\n"
950 << " Entry.next = __llvm_jmpbuf_list;\n"
951 << " if (setjmp(Entry.buf)) {\n"
952 << " __llvm_jmpbuf_list = Entry.next;\n";
953 printBranchToBlock(II.getParent(), II.getExceptionalDest(), 4);
955 << " __llvm_jmpbuf_list = &Entry;\n"
959 << " __llvm_jmpbuf_list = Entry.next;\n"
961 printBranchToBlock(II.getParent(), II.getNormalDest(), 0);
962 emittedInvoke = true;
966 static bool isGotoCodeNeccessary(BasicBlock *From, BasicBlock *To) {
967 // If PHI nodes need copies, we need the copy code...
968 if (isa<PHINode>(To->front()) ||
969 From->getNext() != To) // Not directly successor, need goto
972 // Otherwise we don't need the code.
976 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
978 for (BasicBlock::iterator I = Succ->begin();
979 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
980 // now we have to do the printing
981 Out << std::string(Indent, ' ');
982 Out << " " << Mang->getValueName(I) << "__PHI_TEMPORARY = ";
983 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
984 Out << "; /* for PHI node */\n";
987 if (CurBB->getNext() != Succ) {
988 Out << std::string(Indent, ' ') << " goto ";
994 // Brach instruction printing - Avoid printing out a brach to a basic block that
995 // immediately succeeds the current one.
997 void CWriter::visitBranchInst(BranchInst &I) {
998 if (I.isConditional()) {
999 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(0))) {
1001 writeOperand(I.getCondition());
1004 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1006 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(1))) {
1007 Out << " } else {\n";
1008 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1011 // First goto not neccesary, assume second one is...
1013 writeOperand(I.getCondition());
1016 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1021 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1026 // PHI nodes get copied into temporary values at the end of predecessor basic
1027 // blocks. We now need to copy these temporary values into the REAL value for
1029 void CWriter::visitPHINode(PHINode &I) {
1031 Out << "__PHI_TEMPORARY";
1035 void CWriter::visitBinaryOperator(Instruction &I) {
1036 // binary instructions, shift instructions, setCond instructions.
1037 assert(!isa<PointerType>(I.getType()));
1039 // We must cast the results of binary operations which might be promoted.
1040 bool needsCast = false;
1041 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1042 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1043 || (I.getType() == Type::FloatTy)) {
1046 printType(Out, I.getType(), "", false, false);
1050 writeOperand(I.getOperand(0));
1052 switch (I.getOpcode()) {
1053 case Instruction::Add: Out << " + "; break;
1054 case Instruction::Sub: Out << " - "; break;
1055 case Instruction::Mul: Out << "*"; break;
1056 case Instruction::Div: Out << "/"; break;
1057 case Instruction::Rem: Out << "%"; break;
1058 case Instruction::And: Out << " & "; break;
1059 case Instruction::Or: Out << " | "; break;
1060 case Instruction::Xor: Out << " ^ "; break;
1061 case Instruction::SetEQ: Out << " == "; break;
1062 case Instruction::SetNE: Out << " != "; break;
1063 case Instruction::SetLE: Out << " <= "; break;
1064 case Instruction::SetGE: Out << " >= "; break;
1065 case Instruction::SetLT: Out << " < "; break;
1066 case Instruction::SetGT: Out << " > "; break;
1067 case Instruction::Shl : Out << " << "; break;
1068 case Instruction::Shr : Out << " >> "; break;
1069 default: std::cerr << "Invalid operator type!" << I; abort();
1072 writeOperand(I.getOperand(1));
1079 void CWriter::visitCastInst(CastInst &I) {
1080 if (I.getType() == Type::BoolTy) {
1082 writeOperand(I.getOperand(0));
1087 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1089 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1090 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1091 // Avoid "cast to pointer from integer of different size" warnings
1095 writeOperand(I.getOperand(0));
1098 void CWriter::visitCallInst(CallInst &I) {
1099 // Handle intrinsic function calls first...
1100 if (Function *F = I.getCalledFunction())
1101 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID()) {
1103 default: assert(0 && "Unknown LLVM intrinsic!");
1104 case LLVMIntrinsic::va_start:
1105 Out << "va_start((va_list)*";
1106 writeOperand(I.getOperand(1));
1108 // Output the last argument to the enclosing function...
1109 writeOperand(&I.getParent()->getParent()->aback());
1112 case LLVMIntrinsic::va_end:
1113 Out << "va_end((va_list)*";
1114 writeOperand(I.getOperand(1));
1117 case LLVMIntrinsic::va_copy:
1118 Out << "va_copy((va_list)*";
1119 writeOperand(I.getOperand(1));
1120 Out << ", (va_list)";
1121 writeOperand(I.getOperand(2));
1125 case LLVMIntrinsic::setjmp:
1127 Out << "setjmp(*(jmp_buf*)";
1128 writeOperand(I.getOperand(1));
1132 // For right now, we don't really support non-local jumps. So
1133 // make setjmp() always evaluate to zero for now.
1138 case LLVMIntrinsic::longjmp:
1140 Out << "longjmp(*(jmp_buf*)";
1141 writeOperand(I.getOperand(1));
1143 writeOperand(I.getOperand(2));
1147 // For right now, we don't really support non-local jumps. So
1148 // make longjmp() abort the program.
1158 void CWriter::visitCallSite(CallSite CS) {
1159 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1160 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1161 const Type *RetTy = FTy->getReturnType();
1163 writeOperand(CS.getCalledValue());
1166 if (CS.arg_begin() != CS.arg_end()) {
1167 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1170 for (++AI; AI != AE; ++AI) {
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); }