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);
120 void visitUnwindInst(UnwindInst &I);
122 void visitPHINode(PHINode &I);
123 void visitBinaryOperator(Instruction &I);
125 void visitCastInst (CastInst &I);
126 void visitCallInst (CallInst &I);
127 void visitCallSite (CallSite CS);
128 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
130 void visitMallocInst(MallocInst &I);
131 void visitAllocaInst(AllocaInst &I);
132 void visitFreeInst (FreeInst &I);
133 void visitLoadInst (LoadInst &I);
134 void visitStoreInst (StoreInst &I);
135 void visitGetElementPtrInst(GetElementPtrInst &I);
136 void visitVarArgInst(VarArgInst &I);
138 void visitInstruction(Instruction &I) {
139 std::cerr << "C Writer does not know about " << I;
143 void outputLValue(Instruction *I) {
144 Out << " " << Mang->getValueName(I) << " = ";
146 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
148 void printIndexingExpression(Value *Ptr, User::op_iterator I,
149 User::op_iterator E);
153 // A pointer type should not use parens around *'s alone, e.g., (**)
154 inline bool ptrTypeNameNeedsParens(const std::string &NameSoFar) {
155 return (NameSoFar.find_last_not_of('*') != std::string::npos);
158 // Pass the Type* and the variable name and this prints out the variable
161 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
162 const std::string &NameSoFar,
163 bool IgnoreName, bool namedContext) {
164 if (Ty->isPrimitiveType())
165 switch (Ty->getPrimitiveID()) {
166 case Type::VoidTyID: return Out << "void " << NameSoFar;
167 case Type::BoolTyID: return Out << "bool " << NameSoFar;
168 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
169 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
170 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
171 case Type::ShortTyID: return Out << "short " << NameSoFar;
172 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
173 case Type::IntTyID: return Out << "int " << NameSoFar;
174 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
175 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
176 case Type::FloatTyID: return Out << "float " << NameSoFar;
177 case Type::DoubleTyID: return Out << "double " << NameSoFar;
179 std::cerr << "Unknown primitive type: " << Ty << "\n";
183 // Check to see if the type is named.
184 if (!IgnoreName || isa<OpaqueType>(Ty)) {
185 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
186 if (I != TypeNames.end()) return Out << I->second << " " << NameSoFar;
189 switch (Ty->getPrimitiveID()) {
190 case Type::FunctionTyID: {
191 const FunctionType *MTy = cast<FunctionType>(Ty);
192 std::stringstream FunctionInnards;
193 FunctionInnards << " (" << NameSoFar << ") (";
194 for (FunctionType::ParamTypes::const_iterator
195 I = MTy->getParamTypes().begin(),
196 E = MTy->getParamTypes().end(); I != E; ++I) {
197 if (I != MTy->getParamTypes().begin())
198 FunctionInnards << ", ";
199 printType(FunctionInnards, *I, "");
201 if (MTy->isVarArg()) {
202 if (!MTy->getParamTypes().empty())
203 FunctionInnards << ", ...";
204 } else if (MTy->getParamTypes().empty()) {
205 FunctionInnards << "void";
207 FunctionInnards << ")";
208 std::string tstr = FunctionInnards.str();
209 printType(Out, MTy->getReturnType(), tstr);
212 case Type::StructTyID: {
213 const StructType *STy = cast<StructType>(Ty);
214 Out << NameSoFar + " {\n";
216 for (StructType::ElementTypes::const_iterator
217 I = STy->getElementTypes().begin(),
218 E = STy->getElementTypes().end(); I != E; ++I) {
220 printType(Out, *I, "field" + utostr(Idx++));
226 case Type::PointerTyID: {
227 const PointerType *PTy = cast<PointerType>(Ty);
228 std::string ptrName = "*" + NameSoFar;
230 // Do not need parens around "* NameSoFar" if NameSoFar consists only
231 // of zero or more '*' chars *and* this is not an unnamed pointer type
232 // such as the result type in a cast statement. Otherwise, enclose in ( ).
233 if (ptrTypeNameNeedsParens(NameSoFar) || !namedContext ||
234 PTy->getElementType()->getPrimitiveID() == Type::ArrayTyID)
235 ptrName = "(" + ptrName + ")"; //
237 return printType(Out, PTy->getElementType(), ptrName);
240 case Type::ArrayTyID: {
241 const ArrayType *ATy = cast<ArrayType>(Ty);
242 unsigned NumElements = ATy->getNumElements();
243 return printType(Out, ATy->getElementType(),
244 NameSoFar + "[" + utostr(NumElements) + "]");
247 case Type::OpaqueTyID: {
248 static int Count = 0;
249 std::string TyName = "struct opaque_" + itostr(Count++);
250 assert(TypeNames.find(Ty) == TypeNames.end());
251 TypeNames[Ty] = TyName;
252 return Out << TyName << " " << NameSoFar;
255 assert(0 && "Unhandled case in getTypeProps!");
262 void CWriter::printConstantArray(ConstantArray *CPA) {
264 // As a special case, print the array as a string if it is an array of
265 // ubytes or an array of sbytes with positive values.
267 const Type *ETy = CPA->getType()->getElementType();
268 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
270 // Make sure the last character is a null char, as automatically added by C
271 if (isString && (CPA->getNumOperands() == 0 ||
272 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
277 // Keep track of whether the last number was a hexadecimal escape
278 bool LastWasHex = false;
280 // Do not include the last character, which we know is null
281 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
282 unsigned char C = cast<ConstantInt>(CPA->getOperand(i))->getRawValue();
284 // Print it out literally if it is a printable character. The only thing
285 // to be careful about is when the last letter output was a hex escape
286 // code, in which case we have to be careful not to print out hex digits
287 // explicitly (the C compiler thinks it is a continuation of the previous
288 // character, sheesh...)
290 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
292 if (C == '"' || C == '\\')
299 case '\n': Out << "\\n"; break;
300 case '\t': Out << "\\t"; break;
301 case '\r': Out << "\\r"; break;
302 case '\v': Out << "\\v"; break;
303 case '\a': Out << "\\a"; break;
304 case '\"': Out << "\\\""; break;
305 case '\'': Out << "\\\'"; break;
308 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
309 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
318 if (CPA->getNumOperands()) {
320 printConstant(cast<Constant>(CPA->getOperand(0)));
321 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
323 printConstant(cast<Constant>(CPA->getOperand(i)));
330 /// FPCSafeToPrint - Returns true if we may assume that CFP may be
331 /// written out textually as a double (rather than as a reference to a
332 /// stack-allocated variable). We decide this by converting CFP to a
333 /// string and back into a double, and then checking whether the
334 /// conversion results in a bit-equal double to the original value of
335 /// CFP. This depends on us and the target C compiler agreeing on the
336 /// conversion process (which is pretty likely since we only deal in
337 /// IEEE FP.) This is adapted from similar code in
338 /// lib/VMCore/AsmWriter.cpp:WriteConstantInt().
339 static bool FPCSafeToPrint (const ConstantFP *CFP) {
340 std::string StrVal = ftostr(CFP->getValue());
341 // Check to make sure that the stringized number is not some string like
342 // "Inf" or NaN, that atof will accept, but the lexer will not. Check that
343 // the string matches the "[-+]?[0-9]" regex.
344 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
345 ((StrVal[0] == '-' || StrVal[0] == '+') &&
346 (StrVal[1] >= '0' && StrVal[1] <= '9')))
347 // Reparse stringized version!
348 return (atof(StrVal.c_str()) == CFP->getValue());
352 // printConstant - The LLVM Constant to C Constant converter.
353 void CWriter::printConstant(Constant *CPV) {
354 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
355 switch (CE->getOpcode()) {
356 case Instruction::Cast:
358 printType(Out, CPV->getType());
360 printConstant(CE->getOperand(0));
364 case Instruction::GetElementPtr:
366 printIndexingExpression(CE->getOperand(0),
367 CPV->op_begin()+1, CPV->op_end());
370 case Instruction::Add:
371 case Instruction::Sub:
372 case Instruction::Mul:
373 case Instruction::Div:
374 case Instruction::Rem:
375 case Instruction::SetEQ:
376 case Instruction::SetNE:
377 case Instruction::SetLT:
378 case Instruction::SetLE:
379 case Instruction::SetGT:
380 case Instruction::SetGE:
382 printConstant(CE->getOperand(0));
383 switch (CE->getOpcode()) {
384 case Instruction::Add: Out << " + "; break;
385 case Instruction::Sub: Out << " - "; break;
386 case Instruction::Mul: Out << " * "; break;
387 case Instruction::Div: Out << " / "; break;
388 case Instruction::Rem: Out << " % "; break;
389 case Instruction::SetEQ: Out << " == "; break;
390 case Instruction::SetNE: Out << " != "; break;
391 case Instruction::SetLT: Out << " < "; break;
392 case Instruction::SetLE: Out << " <= "; break;
393 case Instruction::SetGT: Out << " > "; break;
394 case Instruction::SetGE: Out << " >= "; break;
395 default: assert(0 && "Illegal opcode here!");
397 printConstant(CE->getOperand(1));
402 std::cerr << "CWriter Error: Unhandled constant expression: "
408 switch (CPV->getType()->getPrimitiveID()) {
410 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
411 case Type::SByteTyID:
412 case Type::ShortTyID:
413 Out << cast<ConstantSInt>(CPV)->getValue(); break;
415 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
416 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
418 Out << cast<ConstantSInt>(CPV)->getValue();
422 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
424 case Type::UByteTyID:
425 case Type::UShortTyID:
426 Out << cast<ConstantUInt>(CPV)->getValue(); break;
428 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
429 case Type::ULongTyID:
430 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
432 case Type::FloatTyID:
433 case Type::DoubleTyID: {
434 ConstantFP *FPC = cast<ConstantFP>(CPV);
435 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
436 if (I != FPConstantMap.end()) {
437 // Because of FP precision problems we must load from a stack allocated
438 // value that holds the value in hex.
439 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
440 << "*)&FloatConstant" << I->second << ")";
442 if (FPCSafeToPrint (FPC)) {
443 Out << ftostr (FPC->getValue ());
445 Out << FPC->getValue(); // Who knows? Give it our best shot...
451 case Type::ArrayTyID:
452 printConstantArray(cast<ConstantArray>(CPV));
455 case Type::StructTyID: {
457 if (CPV->getNumOperands()) {
459 printConstant(cast<Constant>(CPV->getOperand(0)));
460 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
462 printConstant(cast<Constant>(CPV->getOperand(i)));
469 case Type::PointerTyID:
470 if (isa<ConstantPointerNull>(CPV)) {
472 printType(Out, CPV->getType());
473 Out << ")/*NULL*/0)";
475 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
476 writeOperand(CPR->getValue());
481 std::cerr << "Unknown constant type: " << CPV << "\n";
486 void CWriter::writeOperandInternal(Value *Operand) {
487 if (Instruction *I = dyn_cast<Instruction>(Operand))
488 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
489 // Should we inline this instruction to build a tree?
496 if (Constant *CPV = dyn_cast<Constant>(Operand)) {
499 Out << Mang->getValueName(Operand);
503 void CWriter::writeOperand(Value *Operand) {
504 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
505 Out << "(&"; // Global variables are references as their addresses by llvm
507 writeOperandInternal(Operand);
509 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
513 // nameAllUsedStructureTypes - If there are structure types in the module that
514 // are used but do not have names assigned to them in the symbol table yet then
515 // we assign them names now.
517 bool CWriter::nameAllUsedStructureTypes(Module &M) {
518 // Get a set of types that are used by the program...
519 std::set<const Type *> UT = getAnalysis<FindUsedTypes>().getTypes();
521 // Loop over the module symbol table, removing types from UT that are already
524 SymbolTable &MST = M.getSymbolTable();
525 if (MST.find(Type::TypeTy) != MST.end())
526 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
527 E = MST.type_end(Type::TypeTy); I != E; ++I)
528 UT.erase(cast<Type>(I->second));
530 // UT now contains types that are not named. Loop over it, naming structure
533 bool Changed = false;
534 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
536 if (const StructType *ST = dyn_cast<StructType>(*I)) {
537 ((Value*)ST)->setName("unnamed", &MST);
543 // generateCompilerSpecificCode - This is where we add conditional compilation
544 // directives to cater to specific compilers as need be.
546 static void generateCompilerSpecificCode(std::ostream& Out) {
547 // Alloca is hard to get, and we don't want to include stdlib.h here...
548 Out << "/* get a declaration for alloca */\n"
550 << "extern void *__builtin_alloca(unsigned long);\n"
551 << "#define alloca(x) __builtin_alloca(x)\n"
553 << "#ifndef __FreeBSD__\n"
554 << "#include <alloca.h>\n"
558 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
559 // If we aren't being compiled with GCC, just drop these attributes.
560 Out << "#ifndef __GNUC__ /* Can only support \"linkonce\" vars with GCC */\n"
561 << "#define __attribute__(X)\n"
565 void CWriter::printModule(Module *M) {
566 // Calculate which global values have names that will collide when we throw
567 // away type information.
568 { // Scope to delete the FoundNames set when we are done with it...
569 std::set<std::string> FoundNames;
570 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
571 if (I->hasName()) // If the global has a name...
572 if (FoundNames.count(I->getName())) // And the name is already used
573 MangledGlobals.insert(I); // Mangle the name
575 FoundNames.insert(I->getName()); // Otherwise, keep track of name
577 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
578 if (I->hasName()) // If the global has a name...
579 if (FoundNames.count(I->getName())) // And the name is already used
580 MangledGlobals.insert(I); // Mangle the name
582 FoundNames.insert(I->getName()); // Otherwise, keep track of name
585 // get declaration for alloca
586 Out << "/* Provide Declarations */\n";
587 Out << "#include <stdarg.h>\n";
588 Out << "#include <setjmp.h>\n";
589 generateCompilerSpecificCode(Out);
591 // Provide a definition for `bool' if not compiling with a C++ compiler.
593 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
595 << "\n\n/* Support for floating point constants */\n"
596 << "typedef unsigned long long ConstantDoubleTy;\n"
597 << "typedef unsigned int ConstantFloatTy;\n"
599 << "\n\n/* Support for the invoke instruction */\n"
600 << "extern struct __llvm_jmpbuf_list_t {\n"
601 << " jmp_buf buf; struct __llvm_jmpbuf_list_t *next;\n"
602 << "} *__llvm_jmpbuf_list;\n"
604 << "\n\n/* Global Declarations */\n";
606 // First output all the declarations for the program, because C requires
607 // Functions & globals to be declared before they are used.
610 // Loop over the symbol table, emitting all named constants...
611 printSymbolTable(M->getSymbolTable());
613 // Global variable declarations...
615 Out << "\n/* External Global Variable Declarations */\n";
616 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
617 if (I->hasExternalLinkage()) {
619 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
625 // Function declarations
627 Out << "\n/* Function Declarations */\n";
629 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
630 // If the function is external and the name collides don't print it.
631 // Sometimes the bytecode likes to have multiple "declarations" for
632 // external functions
633 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
634 !I->getIntrinsicID()) {
635 printFunctionSignature(I, true);
641 // Print Malloc prototype if needed
643 Out << "\n/* Malloc to make sun happy */\n";
644 Out << "extern void * malloc();\n\n";
647 // Output the global variable declarations
649 Out << "\n\n/* Global Variable Declarations */\n";
650 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
651 if (!I->isExternal()) {
653 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
659 // Output the global variable definitions and contents...
661 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
662 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
663 if (!I->isExternal()) {
664 if (I->hasInternalLinkage())
666 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
667 if (I->hasLinkOnceLinkage())
668 Out << " __attribute__((common))";
669 if (!I->getInitializer()->isNullValue()) {
671 writeOperand(I->getInitializer());
677 // Output all of the functions...
678 emittedInvoke = false;
680 Out << "\n\n/* Function Bodies */\n";
681 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
685 // If the program included an invoke instruction, we need to output the
686 // support code for it here!
688 Out << "\n/* More support for the invoke instruction */\n"
689 << "struct __llvm_jmpbuf_list_t *__llvm_jmpbuf_list "
690 << "__attribute__((common)) = 0;\n";
695 /// printSymbolTable - Run through symbol table looking for type names. If a
696 /// type name is found, emit it's declaration...
698 void CWriter::printSymbolTable(const SymbolTable &ST) {
699 // If there are no type names, exit early.
700 if (ST.find(Type::TypeTy) == ST.end())
703 // We are only interested in the type plane of the symbol table...
704 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
705 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
707 // Print out forward declarations for structure types before anything else!
708 Out << "/* Structure forward decls */\n";
709 for (; I != End; ++I)
710 if (const Type *STy = dyn_cast<StructType>(I->second)) {
711 std::string Name = "struct l_" + Mangler::makeNameProper(I->first);
712 Out << Name << ";\n";
713 TypeNames.insert(std::make_pair(STy, Name));
718 // Now we can print out typedefs...
719 Out << "/* Typedefs */\n";
720 for (I = ST.type_begin(Type::TypeTy); I != End; ++I) {
721 const Type *Ty = cast<Type>(I->second);
722 std::string Name = "l_" + Mangler::makeNameProper(I->first);
724 printType(Out, Ty, Name);
730 // Keep track of which structures have been printed so far...
731 std::set<const StructType *> StructPrinted;
733 // Loop over all structures then push them into the stack so they are
734 // printed in the correct order.
736 Out << "/* Structure contents */\n";
737 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
738 if (const StructType *STy = dyn_cast<StructType>(I->second))
739 printContainedStructs(STy, StructPrinted);
742 // Push the struct onto the stack and recursively push all structs
743 // this one depends on.
744 void CWriter::printContainedStructs(const Type *Ty,
745 std::set<const StructType*> &StructPrinted){
746 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
747 //Check to see if we have already printed this struct
748 if (StructPrinted.count(STy) == 0) {
749 // Print all contained types first...
750 for (StructType::ElementTypes::const_iterator
751 I = STy->getElementTypes().begin(),
752 E = STy->getElementTypes().end(); I != E; ++I) {
753 const Type *Ty1 = I->get();
754 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
755 printContainedStructs(*I, StructPrinted);
758 //Print structure type out..
759 StructPrinted.insert(STy);
760 std::string Name = TypeNames[STy];
761 printType(Out, STy, Name, true);
765 // If it is an array, check contained types and continue
766 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
767 const Type *Ty1 = ATy->getElementType();
768 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
769 printContainedStructs(Ty1, StructPrinted);
774 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
775 // If the program provides its own malloc prototype we don't need
776 // to include the general one.
777 if (Mang->getValueName(F) == "malloc")
780 if (F->hasInternalLinkage()) Out << "static ";
781 if (F->hasLinkOnceLinkage()) Out << "inline ";
783 // Loop over the arguments, printing them...
784 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
786 std::stringstream FunctionInnards;
788 // Print out the name...
789 FunctionInnards << Mang->getValueName(F) << "(";
791 if (!F->isExternal()) {
794 if (F->abegin()->hasName() || !Prototype)
795 ArgName = Mang->getValueName(F->abegin());
796 printType(FunctionInnards, F->afront().getType(), ArgName);
797 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
799 FunctionInnards << ", ";
800 if (I->hasName() || !Prototype)
801 ArgName = Mang->getValueName(I);
804 printType(FunctionInnards, I->getType(), ArgName);
808 // Loop over the arguments, printing them...
809 for (FunctionType::ParamTypes::const_iterator I =
810 FT->getParamTypes().begin(),
811 E = FT->getParamTypes().end(); I != E; ++I) {
812 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
813 printType(FunctionInnards, *I);
817 // Finish printing arguments... if this is a vararg function, print the ...,
818 // unless there are no known types, in which case, we just emit ().
820 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
821 if (FT->getParamTypes().size()) FunctionInnards << ", ";
822 FunctionInnards << "..."; // Output varargs portion of signature!
824 FunctionInnards << ")";
825 // Print out the return type and the entire signature for that matter
826 printType(Out, F->getReturnType(), FunctionInnards.str());
829 void CWriter::printFunction(Function *F) {
830 if (F->isExternal()) return;
832 printFunctionSignature(F, false);
835 // print local variable information for the function
836 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
837 if (const AllocaInst *AI = isDirectAlloca(*I)) {
839 printType(Out, AI->getAllocatedType(), Mang->getValueName(AI));
840 Out << "; /* Address exposed local */\n";
841 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
843 printType(Out, (*I)->getType(), Mang->getValueName(*I));
846 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
848 printType(Out, (*I)->getType(), Mang->getValueName(*I)+"__PHI_TEMPORARY");
855 // Scan the function for floating point constants. If any FP constant is used
856 // in the function, we want to redirect it here so that we do not depend on
857 // the precision of the printed form, unless the printed form preserves
860 unsigned FPCounter = 0;
861 for (constant_iterator I = constant_begin(F), E = constant_end(F); I != E;++I)
862 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
863 if ((!FPCSafeToPrint(FPC)) // Do not put in FPConstantMap if safe.
864 && (FPConstantMap.find(FPC) == FPConstantMap.end())) {
865 double Val = FPC->getValue();
867 FPConstantMap[FPC] = FPCounter; // Number the FP constants
869 if (FPC->getType() == Type::DoubleTy)
870 Out << " const ConstantDoubleTy FloatConstant" << FPCounter++
871 << " = 0x" << std::hex << *(unsigned long long*)&Val << std::dec
872 << "ULL; /* " << Val << " */\n";
873 else if (FPC->getType() == Type::FloatTy) {
875 Out << " const ConstantFloatTy FloatConstant" << FPCounter++
876 << " = 0x" << std::hex << *(unsigned*)&fVal << std::dec
877 << "U; /* " << Val << " */\n";
879 assert(0 && "Unknown float type!");
884 // print the basic blocks
885 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
886 BasicBlock *Prev = BB->getPrev();
888 // Don't print the label for the basic block if there are no uses, or if the
889 // only terminator use is the precessor basic block's terminator. We have
890 // to scan the use list because PHI nodes use basic blocks too but do not
891 // require a label to be generated.
893 bool NeedsLabel = false;
894 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
896 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
897 if (TI != Prev->getTerminator() ||
898 isa<SwitchInst>(Prev->getTerminator()) ||
899 isa<InvokeInst>(Prev->getTerminator())) {
904 if (NeedsLabel) Out << Mang->getValueName(BB) << ":\n";
906 // Output all of the instructions in the basic block...
907 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
908 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
909 if (II->getType() != Type::VoidTy)
918 // Don't emit prefix or suffix for the terminator...
919 visit(*BB->getTerminator());
923 FPConstantMap.clear();
926 // Specific Instruction type classes... note that all of the casts are
927 // necessary because we use the instruction classes as opaque types...
929 void CWriter::visitReturnInst(ReturnInst &I) {
930 // Don't output a void return if this is the last basic block in the function
931 if (I.getNumOperands() == 0 &&
932 &*--I.getParent()->getParent()->end() == I.getParent() &&
933 !I.getParent()->size() == 1) {
938 if (I.getNumOperands()) {
940 writeOperand(I.getOperand(0));
945 void CWriter::visitSwitchInst(SwitchInst &SI) {
947 writeOperand(SI.getOperand(0));
948 Out << ") {\n default:\n";
949 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
951 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
953 writeOperand(SI.getOperand(i));
955 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
956 printBranchToBlock(SI.getParent(), Succ, 2);
957 if (Succ == SI.getParent()->getNext())
963 void CWriter::visitInvokeInst(InvokeInst &II) {
965 << " struct __llvm_jmpbuf_list_t Entry;\n"
966 << " Entry.next = __llvm_jmpbuf_list;\n"
967 << " if (setjmp(Entry.buf)) {\n"
968 << " __llvm_jmpbuf_list = Entry.next;\n";
969 printBranchToBlock(II.getParent(), II.getExceptionalDest(), 4);
971 << " __llvm_jmpbuf_list = &Entry;\n"
974 if (II.getType() != Type::VoidTy) outputLValue(&II);
977 << " __llvm_jmpbuf_list = Entry.next;\n"
979 printBranchToBlock(II.getParent(), II.getNormalDest(), 0);
980 emittedInvoke = true;
984 void CWriter::visitUnwindInst(UnwindInst &I) {
985 // The unwind instructions causes a control flow transfer out of the current
986 // function, unwinding the stack until a caller who used the invoke
987 // instruction is found. In this context, we code generated the invoke
988 // instruction to add an entry to the top of the jmpbuf_list. Thus, here we
989 // just have to longjmp to the specified handler.
990 Out << " if (__llvm_jmpbuf_list == 0) { /* llvm.unwind */\n"
991 << " printf(\"throw found with no handler!\\n\"); abort();\n"
993 << " longjmp(__llvm_jmpbuf_list->buf, 1);\n";
994 emittedInvoke = true;
997 static bool isGotoCodeNeccessary(BasicBlock *From, BasicBlock *To) {
998 // If PHI nodes need copies, we need the copy code...
999 if (isa<PHINode>(To->front()) ||
1000 From->getNext() != To) // Not directly successor, need goto
1003 // Otherwise we don't need the code.
1007 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
1009 for (BasicBlock::iterator I = Succ->begin();
1010 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1011 // now we have to do the printing
1012 Out << std::string(Indent, ' ');
1013 Out << " " << Mang->getValueName(I) << "__PHI_TEMPORARY = ";
1014 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1015 Out << "; /* for PHI node */\n";
1018 if (CurBB->getNext() != Succ || isa<InvokeInst>(CurBB->getTerminator())) {
1019 Out << std::string(Indent, ' ') << " goto ";
1025 // Brach instruction printing - Avoid printing out a brach to a basic block that
1026 // immediately succeeds the current one.
1028 void CWriter::visitBranchInst(BranchInst &I) {
1029 if (I.isConditional()) {
1030 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(0))) {
1032 writeOperand(I.getCondition());
1035 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1037 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(1))) {
1038 Out << " } else {\n";
1039 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1042 // First goto not necessary, assume second one is...
1044 writeOperand(I.getCondition());
1047 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1052 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1057 // PHI nodes get copied into temporary values at the end of predecessor basic
1058 // blocks. We now need to copy these temporary values into the REAL value for
1060 void CWriter::visitPHINode(PHINode &I) {
1062 Out << "__PHI_TEMPORARY";
1066 void CWriter::visitBinaryOperator(Instruction &I) {
1067 // binary instructions, shift instructions, setCond instructions.
1068 assert(!isa<PointerType>(I.getType()));
1070 // We must cast the results of binary operations which might be promoted.
1071 bool needsCast = false;
1072 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1073 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1074 || (I.getType() == Type::FloatTy)) {
1077 printType(Out, I.getType(), "", false, false);
1081 writeOperand(I.getOperand(0));
1083 switch (I.getOpcode()) {
1084 case Instruction::Add: Out << " + "; break;
1085 case Instruction::Sub: Out << " - "; break;
1086 case Instruction::Mul: Out << "*"; break;
1087 case Instruction::Div: Out << "/"; break;
1088 case Instruction::Rem: Out << "%"; break;
1089 case Instruction::And: Out << " & "; break;
1090 case Instruction::Or: Out << " | "; break;
1091 case Instruction::Xor: Out << " ^ "; break;
1092 case Instruction::SetEQ: Out << " == "; break;
1093 case Instruction::SetNE: Out << " != "; break;
1094 case Instruction::SetLE: Out << " <= "; break;
1095 case Instruction::SetGE: Out << " >= "; break;
1096 case Instruction::SetLT: Out << " < "; break;
1097 case Instruction::SetGT: Out << " > "; break;
1098 case Instruction::Shl : Out << " << "; break;
1099 case Instruction::Shr : Out << " >> "; break;
1100 default: std::cerr << "Invalid operator type!" << I; abort();
1103 writeOperand(I.getOperand(1));
1110 void CWriter::visitCastInst(CastInst &I) {
1111 if (I.getType() == Type::BoolTy) {
1113 writeOperand(I.getOperand(0));
1118 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1120 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1121 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1122 // Avoid "cast to pointer from integer of different size" warnings
1126 writeOperand(I.getOperand(0));
1129 void CWriter::visitCallInst(CallInst &I) {
1130 // Handle intrinsic function calls first...
1131 if (Function *F = I.getCalledFunction())
1132 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID()) {
1134 default: assert(0 && "Unknown LLVM intrinsic!");
1135 case LLVMIntrinsic::va_start:
1136 Out << "va_start(*(va_list*)";
1137 writeOperand(I.getOperand(1));
1139 // Output the last argument to the enclosing function...
1140 writeOperand(&I.getParent()->getParent()->aback());
1143 case LLVMIntrinsic::va_end:
1144 Out << "va_end(*(va_list*)";
1145 writeOperand(I.getOperand(1));
1148 case LLVMIntrinsic::va_copy:
1149 Out << "va_copy(*(va_list*)";
1150 writeOperand(I.getOperand(1));
1151 Out << ", (va_list)";
1152 writeOperand(I.getOperand(2));
1156 case LLVMIntrinsic::setjmp:
1157 case LLVMIntrinsic::sigsetjmp:
1158 // This instrinsic should never exist in the program, but until we get
1159 // setjmp/longjmp transformations going on, we should codegen it to
1160 // something reasonable. This will allow code that never calls longjmp
1164 case LLVMIntrinsic::longjmp:
1165 case LLVMIntrinsic::siglongjmp:
1166 // Longjmp is not implemented, and never will be. It would cause an
1175 void CWriter::visitCallSite(CallSite CS) {
1176 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1177 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1178 const Type *RetTy = FTy->getReturnType();
1180 writeOperand(CS.getCalledValue());
1183 if (CS.arg_begin() != CS.arg_end()) {
1184 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1187 for (++AI; AI != AE; ++AI) {
1195 void CWriter::visitMallocInst(MallocInst &I) {
1197 printType(Out, I.getType());
1198 Out << ")malloc(sizeof(";
1199 printType(Out, I.getType()->getElementType());
1202 if (I.isArrayAllocation()) {
1204 writeOperand(I.getOperand(0));
1209 void CWriter::visitAllocaInst(AllocaInst &I) {
1211 printType(Out, I.getType());
1212 Out << ") alloca(sizeof(";
1213 printType(Out, I.getType()->getElementType());
1215 if (I.isArrayAllocation()) {
1217 writeOperand(I.getOperand(0));
1222 void CWriter::visitFreeInst(FreeInst &I) {
1223 Out << "free((char*)";
1224 writeOperand(I.getOperand(0));
1228 void CWriter::printIndexingExpression(Value *Ptr, User::op_iterator I,
1229 User::op_iterator E) {
1230 bool HasImplicitAddress = false;
1231 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1232 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1233 HasImplicitAddress = true;
1234 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1235 HasImplicitAddress = true;
1236 Ptr = CPR->getValue(); // Get to the global...
1237 } else if (isDirectAlloca(Ptr)) {
1238 HasImplicitAddress = true;
1242 if (!HasImplicitAddress)
1243 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1245 writeOperandInternal(Ptr);
1249 const Constant *CI = dyn_cast<Constant>(I);
1250 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1253 writeOperandInternal(Ptr);
1255 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1257 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1260 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1261 "Can only have implicit address with direct accessing");
1263 if (HasImplicitAddress) {
1265 } else if (CI && CI->isNullValue() && I+1 != E) {
1266 // Print out the -> operator if possible...
1267 if ((*(I+1))->getType() == Type::UByteTy) {
1268 Out << (HasImplicitAddress ? "." : "->");
1269 Out << "field" << cast<ConstantUInt>(*(I+1))->getValue();
1275 if ((*I)->getType() == Type::LongTy) {
1280 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
1284 void CWriter::visitLoadInst(LoadInst &I) {
1286 writeOperand(I.getOperand(0));
1289 void CWriter::visitStoreInst(StoreInst &I) {
1291 writeOperand(I.getPointerOperand());
1293 writeOperand(I.getOperand(0));
1296 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1298 printIndexingExpression(I.getPointerOperand(), I.idx_begin(), I.idx_end());
1301 void CWriter::visitVarArgInst(VarArgInst &I) {
1302 Out << "va_arg((va_list)*";
1303 writeOperand(I.getOperand(0));
1305 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1310 //===----------------------------------------------------------------------===//
1311 // External Interface declaration
1312 //===----------------------------------------------------------------------===//
1314 Pass *createWriteToCPass(std::ostream &o) { return new CWriter(o); }