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 "llvm/Support/CallSite.h"
21 #include "Support/StringExtras.h"
22 #include "Support/STLExtras.h"
28 class CWriter : public Pass, public InstVisitor<CWriter> {
30 SlotCalculator *Table;
31 const Module *TheModule;
32 std::map<const Type *, std::string> TypeNames;
33 std::set<const Value*> MangledGlobals;
34 bool needsMalloc, emittedInvoke;
36 std::map<const ConstantFP *, unsigned> FPConstantMap;
38 CWriter(std::ostream &o) : Out(o) {}
40 void getAnalysisUsage(AnalysisUsage &AU) const {
42 AU.addRequired<FindUsedTypes>();
45 virtual bool run(Module &M) {
47 Table = new SlotCalculator(&M, false);
50 // Ensure that all structure types have names...
51 bool Changed = nameAllUsedStructureTypes(M);
59 MangledGlobals.clear();
63 std::ostream &printType(std::ostream &Out, const Type *Ty,
64 const std::string &VariableName = "",
65 bool IgnoreName = false, bool namedContext = true);
67 void writeOperand(Value *Operand);
68 void writeOperandInternal(Value *Operand);
70 std::string getValueName(const Value *V);
73 bool nameAllUsedStructureTypes(Module &M);
74 void printModule(Module *M);
75 void printSymbolTable(const SymbolTable &ST);
76 void printContainedStructs(const Type *Ty, std::set<const StructType *> &);
77 void printFunctionSignature(const Function *F, bool Prototype);
79 void printFunction(Function *);
81 void printConstant(Constant *CPV);
82 void printConstantArray(ConstantArray *CPA);
84 // isInlinableInst - Attempt to inline instructions into their uses to build
85 // trees as much as possible. To do this, we have to consistently decide
86 // what is acceptable to inline, so that variable declarations don't get
87 // printed and an extra copy of the expr is not emitted.
89 static bool isInlinableInst(const Instruction &I) {
90 // Must be an expression, must be used exactly once. If it is dead, we
91 // emit it inline where it would go.
92 if (I.getType() == Type::VoidTy || I.use_size() != 1 ||
93 isa<TerminatorInst>(I) || isa<CallInst>(I) || isa<PHINode>(I) ||
94 isa<LoadInst>(I)) // Don't inline a load across a store!
97 // Only inline instruction it it's use is in the same BB as the inst.
98 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
101 // isDirectAlloca - Define fixed sized allocas in the entry block as direct
102 // variables which are accessed with the & operator. This causes GCC to
103 // generate significantly better code than to emit alloca calls directly.
105 static const AllocaInst *isDirectAlloca(const Value *V) {
106 const AllocaInst *AI = dyn_cast<AllocaInst>(V);
107 if (!AI) return false;
108 if (AI->isArrayAllocation())
109 return 0; // FIXME: we can also inline fixed size array allocas!
110 if (AI->getParent() != &AI->getParent()->getParent()->getEntryNode())
115 // Instruction visitation functions
116 friend class InstVisitor<CWriter>;
118 void visitReturnInst(ReturnInst &I);
119 void visitBranchInst(BranchInst &I);
120 void visitSwitchInst(SwitchInst &I);
121 void visitInvokeInst(InvokeInst &I);
123 void visitPHINode(PHINode &I);
124 void visitBinaryOperator(Instruction &I);
126 void visitCastInst (CastInst &I);
127 void visitCallInst (CallInst &I);
128 void visitCallSite (CallSite CS);
129 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
131 void visitMallocInst(MallocInst &I);
132 void visitAllocaInst(AllocaInst &I);
133 void visitFreeInst (FreeInst &I);
134 void visitLoadInst (LoadInst &I);
135 void visitStoreInst (StoreInst &I);
136 void visitGetElementPtrInst(GetElementPtrInst &I);
137 void visitVarArgInst(VarArgInst &I);
139 void visitInstruction(Instruction &I) {
140 std::cerr << "C Writer does not know about " << I;
144 void outputLValue(Instruction *I) {
145 Out << " " << getValueName(I) << " = ";
147 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
149 void printIndexingExpression(Value *Ptr, User::op_iterator I,
150 User::op_iterator E);
154 // We dont want identifier names with ., space, - in them.
155 // So we replace them with _
156 static std::string makeNameProper(std::string x) {
158 for (std::string::iterator sI = x.begin(), sEnd = x.end(); sI != sEnd; sI++)
160 case '.': tmp += "d_"; break;
161 case ' ': tmp += "s_"; break;
162 case '-': tmp += "D_"; break;
169 std::string CWriter::getValueName(const Value *V) {
170 if (V->hasName()) { // Print out the label if it exists...
172 // Name mangling occurs as follows:
173 // - If V is not a global, mangling always occurs.
174 // - Otherwise, mangling occurs when any of the following are true:
175 // 1) V has internal linkage
176 // 2) V's name would collide if it is not mangled.
179 if(const GlobalValue* gv = dyn_cast<GlobalValue>(V)) {
180 if(!gv->hasInternalLinkage() && !MangledGlobals.count(gv)) {
181 // No internal linkage, name will not collide -> no mangling.
182 return makeNameProper(gv->getName());
186 // Non-global, or global with internal linkage / colliding name -> mangle.
187 return "l" + utostr(V->getType()->getUniqueID()) + "_" +
188 makeNameProper(V->getName());
191 int Slot = Table->getValSlot(V);
192 assert(Slot >= 0 && "Invalid value!");
193 return "ltmp_" + itostr(Slot) + "_" + utostr(V->getType()->getUniqueID());
196 // A pointer type should not use parens around *'s alone, e.g., (**)
197 inline bool ptrTypeNameNeedsParens(const std::string &NameSoFar) {
198 return (NameSoFar.find_last_not_of('*') != std::string::npos);
201 // Pass the Type* and the variable name and this prints out the variable
204 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
205 const std::string &NameSoFar,
206 bool IgnoreName, bool namedContext) {
207 if (Ty->isPrimitiveType())
208 switch (Ty->getPrimitiveID()) {
209 case Type::VoidTyID: return Out << "void " << NameSoFar;
210 case Type::BoolTyID: return Out << "bool " << NameSoFar;
211 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
212 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
213 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
214 case Type::ShortTyID: return Out << "short " << NameSoFar;
215 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
216 case Type::IntTyID: return Out << "int " << NameSoFar;
217 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
218 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
219 case Type::FloatTyID: return Out << "float " << NameSoFar;
220 case Type::DoubleTyID: return Out << "double " << NameSoFar;
222 std::cerr << "Unknown primitive type: " << Ty << "\n";
226 // Check to see if the type is named.
227 if (!IgnoreName || isa<OpaqueType>(Ty)) {
228 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
229 if (I != TypeNames.end()) {
230 return Out << I->second << " " << NameSoFar;
234 switch (Ty->getPrimitiveID()) {
235 case Type::FunctionTyID: {
236 const FunctionType *MTy = cast<FunctionType>(Ty);
237 std::stringstream FunctionInnards;
238 FunctionInnards << " (" << NameSoFar << ") (";
239 for (FunctionType::ParamTypes::const_iterator
240 I = MTy->getParamTypes().begin(),
241 E = MTy->getParamTypes().end(); I != E; ++I) {
242 if (I != MTy->getParamTypes().begin())
243 FunctionInnards << ", ";
244 printType(FunctionInnards, *I, "");
246 if (MTy->isVarArg()) {
247 if (!MTy->getParamTypes().empty())
248 FunctionInnards << ", ...";
249 } else if (MTy->getParamTypes().empty()) {
250 FunctionInnards << "void";
252 FunctionInnards << ")";
253 std::string tstr = FunctionInnards.str();
254 printType(Out, MTy->getReturnType(), tstr);
257 case Type::StructTyID: {
258 const StructType *STy = cast<StructType>(Ty);
259 Out << NameSoFar + " {\n";
261 for (StructType::ElementTypes::const_iterator
262 I = STy->getElementTypes().begin(),
263 E = STy->getElementTypes().end(); I != E; ++I) {
265 printType(Out, *I, "field" + utostr(Idx++));
271 case Type::PointerTyID: {
272 const PointerType *PTy = cast<PointerType>(Ty);
273 std::string ptrName = "*" + NameSoFar;
275 // Do not need parens around "* NameSoFar" if NameSoFar consists only
276 // of zero or more '*' chars *and* this is not an unnamed pointer type
277 // such as the result type in a cast statement. Otherwise, enclose in ( ).
278 if (ptrTypeNameNeedsParens(NameSoFar) || !namedContext ||
279 PTy->getElementType()->getPrimitiveID() == Type::ArrayTyID)
280 ptrName = "(" + ptrName + ")"; //
282 return printType(Out, PTy->getElementType(), ptrName);
285 case Type::ArrayTyID: {
286 const ArrayType *ATy = cast<ArrayType>(Ty);
287 unsigned NumElements = ATy->getNumElements();
288 return printType(Out, ATy->getElementType(),
289 NameSoFar + "[" + utostr(NumElements) + "]");
292 case Type::OpaqueTyID: {
293 static int Count = 0;
294 std::string TyName = "struct opaque_" + itostr(Count++);
295 assert(TypeNames.find(Ty) == TypeNames.end());
296 TypeNames[Ty] = TyName;
297 return Out << TyName << " " << NameSoFar;
300 assert(0 && "Unhandled case in getTypeProps!");
307 void CWriter::printConstantArray(ConstantArray *CPA) {
309 // As a special case, print the array as a string if it is an array of
310 // ubytes or an array of sbytes with positive values.
312 const Type *ETy = CPA->getType()->getElementType();
313 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
315 // Make sure the last character is a null char, as automatically added by C
316 if (isString && (CPA->getNumOperands() == 0 ||
317 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
322 // Keep track of whether the last number was a hexadecimal escape
323 bool LastWasHex = false;
325 // Do not include the last character, which we know is null
326 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
327 unsigned char C = (ETy == Type::SByteTy) ?
328 (unsigned char)cast<ConstantSInt>(CPA->getOperand(i))->getValue() :
329 (unsigned char)cast<ConstantUInt>(CPA->getOperand(i))->getValue();
331 // Print it out literally if it is a printable character. The only thing
332 // to be careful about is when the last letter output was a hex escape
333 // code, in which case we have to be careful not to print out hex digits
334 // explicitly (the C compiler thinks it is a continuation of the previous
335 // character, sheesh...)
337 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
339 if (C == '"' || C == '\\')
346 case '\n': Out << "\\n"; break;
347 case '\t': Out << "\\t"; break;
348 case '\r': Out << "\\r"; break;
349 case '\v': Out << "\\v"; break;
350 case '\a': Out << "\\a"; break;
351 case '\"': Out << "\\\""; break;
352 case '\'': Out << "\\\'"; break;
355 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
356 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
365 if (CPA->getNumOperands()) {
367 printConstant(cast<Constant>(CPA->getOperand(0)));
368 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
370 printConstant(cast<Constant>(CPA->getOperand(i)));
377 /// FPCSafeToPrint - Returns true if we may assume that CFP may be
378 /// written out textually as a double (rather than as a reference to a
379 /// stack-allocated variable). We decide this by converting CFP to a
380 /// string and back into a double, and then checking whether the
381 /// conversion results in a bit-equal double to the original value of
382 /// CFP. This depends on us and the target C compiler agreeing on the
383 /// conversion process (which is pretty likely since we only deal in
384 /// IEEE FP.) This is adapted from similar code in
385 /// lib/VMCore/AsmWriter.cpp:WriteConstantInt().
386 static bool FPCSafeToPrint (const ConstantFP *CFP) {
387 std::string StrVal = ftostr(CFP->getValue());
388 // Check to make sure that the stringized number is not some string like
389 // "Inf" or NaN, that atof will accept, but the lexer will not. Check that
390 // the string matches the "[-+]?[0-9]" regex.
391 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
392 ((StrVal[0] == '-' || StrVal[0] == '+') &&
393 (StrVal[1] >= '0' && StrVal[1] <= '9')))
394 // Reparse stringized version!
395 return (atof(StrVal.c_str()) == CFP->getValue());
399 // printConstant - The LLVM Constant to C Constant converter.
400 void CWriter::printConstant(Constant *CPV) {
401 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
402 switch (CE->getOpcode()) {
403 case Instruction::Cast:
405 printType(Out, CPV->getType());
407 printConstant(CE->getOperand(0));
411 case Instruction::GetElementPtr:
413 printIndexingExpression(CE->getOperand(0),
414 CPV->op_begin()+1, CPV->op_end());
417 case Instruction::Add:
419 printConstant(CE->getOperand(0));
421 printConstant(CE->getOperand(1));
424 case Instruction::Sub:
426 printConstant(CE->getOperand(0));
428 printConstant(CE->getOperand(1));
433 std::cerr << "CWriter Error: Unhandled constant expression: "
439 switch (CPV->getType()->getPrimitiveID()) {
441 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
442 case Type::SByteTyID:
443 case Type::ShortTyID:
444 Out << cast<ConstantSInt>(CPV)->getValue(); break;
446 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
447 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
449 Out << cast<ConstantSInt>(CPV)->getValue();
453 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
455 case Type::UByteTyID:
456 case Type::UShortTyID:
457 Out << cast<ConstantUInt>(CPV)->getValue(); break;
459 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
460 case Type::ULongTyID:
461 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
463 case Type::FloatTyID:
464 case Type::DoubleTyID: {
465 ConstantFP *FPC = cast<ConstantFP>(CPV);
466 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
467 if (I != FPConstantMap.end()) {
468 // Because of FP precision problems we must load from a stack allocated
469 // value that holds the value in hex.
470 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
471 << "*)&FloatConstant" << I->second << ")";
473 if (FPCSafeToPrint (FPC)) {
474 Out << ftostr (FPC->getValue ());
476 Out << FPC->getValue(); // Who knows? Give it our best shot...
482 case Type::ArrayTyID:
483 printConstantArray(cast<ConstantArray>(CPV));
486 case Type::StructTyID: {
488 if (CPV->getNumOperands()) {
490 printConstant(cast<Constant>(CPV->getOperand(0)));
491 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
493 printConstant(cast<Constant>(CPV->getOperand(i)));
500 case Type::PointerTyID:
501 if (isa<ConstantPointerNull>(CPV)) {
503 printType(Out, CPV->getType());
504 Out << ")/*NULL*/0)";
506 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
507 writeOperand(CPR->getValue());
512 std::cerr << "Unknown constant type: " << CPV << "\n";
517 void CWriter::writeOperandInternal(Value *Operand) {
518 if (Instruction *I = dyn_cast<Instruction>(Operand))
519 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
520 // Should we inline this instruction to build a tree?
527 if (Operand->hasName()) {
528 Out << getValueName(Operand);
529 } else if (Constant *CPV = dyn_cast<Constant>(Operand)) {
532 int Slot = Table->getValSlot(Operand);
533 assert(Slot >= 0 && "Malformed LLVM!");
534 Out << "ltmp_" << Slot << "_" << Operand->getType()->getUniqueID();
538 void CWriter::writeOperand(Value *Operand) {
539 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
540 Out << "(&"; // Global variables are references as their addresses by llvm
542 writeOperandInternal(Operand);
544 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
548 // nameAllUsedStructureTypes - If there are structure types in the module that
549 // are used but do not have names assigned to them in the symbol table yet then
550 // we assign them names now.
552 bool CWriter::nameAllUsedStructureTypes(Module &M) {
553 // Get a set of types that are used by the program...
554 std::set<const Type *> UT = getAnalysis<FindUsedTypes>().getTypes();
556 // Loop over the module symbol table, removing types from UT that are already
559 SymbolTable &MST = M.getSymbolTable();
560 if (MST.find(Type::TypeTy) != MST.end())
561 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
562 E = MST.type_end(Type::TypeTy); I != E; ++I)
563 UT.erase(cast<Type>(I->second));
565 // UT now contains types that are not named. Loop over it, naming structure
568 bool Changed = false;
569 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
571 if (const StructType *ST = dyn_cast<StructType>(*I)) {
572 ((Value*)ST)->setName("unnamed", &MST);
578 // generateCompilerSpecificCode - This is where we add conditional compilation
579 // directives to cater to specific compilers as need be.
581 static void generateCompilerSpecificCode(std::ostream& Out) {
582 // Alloca is hard to get, and we don't want to include stdlib.h here...
583 Out << "/* get a declaration for alloca */\n"
585 << "extern void *__builtin_alloca(unsigned long);\n"
586 << "#define alloca(x) __builtin_alloca(x)\n"
588 << "#ifndef __FreeBSD__\n"
589 << "#include <alloca.h>\n"
593 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
594 // If we aren't being compiled with GCC, just drop these attributes.
595 Out << "#ifndef __GNUC__ /* Can only support \"linkonce\" vars with GCC */\n"
596 << "#define __attribute__(X)\n"
600 void CWriter::printModule(Module *M) {
601 // Calculate which global values have names that will collide when we throw
602 // away type information.
603 { // Scope to delete the FoundNames set when we are done with it...
604 std::set<std::string> FoundNames;
605 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
606 if (I->hasName()) // If the global has a name...
607 if (FoundNames.count(I->getName())) // And the name is already used
608 MangledGlobals.insert(I); // Mangle the name
610 FoundNames.insert(I->getName()); // Otherwise, keep track of name
612 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
613 if (I->hasName()) // If the global has a name...
614 if (FoundNames.count(I->getName())) // And the name is already used
615 MangledGlobals.insert(I); // Mangle the name
617 FoundNames.insert(I->getName()); // Otherwise, keep track of name
620 // get declaration for alloca
621 Out << "/* Provide Declarations */\n";
622 Out << "#include <stdarg.h>\n";
623 Out << "#include <setjmp.h>\n";
624 generateCompilerSpecificCode(Out);
626 // Provide a definition for `bool' if not compiling with a C++ compiler.
628 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
630 << "\n\n/* Support for floating point constants */\n"
631 << "typedef unsigned long long ConstantDoubleTy;\n"
632 << "typedef unsigned int ConstantFloatTy;\n"
634 << "\n\n/* Support for the invoke instruction */\n"
635 << "extern struct __llvm_jmpbuf_list_t {\n"
636 << " jmp_buf buf; struct __llvm_jmpbuf_list_t *next;\n"
637 << "} *__llvm_jmpbuf_list;\n"
639 << "\n\n/* Global Declarations */\n";
641 // First output all the declarations for the program, because C requires
642 // Functions & globals to be declared before they are used.
645 // Loop over the symbol table, emitting all named constants...
646 printSymbolTable(M->getSymbolTable());
648 // Global variable declarations...
650 Out << "\n/* External Global Variable Declarations */\n";
651 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
652 if (I->hasExternalLinkage()) {
654 printType(Out, I->getType()->getElementType(), getValueName(I));
660 // Function declarations
662 Out << "\n/* Function Declarations */\n";
664 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
665 // If the function is external and the name collides don't print it.
666 // Sometimes the bytecode likes to have multiple "declarations" for
667 // external functions
668 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
669 !I->getIntrinsicID()) {
670 printFunctionSignature(I, true);
676 // Print Malloc prototype if needed
678 Out << "\n/* Malloc to make sun happy */\n";
679 Out << "extern void * malloc(size_t);\n\n";
682 // Output the global variable declarations
684 Out << "\n\n/* Global Variable Declarations */\n";
685 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
686 if (!I->isExternal()) {
688 printType(Out, I->getType()->getElementType(), getValueName(I));
695 // Output the global variable definitions and contents...
697 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
698 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
699 if (!I->isExternal()) {
700 if (I->hasInternalLinkage())
702 printType(Out, I->getType()->getElementType(), getValueName(I));
703 if (I->hasLinkOnceLinkage())
704 Out << " __attribute__((common))";
705 if (!I->getInitializer()->isNullValue()) {
707 writeOperand(I->getInitializer());
713 // Output all of the functions...
714 emittedInvoke = false;
716 Out << "\n\n/* Function Bodies */\n";
717 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
721 // If the program included an invoke instruction, we need to output the
722 // support code for it here!
724 Out << "\n/* More support for the invoke instruction */\n"
725 << "struct __llvm_jmpbuf_list_t *__llvm_jmpbuf_list "
726 << "__attribute__((common)) = 0;\n";
731 /// printSymbolTable - Run through symbol table looking for type names. If a
732 /// type name is found, emit it's declaration...
734 void CWriter::printSymbolTable(const SymbolTable &ST) {
735 // If there are no type names, exit early.
736 if (ST.find(Type::TypeTy) == ST.end())
739 // We are only interested in the type plane of the symbol table...
740 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
741 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
743 // Print out forward declarations for structure types before anything else!
744 Out << "/* Structure forward decls */\n";
745 for (; I != End; ++I)
746 if (const Type *STy = dyn_cast<StructType>(I->second)) {
747 std::string Name = "struct l_" + makeNameProper(I->first);
748 Out << Name << ";\n";
749 TypeNames.insert(std::make_pair(STy, Name));
754 // Now we can print out typedefs...
755 Out << "/* Typedefs */\n";
756 for (I = ST.type_begin(Type::TypeTy); I != End; ++I) {
757 const Type *Ty = cast<Type>(I->second);
758 std::string Name = "l_" + makeNameProper(I->first);
760 printType(Out, Ty, Name);
766 // Keep track of which structures have been printed so far...
767 std::set<const StructType *> StructPrinted;
769 // Loop over all structures then push them into the stack so they are
770 // printed in the correct order.
772 Out << "/* Structure contents */\n";
773 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
774 if (const StructType *STy = dyn_cast<StructType>(I->second))
775 printContainedStructs(STy, StructPrinted);
778 // Push the struct onto the stack and recursively push all structs
779 // this one depends on.
780 void CWriter::printContainedStructs(const Type *Ty,
781 std::set<const StructType*> &StructPrinted){
782 if (const StructType *STy = dyn_cast<StructType>(Ty)){
783 //Check to see if we have already printed this struct
784 if (StructPrinted.count(STy) == 0) {
785 // Print all contained types first...
786 for (StructType::ElementTypes::const_iterator
787 I = STy->getElementTypes().begin(),
788 E = STy->getElementTypes().end(); I != E; ++I) {
789 const Type *Ty1 = I->get();
790 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
791 printContainedStructs(*I, StructPrinted);
794 //Print structure type out..
795 StructPrinted.insert(STy);
796 std::string Name = TypeNames[STy];
797 printType(Out, STy, Name, true);
801 // If it is an array, check contained types and continue
802 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
803 const Type *Ty1 = ATy->getElementType();
804 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
805 printContainedStructs(Ty1, StructPrinted);
810 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
811 // If the program provides its own malloc prototype we don't need
812 // to include the general one.
813 if (getValueName(F) == "malloc")
816 if (F->hasInternalLinkage()) Out << "static ";
817 if (F->hasLinkOnceLinkage()) Out << "inline ";
819 // Loop over the arguments, printing them...
820 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
822 std::stringstream FunctionInnards;
824 // Print out the name...
825 FunctionInnards << getValueName(F) << "(";
827 if (!F->isExternal()) {
830 if (F->abegin()->hasName() || !Prototype)
831 ArgName = getValueName(F->abegin());
832 printType(FunctionInnards, F->afront().getType(), ArgName);
833 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
835 FunctionInnards << ", ";
836 if (I->hasName() || !Prototype)
837 ArgName = getValueName(I);
840 printType(FunctionInnards, I->getType(), ArgName);
844 // Loop over the arguments, printing them...
845 for (FunctionType::ParamTypes::const_iterator I =
846 FT->getParamTypes().begin(),
847 E = FT->getParamTypes().end(); I != E; ++I) {
848 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
849 printType(FunctionInnards, *I);
853 // Finish printing arguments... if this is a vararg function, print the ...,
854 // unless there are no known types, in which case, we just emit ().
856 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
857 if (FT->getParamTypes().size()) FunctionInnards << ", ";
858 FunctionInnards << "..."; // Output varargs portion of signature!
860 FunctionInnards << ")";
861 // Print out the return type and the entire signature for that matter
862 printType(Out, F->getReturnType(), FunctionInnards.str());
866 void CWriter::printFunction(Function *F) {
867 if (F->isExternal()) return;
869 Table->incorporateFunction(F);
871 printFunctionSignature(F, false);
874 // print local variable information for the function
875 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
876 if (const AllocaInst *AI = isDirectAlloca(*I)) {
878 printType(Out, AI->getAllocatedType(), getValueName(AI));
879 Out << "; /* Address exposed local */\n";
880 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
882 printType(Out, (*I)->getType(), getValueName(*I));
885 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
887 printType(Out, (*I)->getType(), getValueName(*I)+"__PHI_TEMPORARY");
894 // Scan the function for floating point constants. If any FP constant is used
895 // in the function, we want to redirect it here so that we do not depend on
896 // the precision of the printed form, unless the printed form preserves
899 unsigned FPCounter = 0;
900 for (constant_iterator I = constant_begin(F), E = constant_end(F); I != E;++I)
901 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
902 if ((!FPCSafeToPrint(FPC)) // Do not put in FPConstantMap if safe.
903 && (FPConstantMap.find(FPC) == FPConstantMap.end())) {
904 double Val = FPC->getValue();
906 FPConstantMap[FPC] = FPCounter; // Number the FP constants
908 if (FPC->getType() == Type::DoubleTy)
909 Out << " const ConstantDoubleTy FloatConstant" << FPCounter++
910 << " = 0x" << std::hex << *(unsigned long long*)&Val << std::dec
911 << "; /* " << Val << " */\n";
912 else if (FPC->getType() == Type::FloatTy) {
914 Out << " const ConstantFloatTy FloatConstant" << FPCounter++
915 << " = 0x" << std::hex << *(unsigned*)&fVal << std::dec
916 << "; /* " << Val << " */\n";
918 assert(0 && "Unknown float type!");
923 // print the basic blocks
924 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
925 BasicBlock *Prev = BB->getPrev();
927 // Don't print the label for the basic block if there are no uses, or if the
928 // only terminator use is the precessor basic block's terminator. We have
929 // to scan the use list because PHI nodes use basic blocks too but do not
930 // require a label to be generated.
932 bool NeedsLabel = false;
933 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
935 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
936 if (TI != Prev->getTerminator() ||
937 isa<SwitchInst>(Prev->getTerminator())) {
942 if (NeedsLabel) Out << getValueName(BB) << ":\n";
944 // Output all of the instructions in the basic block...
945 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
946 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
947 if (II->getType() != Type::VoidTy)
956 // Don't emit prefix or suffix for the terminator...
957 visit(*BB->getTerminator());
961 Table->purgeFunction();
962 FPConstantMap.clear();
965 // Specific Instruction type classes... note that all of the casts are
966 // neccesary because we use the instruction classes as opaque types...
968 void CWriter::visitReturnInst(ReturnInst &I) {
969 // Don't output a void return if this is the last basic block in the function
970 if (I.getNumOperands() == 0 &&
971 &*--I.getParent()->getParent()->end() == I.getParent() &&
972 !I.getParent()->size() == 1) {
977 if (I.getNumOperands()) {
979 writeOperand(I.getOperand(0));
984 void CWriter::visitSwitchInst(SwitchInst &SI) {
986 writeOperand(SI.getOperand(0));
987 Out << ") {\n default:\n";
988 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
990 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
992 writeOperand(SI.getOperand(i));
994 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
995 printBranchToBlock(SI.getParent(), Succ, 2);
996 if (Succ == SI.getParent()->getNext())
1002 void CWriter::visitInvokeInst(InvokeInst &II) {
1004 << " struct __llvm_jmpbuf_list_t Entry;\n"
1005 << " Entry.next = __llvm_jmpbuf_list;\n"
1006 << " if (setjmp(Entry.buf)) {\n"
1007 << " __llvm_jmpbuf_list = Entry.next;\n";
1008 printBranchToBlock(II.getParent(), II.getExceptionalDest(), 4);
1010 << " __llvm_jmpbuf_list = &Entry;\n"
1014 << " __llvm_jmpbuf_list = Entry.next;\n"
1016 printBranchToBlock(II.getParent(), II.getNormalDest(), 0);
1017 emittedInvoke = true;
1021 static bool isGotoCodeNeccessary(BasicBlock *From, BasicBlock *To) {
1022 // If PHI nodes need copies, we need the copy code...
1023 if (isa<PHINode>(To->front()) ||
1024 From->getNext() != To) // Not directly successor, need goto
1027 // Otherwise we don't need the code.
1031 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
1033 for (BasicBlock::iterator I = Succ->begin();
1034 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1035 // now we have to do the printing
1036 Out << std::string(Indent, ' ');
1037 Out << " " << getValueName(I) << "__PHI_TEMPORARY = ";
1038 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1039 Out << "; /* for PHI node */\n";
1042 if (CurBB->getNext() != Succ) {
1043 Out << std::string(Indent, ' ') << " goto ";
1049 // Brach instruction printing - Avoid printing out a brach to a basic block that
1050 // immediately succeeds the current one.
1052 void CWriter::visitBranchInst(BranchInst &I) {
1053 if (I.isConditional()) {
1054 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(0))) {
1056 writeOperand(I.getCondition());
1059 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1061 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(1))) {
1062 Out << " } else {\n";
1063 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1066 // First goto not neccesary, assume second one is...
1068 writeOperand(I.getCondition());
1071 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1076 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1081 // PHI nodes get copied into temporary values at the end of predecessor basic
1082 // blocks. We now need to copy these temporary values into the REAL value for
1084 void CWriter::visitPHINode(PHINode &I) {
1086 Out << "__PHI_TEMPORARY";
1090 void CWriter::visitBinaryOperator(Instruction &I) {
1091 // binary instructions, shift instructions, setCond instructions.
1092 assert(!isa<PointerType>(I.getType()));
1094 // We must cast the results of binary operations which might be promoted.
1095 bool needsCast = false;
1096 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1097 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1098 || (I.getType() == Type::FloatTy)) {
1101 printType(Out, I.getType(), "", false, false);
1105 writeOperand(I.getOperand(0));
1107 switch (I.getOpcode()) {
1108 case Instruction::Add: Out << " + "; break;
1109 case Instruction::Sub: Out << " - "; break;
1110 case Instruction::Mul: Out << "*"; break;
1111 case Instruction::Div: Out << "/"; break;
1112 case Instruction::Rem: Out << "%"; break;
1113 case Instruction::And: Out << " & "; break;
1114 case Instruction::Or: Out << " | "; break;
1115 case Instruction::Xor: Out << " ^ "; break;
1116 case Instruction::SetEQ: Out << " == "; break;
1117 case Instruction::SetNE: Out << " != "; break;
1118 case Instruction::SetLE: Out << " <= "; break;
1119 case Instruction::SetGE: Out << " >= "; break;
1120 case Instruction::SetLT: Out << " < "; break;
1121 case Instruction::SetGT: Out << " > "; break;
1122 case Instruction::Shl : Out << " << "; break;
1123 case Instruction::Shr : Out << " >> "; break;
1124 default: std::cerr << "Invalid operator type!" << I; abort();
1127 writeOperand(I.getOperand(1));
1134 void CWriter::visitCastInst(CastInst &I) {
1135 if (I.getType() == Type::BoolTy) {
1137 writeOperand(I.getOperand(0));
1142 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1144 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1145 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1146 // Avoid "cast to pointer from integer of different size" warnings
1150 writeOperand(I.getOperand(0));
1153 void CWriter::visitCallInst(CallInst &I) {
1154 // Handle intrinsic function calls first...
1155 if (Function *F = I.getCalledFunction())
1156 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID()) {
1158 default: assert(0 && "Unknown LLVM intrinsic!");
1159 case LLVMIntrinsic::va_start:
1160 Out << "va_start((va_list)*";
1161 writeOperand(I.getOperand(1));
1163 // Output the last argument to the enclosing function...
1164 writeOperand(&I.getParent()->getParent()->aback());
1167 case LLVMIntrinsic::va_end:
1168 Out << "va_end((va_list)*";
1169 writeOperand(I.getOperand(1));
1172 case LLVMIntrinsic::va_copy:
1173 Out << "va_copy((va_list)*";
1174 writeOperand(I.getOperand(1));
1175 Out << ", (va_list)";
1176 writeOperand(I.getOperand(2));
1180 case LLVMIntrinsic::setjmp:
1181 Out << "setjmp(*(jmp_buf*)";
1182 writeOperand(I.getOperand(1));
1185 case LLVMIntrinsic::longjmp:
1186 Out << "longjmp(*(jmp_buf*)";
1187 writeOperand(I.getOperand(1));
1189 writeOperand(I.getOperand(2));
1197 void CWriter::visitCallSite(CallSite CS) {
1198 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1199 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1200 const Type *RetTy = FTy->getReturnType();
1202 writeOperand(CS.getCalledValue());
1205 if (CS.arg_begin() != CS.arg_end()) {
1206 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1209 for (++AI; AI != AE; ++AI) {
1217 void CWriter::visitMallocInst(MallocInst &I) {
1219 printType(Out, I.getType());
1220 Out << ")malloc(sizeof(";
1221 printType(Out, I.getType()->getElementType());
1224 if (I.isArrayAllocation()) {
1226 writeOperand(I.getOperand(0));
1231 void CWriter::visitAllocaInst(AllocaInst &I) {
1233 printType(Out, I.getType());
1234 Out << ") alloca(sizeof(";
1235 printType(Out, I.getType()->getElementType());
1237 if (I.isArrayAllocation()) {
1239 writeOperand(I.getOperand(0));
1244 void CWriter::visitFreeInst(FreeInst &I) {
1246 writeOperand(I.getOperand(0));
1250 void CWriter::printIndexingExpression(Value *Ptr, User::op_iterator I,
1251 User::op_iterator E) {
1252 bool HasImplicitAddress = false;
1253 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1254 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1255 HasImplicitAddress = true;
1256 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1257 HasImplicitAddress = true;
1258 Ptr = CPR->getValue(); // Get to the global...
1259 } else if (isDirectAlloca(Ptr)) {
1260 HasImplicitAddress = true;
1264 if (!HasImplicitAddress)
1265 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1267 writeOperandInternal(Ptr);
1271 const Constant *CI = dyn_cast<Constant>(I);
1272 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1275 writeOperandInternal(Ptr);
1277 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1279 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1282 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1283 "Can only have implicit address with direct accessing");
1285 if (HasImplicitAddress) {
1287 } else if (CI && CI->isNullValue() && I+1 != E) {
1288 // Print out the -> operator if possible...
1289 if ((*(I+1))->getType() == Type::UByteTy) {
1290 Out << (HasImplicitAddress ? "." : "->");
1291 Out << "field" << cast<ConstantUInt>(*(I+1))->getValue();
1297 if ((*I)->getType() == Type::LongTy) {
1302 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
1306 void CWriter::visitLoadInst(LoadInst &I) {
1308 writeOperand(I.getOperand(0));
1311 void CWriter::visitStoreInst(StoreInst &I) {
1313 writeOperand(I.getPointerOperand());
1315 writeOperand(I.getOperand(0));
1318 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1320 printIndexingExpression(I.getPointerOperand(), I.idx_begin(), I.idx_end());
1323 void CWriter::visitVarArgInst(VarArgInst &I) {
1324 Out << "va_arg((va_list)*";
1325 writeOperand(I.getOperand(0));
1327 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1332 //===----------------------------------------------------------------------===//
1333 // External Interface declaration
1334 //===----------------------------------------------------------------------===//
1336 Pass *createWriteToCPass(std::ostream &o) { return new CWriter(o); }