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) || isa<VarArgInst>(I))
95 // Don't inline a load across a store or other bad things!
98 // Only inline instruction it it's use is in the same BB as the inst.
99 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
102 // isDirectAlloca - Define fixed sized allocas in the entry block as direct
103 // variables which are accessed with the & operator. This causes GCC to
104 // generate significantly better code than to emit alloca calls directly.
106 static const AllocaInst *isDirectAlloca(const Value *V) {
107 const AllocaInst *AI = dyn_cast<AllocaInst>(V);
108 if (!AI) return false;
109 if (AI->isArrayAllocation())
110 return 0; // FIXME: we can also inline fixed size array allocas!
111 if (AI->getParent() != &AI->getParent()->getParent()->getEntryNode())
116 // Instruction visitation functions
117 friend class InstVisitor<CWriter>;
119 void visitReturnInst(ReturnInst &I);
120 void visitBranchInst(BranchInst &I);
121 void visitSwitchInst(SwitchInst &I);
122 void visitInvokeInst(InvokeInst &I);
124 void visitPHINode(PHINode &I);
125 void visitBinaryOperator(Instruction &I);
127 void visitCastInst (CastInst &I);
128 void visitCallInst (CallInst &I);
129 void visitCallSite (CallSite CS);
130 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
132 void visitMallocInst(MallocInst &I);
133 void visitAllocaInst(AllocaInst &I);
134 void visitFreeInst (FreeInst &I);
135 void visitLoadInst (LoadInst &I);
136 void visitStoreInst (StoreInst &I);
137 void visitGetElementPtrInst(GetElementPtrInst &I);
138 void visitVarArgInst(VarArgInst &I);
140 void visitInstruction(Instruction &I) {
141 std::cerr << "C Writer does not know about " << I;
145 void outputLValue(Instruction *I) {
146 Out << " " << getValueName(I) << " = ";
148 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
150 void printIndexingExpression(Value *Ptr, User::op_iterator I,
151 User::op_iterator E);
155 // We dont want identifier names with ., space, - in them.
156 // So we replace them with _
157 static std::string makeNameProper(std::string x) {
159 for (std::string::iterator sI = x.begin(), sEnd = x.end(); sI != sEnd; sI++)
161 case '.': tmp += "d_"; break;
162 case ' ': tmp += "s_"; break;
163 case '-': tmp += "D_"; break;
170 std::string CWriter::getValueName(const Value *V) {
171 if (V->hasName()) { // Print out the label if it exists...
173 // Name mangling occurs as follows:
174 // - If V is not a global, mangling always occurs.
175 // - Otherwise, mangling occurs when any of the following are true:
176 // 1) V has internal linkage
177 // 2) V's name would collide if it is not mangled.
180 if(const GlobalValue* gv = dyn_cast<GlobalValue>(V)) {
181 if(!gv->hasInternalLinkage() && !MangledGlobals.count(gv)) {
182 // No internal linkage, name will not collide -> no mangling.
183 return makeNameProper(gv->getName());
187 // Non-global, or global with internal linkage / colliding name -> mangle.
188 return "l" + utostr(V->getType()->getUniqueID()) + "_" +
189 makeNameProper(V->getName());
192 int Slot = Table->getValSlot(V);
193 assert(Slot >= 0 && "Invalid value!");
194 return "ltmp_" + itostr(Slot) + "_" + utostr(V->getType()->getUniqueID());
197 // A pointer type should not use parens around *'s alone, e.g., (**)
198 inline bool ptrTypeNameNeedsParens(const std::string &NameSoFar) {
199 return (NameSoFar.find_last_not_of('*') != std::string::npos);
202 // Pass the Type* and the variable name and this prints out the variable
205 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
206 const std::string &NameSoFar,
207 bool IgnoreName, bool namedContext) {
208 if (Ty->isPrimitiveType())
209 switch (Ty->getPrimitiveID()) {
210 case Type::VoidTyID: return Out << "void " << NameSoFar;
211 case Type::BoolTyID: return Out << "bool " << NameSoFar;
212 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
213 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
214 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
215 case Type::ShortTyID: return Out << "short " << NameSoFar;
216 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
217 case Type::IntTyID: return Out << "int " << NameSoFar;
218 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
219 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
220 case Type::FloatTyID: return Out << "float " << NameSoFar;
221 case Type::DoubleTyID: return Out << "double " << NameSoFar;
223 std::cerr << "Unknown primitive type: " << Ty << "\n";
227 // Check to see if the type is named.
228 if (!IgnoreName || isa<OpaqueType>(Ty)) {
229 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
230 if (I != TypeNames.end()) {
231 return Out << I->second << " " << NameSoFar;
235 switch (Ty->getPrimitiveID()) {
236 case Type::FunctionTyID: {
237 const FunctionType *MTy = cast<FunctionType>(Ty);
238 std::stringstream FunctionInnards;
239 FunctionInnards << " (" << NameSoFar << ") (";
240 for (FunctionType::ParamTypes::const_iterator
241 I = MTy->getParamTypes().begin(),
242 E = MTy->getParamTypes().end(); I != E; ++I) {
243 if (I != MTy->getParamTypes().begin())
244 FunctionInnards << ", ";
245 printType(FunctionInnards, *I, "");
247 if (MTy->isVarArg()) {
248 if (!MTy->getParamTypes().empty())
249 FunctionInnards << ", ...";
250 } else if (MTy->getParamTypes().empty()) {
251 FunctionInnards << "void";
253 FunctionInnards << ")";
254 std::string tstr = FunctionInnards.str();
255 printType(Out, MTy->getReturnType(), tstr);
258 case Type::StructTyID: {
259 const StructType *STy = cast<StructType>(Ty);
260 Out << NameSoFar + " {\n";
262 for (StructType::ElementTypes::const_iterator
263 I = STy->getElementTypes().begin(),
264 E = STy->getElementTypes().end(); I != E; ++I) {
266 printType(Out, *I, "field" + utostr(Idx++));
272 case Type::PointerTyID: {
273 const PointerType *PTy = cast<PointerType>(Ty);
274 std::string ptrName = "*" + NameSoFar;
276 // Do not need parens around "* NameSoFar" if NameSoFar consists only
277 // of zero or more '*' chars *and* this is not an unnamed pointer type
278 // such as the result type in a cast statement. Otherwise, enclose in ( ).
279 if (ptrTypeNameNeedsParens(NameSoFar) || !namedContext ||
280 PTy->getElementType()->getPrimitiveID() == Type::ArrayTyID)
281 ptrName = "(" + ptrName + ")"; //
283 return printType(Out, PTy->getElementType(), ptrName);
286 case Type::ArrayTyID: {
287 const ArrayType *ATy = cast<ArrayType>(Ty);
288 unsigned NumElements = ATy->getNumElements();
289 return printType(Out, ATy->getElementType(),
290 NameSoFar + "[" + utostr(NumElements) + "]");
293 case Type::OpaqueTyID: {
294 static int Count = 0;
295 std::string TyName = "struct opaque_" + itostr(Count++);
296 assert(TypeNames.find(Ty) == TypeNames.end());
297 TypeNames[Ty] = TyName;
298 return Out << TyName << " " << NameSoFar;
301 assert(0 && "Unhandled case in getTypeProps!");
308 void CWriter::printConstantArray(ConstantArray *CPA) {
310 // As a special case, print the array as a string if it is an array of
311 // ubytes or an array of sbytes with positive values.
313 const Type *ETy = CPA->getType()->getElementType();
314 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
316 // Make sure the last character is a null char, as automatically added by C
317 if (isString && (CPA->getNumOperands() == 0 ||
318 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
323 // Keep track of whether the last number was a hexadecimal escape
324 bool LastWasHex = false;
326 // Do not include the last character, which we know is null
327 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
328 unsigned char C = cast<ConstantInt>(CPA->getOperand(i))->getRawValue();
330 // Print it out literally if it is a printable character. The only thing
331 // to be careful about is when the last letter output was a hex escape
332 // code, in which case we have to be careful not to print out hex digits
333 // explicitly (the C compiler thinks it is a continuation of the previous
334 // character, sheesh...)
336 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
338 if (C == '"' || C == '\\')
345 case '\n': Out << "\\n"; break;
346 case '\t': Out << "\\t"; break;
347 case '\r': Out << "\\r"; break;
348 case '\v': Out << "\\v"; break;
349 case '\a': Out << "\\a"; break;
350 case '\"': Out << "\\\""; break;
351 case '\'': Out << "\\\'"; break;
354 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
355 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
364 if (CPA->getNumOperands()) {
366 printConstant(cast<Constant>(CPA->getOperand(0)));
367 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
369 printConstant(cast<Constant>(CPA->getOperand(i)));
376 /// FPCSafeToPrint - Returns true if we may assume that CFP may be
377 /// written out textually as a double (rather than as a reference to a
378 /// stack-allocated variable). We decide this by converting CFP to a
379 /// string and back into a double, and then checking whether the
380 /// conversion results in a bit-equal double to the original value of
381 /// CFP. This depends on us and the target C compiler agreeing on the
382 /// conversion process (which is pretty likely since we only deal in
383 /// IEEE FP.) This is adapted from similar code in
384 /// lib/VMCore/AsmWriter.cpp:WriteConstantInt().
385 static bool FPCSafeToPrint (const ConstantFP *CFP) {
386 std::string StrVal = ftostr(CFP->getValue());
387 // Check to make sure that the stringized number is not some string like
388 // "Inf" or NaN, that atof will accept, but the lexer will not. Check that
389 // the string matches the "[-+]?[0-9]" regex.
390 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
391 ((StrVal[0] == '-' || StrVal[0] == '+') &&
392 (StrVal[1] >= '0' && StrVal[1] <= '9')))
393 // Reparse stringized version!
394 return (atof(StrVal.c_str()) == CFP->getValue());
398 // printConstant - The LLVM Constant to C Constant converter.
399 void CWriter::printConstant(Constant *CPV) {
400 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
401 switch (CE->getOpcode()) {
402 case Instruction::Cast:
404 printType(Out, CPV->getType());
406 printConstant(CE->getOperand(0));
410 case Instruction::GetElementPtr:
412 printIndexingExpression(CE->getOperand(0),
413 CPV->op_begin()+1, CPV->op_end());
416 case Instruction::Add:
418 printConstant(CE->getOperand(0));
420 printConstant(CE->getOperand(1));
423 case Instruction::Sub:
425 printConstant(CE->getOperand(0));
427 printConstant(CE->getOperand(1));
432 std::cerr << "CWriter Error: Unhandled constant expression: "
438 switch (CPV->getType()->getPrimitiveID()) {
440 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
441 case Type::SByteTyID:
442 case Type::ShortTyID:
443 Out << cast<ConstantSInt>(CPV)->getValue(); break;
445 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
446 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
448 Out << cast<ConstantSInt>(CPV)->getValue();
452 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
454 case Type::UByteTyID:
455 case Type::UShortTyID:
456 Out << cast<ConstantUInt>(CPV)->getValue(); break;
458 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
459 case Type::ULongTyID:
460 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
462 case Type::FloatTyID:
463 case Type::DoubleTyID: {
464 ConstantFP *FPC = cast<ConstantFP>(CPV);
465 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
466 if (I != FPConstantMap.end()) {
467 // Because of FP precision problems we must load from a stack allocated
468 // value that holds the value in hex.
469 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
470 << "*)&FloatConstant" << I->second << ")";
472 if (FPCSafeToPrint (FPC)) {
473 Out << ftostr (FPC->getValue ());
475 Out << FPC->getValue(); // Who knows? Give it our best shot...
481 case Type::ArrayTyID:
482 printConstantArray(cast<ConstantArray>(CPV));
485 case Type::StructTyID: {
487 if (CPV->getNumOperands()) {
489 printConstant(cast<Constant>(CPV->getOperand(0)));
490 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
492 printConstant(cast<Constant>(CPV->getOperand(i)));
499 case Type::PointerTyID:
500 if (isa<ConstantPointerNull>(CPV)) {
502 printType(Out, CPV->getType());
503 Out << ")/*NULL*/0)";
505 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
506 writeOperand(CPR->getValue());
511 std::cerr << "Unknown constant type: " << CPV << "\n";
516 void CWriter::writeOperandInternal(Value *Operand) {
517 if (Instruction *I = dyn_cast<Instruction>(Operand))
518 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
519 // Should we inline this instruction to build a tree?
526 if (Operand->hasName()) {
527 Out << getValueName(Operand);
528 } else if (Constant *CPV = dyn_cast<Constant>(Operand)) {
531 int Slot = Table->getValSlot(Operand);
532 assert(Slot >= 0 && "Malformed LLVM!");
533 Out << "ltmp_" << Slot << "_" << Operand->getType()->getUniqueID();
537 void CWriter::writeOperand(Value *Operand) {
538 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
539 Out << "(&"; // Global variables are references as their addresses by llvm
541 writeOperandInternal(Operand);
543 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
547 // nameAllUsedStructureTypes - If there are structure types in the module that
548 // are used but do not have names assigned to them in the symbol table yet then
549 // we assign them names now.
551 bool CWriter::nameAllUsedStructureTypes(Module &M) {
552 // Get a set of types that are used by the program...
553 std::set<const Type *> UT = getAnalysis<FindUsedTypes>().getTypes();
555 // Loop over the module symbol table, removing types from UT that are already
558 SymbolTable &MST = M.getSymbolTable();
559 if (MST.find(Type::TypeTy) != MST.end())
560 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
561 E = MST.type_end(Type::TypeTy); I != E; ++I)
562 UT.erase(cast<Type>(I->second));
564 // UT now contains types that are not named. Loop over it, naming structure
567 bool Changed = false;
568 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
570 if (const StructType *ST = dyn_cast<StructType>(*I)) {
571 ((Value*)ST)->setName("unnamed", &MST);
577 // generateCompilerSpecificCode - This is where we add conditional compilation
578 // directives to cater to specific compilers as need be.
580 static void generateCompilerSpecificCode(std::ostream& Out) {
581 // Alloca is hard to get, and we don't want to include stdlib.h here...
582 Out << "/* get a declaration for alloca */\n"
584 << "extern void *__builtin_alloca(unsigned long);\n"
585 << "#define alloca(x) __builtin_alloca(x)\n"
587 << "#ifndef __FreeBSD__\n"
588 << "#include <alloca.h>\n"
592 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
593 // If we aren't being compiled with GCC, just drop these attributes.
594 Out << "#ifndef __GNUC__ /* Can only support \"linkonce\" vars with GCC */\n"
595 << "#define __attribute__(X)\n"
599 void CWriter::printModule(Module *M) {
600 // Calculate which global values have names that will collide when we throw
601 // away type information.
602 { // Scope to delete the FoundNames set when we are done with it...
603 std::set<std::string> FoundNames;
604 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
605 if (I->hasName()) // If the global has a name...
606 if (FoundNames.count(I->getName())) // And the name is already used
607 MangledGlobals.insert(I); // Mangle the name
609 FoundNames.insert(I->getName()); // Otherwise, keep track of name
611 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
612 if (I->hasName()) // If the global has a name...
613 if (FoundNames.count(I->getName())) // And the name is already used
614 MangledGlobals.insert(I); // Mangle the name
616 FoundNames.insert(I->getName()); // Otherwise, keep track of name
619 // get declaration for alloca
620 Out << "/* Provide Declarations */\n";
621 Out << "#include <stdarg.h>\n";
622 Out << "#include <setjmp.h>\n";
623 generateCompilerSpecificCode(Out);
625 // Provide a definition for `bool' if not compiling with a C++ compiler.
627 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
629 << "\n\n/* Support for floating point constants */\n"
630 << "typedef unsigned long long ConstantDoubleTy;\n"
631 << "typedef unsigned int ConstantFloatTy;\n"
633 << "\n\n/* Support for the invoke instruction */\n"
634 << "extern struct __llvm_jmpbuf_list_t {\n"
635 << " jmp_buf buf; struct __llvm_jmpbuf_list_t *next;\n"
636 << "} *__llvm_jmpbuf_list;\n"
638 << "\n\n/* Global Declarations */\n";
640 // First output all the declarations for the program, because C requires
641 // Functions & globals to be declared before they are used.
644 // Loop over the symbol table, emitting all named constants...
645 printSymbolTable(M->getSymbolTable());
647 // Global variable declarations...
649 Out << "\n/* External Global Variable Declarations */\n";
650 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
651 if (I->hasExternalLinkage()) {
653 printType(Out, I->getType()->getElementType(), getValueName(I));
659 // Function declarations
661 Out << "\n/* Function Declarations */\n";
663 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
664 // If the function is external and the name collides don't print it.
665 // Sometimes the bytecode likes to have multiple "declarations" for
666 // external functions
667 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
668 !I->getIntrinsicID()) {
669 printFunctionSignature(I, true);
675 // Print Malloc prototype if needed
677 Out << "\n/* Malloc to make sun happy */\n";
678 Out << "extern void * malloc();\n\n";
681 // Output the global variable declarations
683 Out << "\n\n/* Global Variable Declarations */\n";
684 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
685 if (!I->isExternal()) {
687 printType(Out, I->getType()->getElementType(), getValueName(I));
693 // Output the global variable definitions and contents...
695 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
696 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
697 if (!I->isExternal()) {
698 if (I->hasInternalLinkage())
700 printType(Out, I->getType()->getElementType(), getValueName(I));
701 if (I->hasLinkOnceLinkage())
702 Out << " __attribute__((common))";
703 if (!I->getInitializer()->isNullValue()) {
705 writeOperand(I->getInitializer());
711 // Output all of the functions...
712 emittedInvoke = false;
714 Out << "\n\n/* Function Bodies */\n";
715 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
719 // If the program included an invoke instruction, we need to output the
720 // support code for it here!
722 Out << "\n/* More support for the invoke instruction */\n"
723 << "struct __llvm_jmpbuf_list_t *__llvm_jmpbuf_list "
724 << "__attribute__((common)) = 0;\n";
729 /// printSymbolTable - Run through symbol table looking for type names. If a
730 /// type name is found, emit it's declaration...
732 void CWriter::printSymbolTable(const SymbolTable &ST) {
733 // If there are no type names, exit early.
734 if (ST.find(Type::TypeTy) == ST.end())
737 // We are only interested in the type plane of the symbol table...
738 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
739 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
741 // Print out forward declarations for structure types before anything else!
742 Out << "/* Structure forward decls */\n";
743 for (; I != End; ++I)
744 if (const Type *STy = dyn_cast<StructType>(I->second)) {
745 std::string Name = "struct l_" + makeNameProper(I->first);
746 Out << Name << ";\n";
747 TypeNames.insert(std::make_pair(STy, Name));
752 // Now we can print out typedefs...
753 Out << "/* Typedefs */\n";
754 for (I = ST.type_begin(Type::TypeTy); I != End; ++I) {
755 const Type *Ty = cast<Type>(I->second);
756 std::string Name = "l_" + makeNameProper(I->first);
758 printType(Out, Ty, Name);
764 // Keep track of which structures have been printed so far...
765 std::set<const StructType *> StructPrinted;
767 // Loop over all structures then push them into the stack so they are
768 // printed in the correct order.
770 Out << "/* Structure contents */\n";
771 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
772 if (const StructType *STy = dyn_cast<StructType>(I->second))
773 printContainedStructs(STy, StructPrinted);
776 // Push the struct onto the stack and recursively push all structs
777 // this one depends on.
778 void CWriter::printContainedStructs(const Type *Ty,
779 std::set<const StructType*> &StructPrinted){
780 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
781 //Check to see if we have already printed this struct
782 if (StructPrinted.count(STy) == 0) {
783 // Print all contained types first...
784 for (StructType::ElementTypes::const_iterator
785 I = STy->getElementTypes().begin(),
786 E = STy->getElementTypes().end(); I != E; ++I) {
787 const Type *Ty1 = I->get();
788 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
789 printContainedStructs(*I, StructPrinted);
792 //Print structure type out..
793 StructPrinted.insert(STy);
794 std::string Name = TypeNames[STy];
795 printType(Out, STy, Name, true);
799 // If it is an array, check contained types and continue
800 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
801 const Type *Ty1 = ATy->getElementType();
802 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
803 printContainedStructs(Ty1, StructPrinted);
808 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
809 // If the program provides its own malloc prototype we don't need
810 // to include the general one.
811 if (getValueName(F) == "malloc")
814 if (F->hasInternalLinkage()) Out << "static ";
815 if (F->hasLinkOnceLinkage()) Out << "inline ";
817 // Loop over the arguments, printing them...
818 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
820 std::stringstream FunctionInnards;
822 // Print out the name...
823 FunctionInnards << getValueName(F) << "(";
825 if (!F->isExternal()) {
828 if (F->abegin()->hasName() || !Prototype)
829 ArgName = getValueName(F->abegin());
830 printType(FunctionInnards, F->afront().getType(), ArgName);
831 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
833 FunctionInnards << ", ";
834 if (I->hasName() || !Prototype)
835 ArgName = getValueName(I);
838 printType(FunctionInnards, I->getType(), ArgName);
842 // Loop over the arguments, printing them...
843 for (FunctionType::ParamTypes::const_iterator I =
844 FT->getParamTypes().begin(),
845 E = FT->getParamTypes().end(); I != E; ++I) {
846 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
847 printType(FunctionInnards, *I);
851 // Finish printing arguments... if this is a vararg function, print the ...,
852 // unless there are no known types, in which case, we just emit ().
854 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
855 if (FT->getParamTypes().size()) FunctionInnards << ", ";
856 FunctionInnards << "..."; // Output varargs portion of signature!
858 FunctionInnards << ")";
859 // Print out the return type and the entire signature for that matter
860 printType(Out, F->getReturnType(), FunctionInnards.str());
863 void CWriter::printFunction(Function *F) {
864 if (F->isExternal()) return;
866 Table->incorporateFunction(F);
868 printFunctionSignature(F, false);
871 // print local variable information for the function
872 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
873 if (const AllocaInst *AI = isDirectAlloca(*I)) {
875 printType(Out, AI->getAllocatedType(), getValueName(AI));
876 Out << "; /* Address exposed local */\n";
877 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
879 printType(Out, (*I)->getType(), getValueName(*I));
882 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
884 printType(Out, (*I)->getType(), getValueName(*I)+"__PHI_TEMPORARY");
891 // Scan the function for floating point constants. If any FP constant is used
892 // in the function, we want to redirect it here so that we do not depend on
893 // the precision of the printed form, unless the printed form preserves
896 unsigned FPCounter = 0;
897 for (constant_iterator I = constant_begin(F), E = constant_end(F); I != E;++I)
898 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
899 if ((!FPCSafeToPrint(FPC)) // Do not put in FPConstantMap if safe.
900 && (FPConstantMap.find(FPC) == FPConstantMap.end())) {
901 double Val = FPC->getValue();
903 FPConstantMap[FPC] = FPCounter; // Number the FP constants
905 if (FPC->getType() == Type::DoubleTy)
906 Out << " const ConstantDoubleTy FloatConstant" << FPCounter++
907 << " = 0x" << std::hex << *(unsigned long long*)&Val << std::dec
908 << "; /* " << Val << " */\n";
909 else if (FPC->getType() == Type::FloatTy) {
911 Out << " const ConstantFloatTy FloatConstant" << FPCounter++
912 << " = 0x" << std::hex << *(unsigned*)&fVal << std::dec
913 << "; /* " << Val << " */\n";
915 assert(0 && "Unknown float type!");
920 // print the basic blocks
921 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
922 BasicBlock *Prev = BB->getPrev();
924 // Don't print the label for the basic block if there are no uses, or if the
925 // only terminator use is the precessor basic block's terminator. We have
926 // to scan the use list because PHI nodes use basic blocks too but do not
927 // require a label to be generated.
929 bool NeedsLabel = false;
930 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
932 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
933 if (TI != Prev->getTerminator() ||
934 isa<SwitchInst>(Prev->getTerminator())) {
939 if (NeedsLabel) Out << getValueName(BB) << ":\n";
941 // Output all of the instructions in the basic block...
942 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
943 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
944 if (II->getType() != Type::VoidTy)
953 // Don't emit prefix or suffix for the terminator...
954 visit(*BB->getTerminator());
958 Table->purgeFunction();
959 FPConstantMap.clear();
962 // Specific Instruction type classes... note that all of the casts are
963 // neccesary because we use the instruction classes as opaque types...
965 void CWriter::visitReturnInst(ReturnInst &I) {
966 // Don't output a void return if this is the last basic block in the function
967 if (I.getNumOperands() == 0 &&
968 &*--I.getParent()->getParent()->end() == I.getParent() &&
969 !I.getParent()->size() == 1) {
974 if (I.getNumOperands()) {
976 writeOperand(I.getOperand(0));
981 void CWriter::visitSwitchInst(SwitchInst &SI) {
983 writeOperand(SI.getOperand(0));
984 Out << ") {\n default:\n";
985 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
987 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
989 writeOperand(SI.getOperand(i));
991 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
992 printBranchToBlock(SI.getParent(), Succ, 2);
993 if (Succ == SI.getParent()->getNext())
999 void CWriter::visitInvokeInst(InvokeInst &II) {
1001 << " struct __llvm_jmpbuf_list_t Entry;\n"
1002 << " Entry.next = __llvm_jmpbuf_list;\n"
1003 << " if (setjmp(Entry.buf)) {\n"
1004 << " __llvm_jmpbuf_list = Entry.next;\n";
1005 printBranchToBlock(II.getParent(), II.getExceptionalDest(), 4);
1007 << " __llvm_jmpbuf_list = &Entry;\n"
1011 << " __llvm_jmpbuf_list = Entry.next;\n"
1013 printBranchToBlock(II.getParent(), II.getNormalDest(), 0);
1014 emittedInvoke = true;
1018 static bool isGotoCodeNeccessary(BasicBlock *From, BasicBlock *To) {
1019 // If PHI nodes need copies, we need the copy code...
1020 if (isa<PHINode>(To->front()) ||
1021 From->getNext() != To) // Not directly successor, need goto
1024 // Otherwise we don't need the code.
1028 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
1030 for (BasicBlock::iterator I = Succ->begin();
1031 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1032 // now we have to do the printing
1033 Out << std::string(Indent, ' ');
1034 Out << " " << getValueName(I) << "__PHI_TEMPORARY = ";
1035 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1036 Out << "; /* for PHI node */\n";
1039 if (CurBB->getNext() != Succ) {
1040 Out << std::string(Indent, ' ') << " goto ";
1046 // Brach instruction printing - Avoid printing out a brach to a basic block that
1047 // immediately succeeds the current one.
1049 void CWriter::visitBranchInst(BranchInst &I) {
1050 if (I.isConditional()) {
1051 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(0))) {
1053 writeOperand(I.getCondition());
1056 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1058 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(1))) {
1059 Out << " } else {\n";
1060 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1063 // First goto not neccesary, assume second one is...
1065 writeOperand(I.getCondition());
1068 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1073 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1078 // PHI nodes get copied into temporary values at the end of predecessor basic
1079 // blocks. We now need to copy these temporary values into the REAL value for
1081 void CWriter::visitPHINode(PHINode &I) {
1083 Out << "__PHI_TEMPORARY";
1087 void CWriter::visitBinaryOperator(Instruction &I) {
1088 // binary instructions, shift instructions, setCond instructions.
1089 assert(!isa<PointerType>(I.getType()));
1091 // We must cast the results of binary operations which might be promoted.
1092 bool needsCast = false;
1093 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1094 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1095 || (I.getType() == Type::FloatTy)) {
1098 printType(Out, I.getType(), "", false, false);
1102 writeOperand(I.getOperand(0));
1104 switch (I.getOpcode()) {
1105 case Instruction::Add: Out << " + "; break;
1106 case Instruction::Sub: Out << " - "; break;
1107 case Instruction::Mul: Out << "*"; break;
1108 case Instruction::Div: Out << "/"; break;
1109 case Instruction::Rem: Out << "%"; break;
1110 case Instruction::And: Out << " & "; break;
1111 case Instruction::Or: Out << " | "; break;
1112 case Instruction::Xor: Out << " ^ "; break;
1113 case Instruction::SetEQ: Out << " == "; break;
1114 case Instruction::SetNE: Out << " != "; break;
1115 case Instruction::SetLE: Out << " <= "; break;
1116 case Instruction::SetGE: Out << " >= "; break;
1117 case Instruction::SetLT: Out << " < "; break;
1118 case Instruction::SetGT: Out << " > "; break;
1119 case Instruction::Shl : Out << " << "; break;
1120 case Instruction::Shr : Out << " >> "; break;
1121 default: std::cerr << "Invalid operator type!" << I; abort();
1124 writeOperand(I.getOperand(1));
1131 void CWriter::visitCastInst(CastInst &I) {
1132 if (I.getType() == Type::BoolTy) {
1134 writeOperand(I.getOperand(0));
1139 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1141 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1142 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1143 // Avoid "cast to pointer from integer of different size" warnings
1147 writeOperand(I.getOperand(0));
1150 void CWriter::visitCallInst(CallInst &I) {
1151 // Handle intrinsic function calls first...
1152 if (Function *F = I.getCalledFunction())
1153 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID()) {
1155 default: assert(0 && "Unknown LLVM intrinsic!");
1156 case LLVMIntrinsic::va_start:
1157 Out << "va_start((va_list)*";
1158 writeOperand(I.getOperand(1));
1160 // Output the last argument to the enclosing function...
1161 writeOperand(&I.getParent()->getParent()->aback());
1164 case LLVMIntrinsic::va_end:
1165 Out << "va_end((va_list)*";
1166 writeOperand(I.getOperand(1));
1169 case LLVMIntrinsic::va_copy:
1170 Out << "va_copy((va_list)*";
1171 writeOperand(I.getOperand(1));
1172 Out << ", (va_list)";
1173 writeOperand(I.getOperand(2));
1177 case LLVMIntrinsic::setjmp:
1178 Out << "setjmp(*(jmp_buf*)";
1179 writeOperand(I.getOperand(1));
1182 case LLVMIntrinsic::longjmp:
1183 Out << "longjmp(*(jmp_buf*)";
1184 writeOperand(I.getOperand(1));
1186 writeOperand(I.getOperand(2));
1194 void CWriter::visitCallSite(CallSite CS) {
1195 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1196 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1197 const Type *RetTy = FTy->getReturnType();
1199 writeOperand(CS.getCalledValue());
1202 if (CS.arg_begin() != CS.arg_end()) {
1203 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1206 for (++AI; AI != AE; ++AI) {
1214 void CWriter::visitMallocInst(MallocInst &I) {
1216 printType(Out, I.getType());
1217 Out << ")malloc(sizeof(";
1218 printType(Out, I.getType()->getElementType());
1221 if (I.isArrayAllocation()) {
1223 writeOperand(I.getOperand(0));
1228 void CWriter::visitAllocaInst(AllocaInst &I) {
1230 printType(Out, I.getType());
1231 Out << ") alloca(sizeof(";
1232 printType(Out, I.getType()->getElementType());
1234 if (I.isArrayAllocation()) {
1236 writeOperand(I.getOperand(0));
1241 void CWriter::visitFreeInst(FreeInst &I) {
1243 writeOperand(I.getOperand(0));
1247 void CWriter::printIndexingExpression(Value *Ptr, User::op_iterator I,
1248 User::op_iterator E) {
1249 bool HasImplicitAddress = false;
1250 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1251 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1252 HasImplicitAddress = true;
1253 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1254 HasImplicitAddress = true;
1255 Ptr = CPR->getValue(); // Get to the global...
1256 } else if (isDirectAlloca(Ptr)) {
1257 HasImplicitAddress = true;
1261 if (!HasImplicitAddress)
1262 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1264 writeOperandInternal(Ptr);
1268 const Constant *CI = dyn_cast<Constant>(I);
1269 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1272 writeOperandInternal(Ptr);
1274 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1276 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1279 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1280 "Can only have implicit address with direct accessing");
1282 if (HasImplicitAddress) {
1284 } else if (CI && CI->isNullValue() && I+1 != E) {
1285 // Print out the -> operator if possible...
1286 if ((*(I+1))->getType() == Type::UByteTy) {
1287 Out << (HasImplicitAddress ? "." : "->");
1288 Out << "field" << cast<ConstantUInt>(*(I+1))->getValue();
1294 if ((*I)->getType() == Type::LongTy) {
1299 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
1303 void CWriter::visitLoadInst(LoadInst &I) {
1305 writeOperand(I.getOperand(0));
1308 void CWriter::visitStoreInst(StoreInst &I) {
1310 writeOperand(I.getPointerOperand());
1312 writeOperand(I.getOperand(0));
1315 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1317 printIndexingExpression(I.getPointerOperand(), I.idx_begin(), I.idx_end());
1320 void CWriter::visitVarArgInst(VarArgInst &I) {
1321 Out << "va_arg((va_list)*";
1322 writeOperand(I.getOperand(0));
1324 printType(Out, I.getType(), "", /*ignoreName*/false, /*namedContext*/false);
1329 //===----------------------------------------------------------------------===//
1330 // External Interface declaration
1331 //===----------------------------------------------------------------------===//
1333 Pass *createWriteToCPass(std::ostream &o) { return new CWriter(o); }