1 //===-- Writer.cpp - Library for converting LLVM code to C ----------------===//
3 // This library implements the functionality defined in llvm/Assembly/CWriter.h
5 // TODO : Recursive types.
7 //===-----------------------------------------------------------------------==//
9 #include "llvm/Assembly/CWriter.h"
10 #include "llvm/Constants.h"
11 #include "llvm/DerivedTypes.h"
12 #include "llvm/Module.h"
13 #include "llvm/iMemory.h"
14 #include "llvm/iTerminators.h"
15 #include "llvm/iPHINode.h"
16 #include "llvm/iOther.h"
17 #include "llvm/iOperators.h"
18 #include "llvm/SymbolTable.h"
19 #include "llvm/SlotCalculator.h"
20 #include "llvm/Support/InstVisitor.h"
21 #include "llvm/Support/InstIterator.h"
22 #include "Support/StringExtras.h"
23 #include "Support/STLExtras.h"
30 static std::string getConstStrValue(const Constant* CPV);
33 static std::string getConstArrayStrValue(const Constant* CPV) {
36 // As a special case, print the array as a string if it is an array of
37 // ubytes or an array of sbytes with positive values.
39 const Type *ETy = cast<ArrayType>(CPV->getType())->getElementType();
40 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
42 // Make sure the last character is a null char, as automatically added by C
43 if (CPV->getNumOperands() == 0 ||
44 !cast<Constant>(*(CPV->op_end()-1))->isNullValue())
49 // Do not include the last character, which we know is null
50 for (unsigned i = 0, e = CPV->getNumOperands()-1; i != e; ++i) {
51 unsigned char C = (ETy == Type::SByteTy) ?
52 (unsigned char)cast<ConstantSInt>(CPV->getOperand(i))->getValue() :
53 (unsigned char)cast<ConstantUInt>(CPV->getOperand(i))->getValue();
59 case '\n': Result += "\\n"; break;
60 case '\t': Result += "\\t"; break;
61 case '\r': Result += "\\r"; break;
62 case '\v': Result += "\\v"; break;
63 case '\a': Result += "\\a"; break;
66 Result += ( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A');
67 Result += ((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A');
75 if (CPV->getNumOperands()) {
76 Result += " " + getConstStrValue(cast<Constant>(CPV->getOperand(0)));
77 for (unsigned i = 1; i < CPV->getNumOperands(); i++)
78 Result += ", " + getConstStrValue(cast<Constant>(CPV->getOperand(i)));
86 static std::string getConstStrValue(const Constant* CPV) {
87 switch (CPV->getType()->getPrimitiveID()) {
88 case Type::BoolTyID: return CPV == ConstantBool::False ? "0" : "1";
91 case Type::IntTyID: return itostr(cast<ConstantSInt>(CPV)->getValue());
92 case Type::LongTyID: return itostr(cast<ConstantSInt>(CPV)->getValue())+"ll";
95 case Type::UShortTyID:return utostr(cast<ConstantUInt>(CPV)->getValue());
96 case Type::UIntTyID: return utostr(cast<ConstantUInt>(CPV)->getValue())+"u";
97 case Type::ULongTyID:return utostr(cast<ConstantUInt>(CPV)->getValue())+"ull";
100 case Type::DoubleTyID: return ftostr(cast<ConstantFP>(CPV)->getValue());
102 case Type::ArrayTyID: return getConstArrayStrValue(CPV);
104 case Type::StructTyID: {
105 std::string Result = "{";
106 if (CPV->getNumOperands()) {
107 Result += " " + getConstStrValue(cast<Constant>(CPV->getOperand(0)));
108 for (unsigned i = 1; i < CPV->getNumOperands(); i++)
109 Result += ", " + getConstStrValue(cast<Constant>(CPV->getOperand(i)));
111 return Result + " }";
115 std::cerr << "Unknown constant type: " << CPV << "\n";
121 // Pass the Type* variable and and the variable name and this prints out the
122 // variable declaration.
124 static string calcTypeNameVar(const Type *Ty,
125 map<const Type *, string> &TypeNames,
126 const string &NameSoFar, bool ignoreName = false){
127 if (Ty->isPrimitiveType())
128 switch (Ty->getPrimitiveID()) {
129 case Type::VoidTyID: return "void " + NameSoFar;
130 case Type::BoolTyID: return "bool " + NameSoFar;
131 case Type::UByteTyID: return "unsigned char " + NameSoFar;
132 case Type::SByteTyID: return "signed char " + NameSoFar;
133 case Type::UShortTyID: return "unsigned short " + NameSoFar;
134 case Type::ShortTyID: return "short " + NameSoFar;
135 case Type::UIntTyID: return "unsigned " + NameSoFar;
136 case Type::IntTyID: return "int " + NameSoFar;
137 case Type::ULongTyID: return "unsigned long long " + NameSoFar;
138 case Type::LongTyID: return "signed long long " + NameSoFar;
139 case Type::FloatTyID: return "float " + NameSoFar;
140 case Type::DoubleTyID: return "double " + NameSoFar;
142 std::cerr << "Unknown primitive type: " << Ty << "\n";
146 // Check to see if the type is named.
148 map<const Type *, string>::iterator I = TypeNames.find(Ty);
149 if (I != TypeNames.end())
150 return I->second + " " + NameSoFar;
154 switch (Ty->getPrimitiveID()) {
155 case Type::FunctionTyID: {
156 const FunctionType *MTy = cast<FunctionType>(Ty);
157 Result += calcTypeNameVar(MTy->getReturnType(), TypeNames, "");
158 Result += " " + NameSoFar + " (";
159 for (FunctionType::ParamTypes::const_iterator
160 I = MTy->getParamTypes().begin(),
161 E = MTy->getParamTypes().end(); I != E; ++I) {
162 if (I != MTy->getParamTypes().begin())
164 Result += calcTypeNameVar(*I, TypeNames, "");
166 if (MTy->isVarArg()) {
167 if (!MTy->getParamTypes().empty())
173 case Type::StructTyID: {
174 const StructType *STy = cast<const StructType>(Ty);
175 Result = NameSoFar + " {\n";
177 for (StructType::ElementTypes::const_iterator
178 I = STy->getElementTypes().begin(),
179 E = STy->getElementTypes().end(); I != E; ++I) {
180 Result += " " +calcTypeNameVar(*I, TypeNames, "field" + utostr(indx++));
186 case Type::PointerTyID:
187 return calcTypeNameVar(cast<const PointerType>(Ty)->getElementType(),
188 TypeNames, "*" + NameSoFar);
190 case Type::ArrayTyID: {
191 const ArrayType *ATy = cast<const ArrayType>(Ty);
192 int NumElements = ATy->getNumElements();
193 return calcTypeNameVar(ATy->getElementType(), TypeNames,
194 NameSoFar + "[" + itostr(NumElements) + "]");
197 assert(0 && "Unhandled case in getTypeProps!");
205 class CWriter : public InstVisitor<CWriter> {
207 SlotCalculator &Table;
208 const Module *TheModule;
209 map<const Type *, string> TypeNames;
210 std::set<const Value*> MangledGlobals;
212 inline CWriter(ostream &o, SlotCalculator &Tab, const Module *M)
213 : Out(o), Table(Tab), TheModule(M) {
216 inline void write(Module *M) { printModule(M); }
218 ostream& printType(const Type *Ty, const string &VariableName = "") {
219 return Out << calcTypeNameVar(Ty, TypeNames, VariableName);
222 void writeOperand(Value *Operand);
223 void writeOperandInternal(Value *Operand);
225 string getValueName(const Value *V);
228 void printModule(Module *M);
229 void printSymbolTable(const SymbolTable &ST);
230 void printGlobal(const GlobalVariable *GV);
231 void printFunctionSignature(const Function *F);
232 void printFunctionDecl(const Function *F); // Print just the forward decl
234 void printFunction(Function *);
236 // isInlinableInst - Attempt to inline instructions into their uses to build
237 // trees as much as possible. To do this, we have to consistently decide
238 // what is acceptable to inline, so that variable declarations don't get
239 // printed and an extra copy of the expr is not emitted.
241 static bool isInlinableInst(const Instruction &I) {
242 // Must be an expression, must be used exactly once. If it is dead, we
243 // emit it inline where it would go.
244 if (I.getType() == Type::VoidTy || I.use_size() != 1 ||
245 isa<TerminatorInst>(I) || isa<CallInst>(I) || isa<PHINode>(I))
248 // Only inline instruction it it's use is in the same BB as the inst.
249 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
252 // Instruction visitation functions
253 friend class InstVisitor<CWriter>;
255 void visitReturnInst(ReturnInst &I);
256 void visitBranchInst(BranchInst &I);
258 void visitPHINode(PHINode &I) {}
259 void visitNot(GenericUnaryInst &I);
260 void visitBinaryOperator(Instruction &I);
262 void visitCastInst (CastInst &I);
263 void visitCallInst (CallInst &I);
264 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
266 void visitMallocInst(MallocInst &I);
267 void visitAllocaInst(AllocaInst &I);
268 void visitFreeInst (FreeInst &I);
269 void visitLoadInst (LoadInst &I);
270 void visitStoreInst (StoreInst &I);
271 void visitGetElementPtrInst(GetElementPtrInst &I);
273 void visitInstruction(Instruction &I) {
274 std::cerr << "C Writer does not know about " << I;
278 void outputLValue(Instruction *I) {
279 Out << " " << getValueName(I) << " = ";
281 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
283 void printIndexingExpr(MemAccessInst &MAI);
287 // We dont want identifier names with ., space, - in them.
288 // So we replace them with _
289 static string makeNameProper(string x) {
291 for (string::iterator sI = x.begin(), sEnd = x.end(); sI != sEnd; sI++)
293 case '.': tmp += "d_"; break;
294 case ' ': tmp += "s_"; break;
295 case '-': tmp += "D_"; break;
302 string CWriter::getValueName(const Value *V) {
303 if (V->hasName()) { // Print out the label if it exists...
304 if (isa<GlobalValue>(V) && // Do not mangle globals...
305 cast<GlobalValue>(V)->hasExternalLinkage() && // Unless it's internal or
306 !MangledGlobals.count(V)) // Unless the name would collide if we don't
307 return makeNameProper(V->getName());
309 return "l" + utostr(V->getType()->getUniqueID()) + "_" +
310 makeNameProper(V->getName());
313 int Slot = Table.getValSlot(V);
314 assert(Slot >= 0 && "Invalid value!");
315 return "ltmp_" + itostr(Slot) + "_" + utostr(V->getType()->getUniqueID());
318 void CWriter::writeOperandInternal(Value *Operand) {
319 if (Operand->hasName()) {
320 Out << getValueName(Operand);
321 } else if (Constant *CPV = dyn_cast<Constant>(Operand)) {
322 if (isa<ConstantPointerNull>(CPV)) {
324 printType(CPV->getType(), "");
327 Out << getConstStrValue(CPV);
329 int Slot = Table.getValSlot(Operand);
330 assert(Slot >= 0 && "Malformed LLVM!");
331 Out << "ltmp_" << Slot << "_" << Operand->getType()->getUniqueID();
335 void CWriter::writeOperand(Value *Operand) {
336 if (Instruction *I = dyn_cast<Instruction>(Operand))
337 if (isInlinableInst(*I)) {
338 // Should we inline this instruction to build a tree?
345 if (isa<GlobalVariable>(Operand))
346 Out << "(&"; // Global variables are references as their addresses by llvm
348 writeOperandInternal(Operand);
350 if (isa<GlobalVariable>(Operand))
354 void CWriter::printModule(Module *M) {
355 // Calculate which global values have names that will collide when we throw
356 // away type information.
357 { // Scope to delete the FoundNames set when we are done with it...
358 std::set<string> FoundNames;
359 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
360 if (I->hasName()) // If the global has a name...
361 if (FoundNames.count(I->getName())) // And the name is already used
362 MangledGlobals.insert(I); // Mangle the name
364 FoundNames.insert(I->getName()); // Otherwise, keep track of name
366 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
367 if (I->hasName()) // If the global has a name...
368 if (FoundNames.count(I->getName())) // And the name is already used
369 MangledGlobals.insert(I); // Mangle the name
371 FoundNames.insert(I->getName()); // Otherwise, keep track of name
375 // printing stdlib inclusion
376 // Out << "#include <stdlib.h>\n";
378 // get declaration for alloca
379 Out << "/* Provide Declarations */\n"
380 << "#include <malloc.h>\n"
381 << "#include <alloca.h>\n\n"
383 // Provide a definition for null if one does not already exist.
384 << "#ifndef NULL\n#define NULL 0\n#endif\n\n"
385 << "typedef unsigned char bool;\n"
387 << "\n\n/* Global Symbols */\n";
389 // Loop over the symbol table, emitting all named constants...
390 if (M->hasSymbolTable())
391 printSymbolTable(*M->getSymbolTable());
393 Out << "\n\n/* Global Data */\n";
394 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
395 if (I->hasInternalLinkage()) Out << "static ";
396 printType(I->getType()->getElementType(), getValueName(I));
398 if (I->hasInitializer()) {
400 writeOperand(I->getInitializer());
405 // First output all the declarations of the functions as C requires Functions
406 // be declared before they are used.
408 Out << "\n\n/* Function Declarations */\n";
409 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
410 printFunctionDecl(I);
412 // Output all of the functions...
413 Out << "\n\n/* Function Bodies */\n";
414 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
419 // printSymbolTable - Run through symbol table looking for named constants
420 // if a named constant is found, emit it's declaration...
421 // Assuming that symbol table has only types and constants.
422 void CWriter::printSymbolTable(const SymbolTable &ST) {
423 for (SymbolTable::const_iterator TI = ST.begin(); TI != ST.end(); ++TI) {
424 SymbolTable::type_const_iterator I = ST.type_begin(TI->first);
425 SymbolTable::type_const_iterator End = ST.type_end(TI->first);
427 for (; I != End; ++I)
428 if (const Type *Ty = dyn_cast<StructType>(I->second)) {
429 string Name = "struct l_" + makeNameProper(I->first);
430 Out << Name << ";\n";
432 TypeNames.insert(std::make_pair(Ty, Name));
438 for (SymbolTable::const_iterator TI = ST.begin(); TI != ST.end(); ++TI) {
439 SymbolTable::type_const_iterator I = ST.type_begin(TI->first);
440 SymbolTable::type_const_iterator End = ST.type_end(TI->first);
442 for (; I != End; ++I) {
443 const Value *V = I->second;
444 if (const Type *Ty = dyn_cast<Type>(V)) {
445 string Name = "l_" + makeNameProper(I->first);
446 if (isa<StructType>(Ty))
447 Name = "struct " + makeNameProper(Name);
451 Out << calcTypeNameVar(Ty, TypeNames, Name, true) << ";\n";
458 // printFunctionDecl - Print function declaration
460 void CWriter::printFunctionDecl(const Function *F) {
461 printFunctionSignature(F);
465 void CWriter::printFunctionSignature(const Function *F) {
466 if (F->hasInternalLinkage()) Out << "static ";
468 // Loop over the arguments, printing them...
469 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
471 // Print out the return type and name...
472 printType(F->getReturnType());
473 Out << getValueName(F) << "(";
475 if (!F->isExternal()) {
477 printType(F->afront().getType(), getValueName(F->abegin()));
479 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
482 printType(I->getType(), getValueName(I));
486 // Loop over the arguments, printing them...
487 for (FunctionType::ParamTypes::const_iterator I =
488 FT->getParamTypes().begin(),
489 E = FT->getParamTypes().end(); I != E; ++I) {
490 if (I != FT->getParamTypes().begin()) Out << ", ";
495 // Finish printing arguments...
496 if (FT->isVarArg()) {
497 if (FT->getParamTypes().size()) Out << ", ";
498 Out << "..."; // Output varargs portion of signature!
504 void CWriter::printFunction(Function *F) {
505 if (F->isExternal()) return;
507 Table.incorporateFunction(F);
509 printFunctionSignature(F);
512 // print local variable information for the function
513 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
514 if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
516 printType((*I)->getType(), getValueName(*I));
520 // print the basic blocks
521 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
522 BasicBlock *Prev = BB->getPrev();
524 // Don't print the label for the basic block if there are no uses, or if the
525 // only terminator use is the precessor basic block's terminator. We have
526 // to scan the use list because PHI nodes use basic blocks too but do not
527 // require a label to be generated.
529 bool NeedsLabel = false;
530 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
532 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
533 if (TI != Prev->getTerminator()) {
538 if (NeedsLabel) Out << getValueName(BB) << ":\n";
540 // Output all of the instructions in the basic block...
541 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
542 if (!isInlinableInst(*II) && !isa<PHINode>(*II)) {
543 if (II->getType() != Type::VoidTy)
552 // Don't emit prefix or suffix for the terminator...
553 visit(*BB->getTerminator());
557 Table.purgeFunction();
560 // Specific Instruction type classes... note that all of the casts are
561 // neccesary because we use the instruction classes as opaque types...
563 void CWriter::visitReturnInst(ReturnInst &I) {
564 // Don't output a void return if this is the last basic block in the function
565 if (I.getNumOperands() == 0 &&
566 &*--I.getParent()->getParent()->end() == I.getParent() &&
567 !I.getParent()->size() == 1) {
572 if (I.getNumOperands()) {
574 writeOperand(I.getOperand(0));
579 static bool isGotoCodeNeccessary(BasicBlock *From, BasicBlock *To) {
580 // If PHI nodes need copies, we need the copy code...
581 if (isa<PHINode>(To->front()) ||
582 From->getNext() != To) // Not directly successor, need goto
585 // Otherwise we don't need the code.
589 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
591 for (BasicBlock::iterator I = Succ->begin();
592 PHINode *PN = dyn_cast<PHINode>(&*I); ++I) {
593 // now we have to do the printing
594 Out << string(Indent, ' ');
596 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
597 Out << "; /* for PHI node */\n";
600 if (CurBB->getNext() != Succ) {
601 Out << string(Indent, ' ') << " goto ";
607 // Brach instruction printing - Avoid printing out a brach to a basic block that
608 // immediately succeeds the current one.
610 void CWriter::visitBranchInst(BranchInst &I) {
611 if (I.isConditional()) {
612 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(0))) {
614 writeOperand(I.getCondition());
617 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
619 if (isGotoCodeNeccessary(I.getParent(), I.getSuccessor(1))) {
620 Out << " } else {\n";
621 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
624 // First goto not neccesary, assume second one is...
626 writeOperand(I.getCondition());
629 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
634 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
640 void CWriter::visitNot(GenericUnaryInst &I) {
642 writeOperand(I.getOperand(0));
645 void CWriter::visitBinaryOperator(Instruction &I) {
646 // binary instructions, shift instructions, setCond instructions.
647 if (isa<PointerType>(I.getType())) {
649 printType(I.getType());
653 if (isa<PointerType>(I.getType())) Out << "(long long)";
654 writeOperand(I.getOperand(0));
656 switch (I.getOpcode()) {
657 case Instruction::Add: Out << " + "; break;
658 case Instruction::Sub: Out << " - "; break;
659 case Instruction::Mul: Out << "*"; break;
660 case Instruction::Div: Out << "/"; break;
661 case Instruction::Rem: Out << "%"; break;
662 case Instruction::And: Out << " & "; break;
663 case Instruction::Or: Out << " | "; break;
664 case Instruction::Xor: Out << " ^ "; break;
665 case Instruction::SetEQ: Out << " == "; break;
666 case Instruction::SetNE: Out << " != "; break;
667 case Instruction::SetLE: Out << " <= "; break;
668 case Instruction::SetGE: Out << " >= "; break;
669 case Instruction::SetLT: Out << " < "; break;
670 case Instruction::SetGT: Out << " > "; break;
671 case Instruction::Shl : Out << " << "; break;
672 case Instruction::Shr : Out << " >> "; break;
673 default: std::cerr << "Invalid operator type!" << I; abort();
676 if (isa<PointerType>(I.getType())) Out << "(long long)";
677 writeOperand(I.getOperand(1));
680 void CWriter::visitCastInst(CastInst &I) {
682 printType(I.getType());
684 writeOperand(I.getOperand(0));
687 void CWriter::visitCallInst(CallInst &I) {
688 const PointerType *PTy = cast<PointerType>(I.getCalledValue()->getType());
689 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
690 const Type *RetTy = FTy->getReturnType();
692 Out << getValueName(I.getOperand(0)) << "(";
694 if (I.getNumOperands() > 1) {
695 writeOperand(I.getOperand(1));
697 for (unsigned op = 2, Eop = I.getNumOperands(); op != Eop; ++op) {
699 writeOperand(I.getOperand(op));
705 void CWriter::visitMallocInst(MallocInst &I) {
707 printType(I.getType());
708 Out << ")malloc(sizeof(";
709 printType(I.getType()->getElementType());
712 if (I.isArrayAllocation()) {
714 writeOperand(I.getOperand(0));
719 void CWriter::visitAllocaInst(AllocaInst &I) {
721 printType(I.getType());
722 Out << ") alloca(sizeof(";
723 printType(I.getType()->getElementType());
725 if (I.isArrayAllocation()) {
727 writeOperand(I.getOperand(0));
732 void CWriter::visitFreeInst(FreeInst &I) {
734 writeOperand(I.getOperand(0));
738 void CWriter::printIndexingExpr(MemAccessInst &MAI) {
739 MemAccessInst::op_iterator I = MAI.idx_begin(), E = MAI.idx_end();
741 // If accessing a global value with no indexing, avoid *(&GV) syndrome
742 if (GlobalValue *V = dyn_cast<GlobalValue>(MAI.getPointerOperand())) {
743 writeOperandInternal(V);
747 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
750 writeOperand(MAI.getPointerOperand());
754 // Print out the -> operator if possible...
755 const Constant *CI = dyn_cast<Constant>(I->get());
756 if (CI && CI->isNullValue() && I+1 != E &&
757 (*(I+1))->getType() == Type::UByteTy) {
758 Out << "->field" << cast<ConstantUInt>(*(I+1))->getValue();
763 if ((*I)->getType() == Type::UIntTy) {
768 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
772 void CWriter::visitLoadInst(LoadInst &I) {
773 printIndexingExpr(I);
776 void CWriter::visitStoreInst(StoreInst &I) {
777 printIndexingExpr(I);
779 writeOperand(I.getOperand(0));
782 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
784 printIndexingExpr(I);
787 //===----------------------------------------------------------------------===//
788 // External Interface declaration
789 //===----------------------------------------------------------------------===//
791 void WriteToC(const Module *M, ostream &Out) {
792 assert(M && "You can't write a null module!!");
793 SlotCalculator SlotTable(M, false);
794 CWriter W(Out, SlotTable, M);