1 //===-- Writer.cpp - Library for converting LLVM code to C ----------------===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the LLVM research group and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This library converts LLVM code to C code, compilable by GCC and other C
13 //===----------------------------------------------------------------------===//
15 #include "llvm/Assembly/CWriter.h"
16 #include "llvm/Constants.h"
17 #include "llvm/DerivedTypes.h"
18 #include "llvm/Module.h"
19 #include "llvm/Instructions.h"
20 #include "llvm/Pass.h"
21 #include "llvm/PassManager.h"
22 #include "llvm/SymbolTable.h"
23 #include "llvm/Intrinsics.h"
24 #include "llvm/Analysis/FindUsedTypes.h"
25 #include "llvm/Analysis/ConstantsScanner.h"
26 #include "llvm/Transforms/Scalar.h"
27 #include "llvm/Support/CallSite.h"
28 #include "llvm/Support/GetElementPtrTypeIterator.h"
29 #include "llvm/Support/InstVisitor.h"
30 #include "llvm/Support/Mangler.h"
31 #include "Support/StringExtras.h"
37 class CWriter : public Pass, public InstVisitor<CWriter> {
40 const Module *TheModule;
43 std::map<const Type *, std::string> TypeNames;
44 std::set<const Value*> MangledGlobals;
47 std::map<const ConstantFP *, unsigned> FPConstantMap;
49 CWriter(std::ostream &o) : Out(o) {}
51 void getAnalysisUsage(AnalysisUsage &AU) const {
52 AU.addRequired<FindUsedTypes>();
55 virtual const char *getPassName() const { return "C backend"; }
57 bool doInitialization(Module &M);
61 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
68 MangledGlobals.clear();
72 std::ostream &printType(std::ostream &Out, const Type *Ty,
73 const std::string &VariableName = "",
74 bool IgnoreName = false);
76 void writeOperand(Value *Operand);
77 void writeOperandInternal(Value *Operand);
80 bool nameAllUsedStructureTypes(Module &M);
81 void printModule(Module *M);
82 void printFloatingPointConstants(Module &M);
83 void printSymbolTable(const SymbolTable &ST);
84 void printContainedStructs(const Type *Ty, std::set<const StructType *> &);
85 void printFunctionSignature(const Function *F, bool Prototype);
87 void printFunction(Function &);
89 void printConstant(Constant *CPV);
90 void printConstantArray(ConstantArray *CPA);
92 // isInlinableInst - Attempt to inline instructions into their uses to build
93 // trees as much as possible. To do this, we have to consistently decide
94 // what is acceptable to inline, so that variable declarations don't get
95 // printed and an extra copy of the expr is not emitted.
97 static bool isInlinableInst(const Instruction &I) {
98 // Must be an expression, must be used exactly once. If it is dead, we
99 // emit it inline where it would go.
100 if (I.getType() == Type::VoidTy || !I.hasOneUse() ||
101 isa<TerminatorInst>(I) || isa<CallInst>(I) || isa<PHINode>(I) ||
102 isa<LoadInst>(I) || isa<VAArgInst>(I) || isa<VANextInst>(I))
103 // Don't inline a load across a store or other bad things!
106 // Only inline instruction it it's use is in the same BB as the inst.
107 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
110 // isDirectAlloca - Define fixed sized allocas in the entry block as direct
111 // variables which are accessed with the & operator. This causes GCC to
112 // generate significantly better code than to emit alloca calls directly.
114 static const AllocaInst *isDirectAlloca(const Value *V) {
115 const AllocaInst *AI = dyn_cast<AllocaInst>(V);
116 if (!AI) return false;
117 if (AI->isArrayAllocation())
118 return 0; // FIXME: we can also inline fixed size array allocas!
119 if (AI->getParent() != &AI->getParent()->getParent()->getEntryBlock())
124 // Instruction visitation functions
125 friend class InstVisitor<CWriter>;
127 void visitReturnInst(ReturnInst &I);
128 void visitBranchInst(BranchInst &I);
129 void visitSwitchInst(SwitchInst &I);
130 void visitInvokeInst(InvokeInst &I);
131 void visitUnwindInst(UnwindInst &I);
133 void visitPHINode(PHINode &I);
134 void visitBinaryOperator(Instruction &I);
136 void visitCastInst (CastInst &I);
137 void visitCallInst (CallInst &I);
138 void visitCallSite (CallSite CS);
139 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
141 void visitMallocInst(MallocInst &I);
142 void visitAllocaInst(AllocaInst &I);
143 void visitFreeInst (FreeInst &I);
144 void visitLoadInst (LoadInst &I);
145 void visitStoreInst (StoreInst &I);
146 void visitGetElementPtrInst(GetElementPtrInst &I);
147 void visitVANextInst(VANextInst &I);
148 void visitVAArgInst (VAArgInst &I);
150 void visitInstruction(Instruction &I) {
151 std::cerr << "C Writer does not know about " << I;
155 void outputLValue(Instruction *I) {
156 Out << " " << Mang->getValueName(I) << " = ";
158 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
160 void printIndexingExpression(Value *Ptr, gep_type_iterator I,
161 gep_type_iterator E);
164 // Pass the Type* and the variable name and this prints out the variable
167 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
168 const std::string &NameSoFar,
170 if (Ty->isPrimitiveType())
171 switch (Ty->getPrimitiveID()) {
172 case Type::VoidTyID: return Out << "void " << NameSoFar;
173 case Type::BoolTyID: return Out << "bool " << NameSoFar;
174 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
175 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
176 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
177 case Type::ShortTyID: return Out << "short " << NameSoFar;
178 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
179 case Type::IntTyID: return Out << "int " << NameSoFar;
180 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
181 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
182 case Type::FloatTyID: return Out << "float " << NameSoFar;
183 case Type::DoubleTyID: return Out << "double " << NameSoFar;
185 std::cerr << "Unknown primitive type: " << Ty << "\n";
189 // Check to see if the type is named.
190 if (!IgnoreName || isa<OpaqueType>(Ty)) {
191 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
192 if (I != TypeNames.end()) return Out << I->second << " " << NameSoFar;
195 switch (Ty->getPrimitiveID()) {
196 case Type::FunctionTyID: {
197 const FunctionType *MTy = cast<FunctionType>(Ty);
198 std::stringstream FunctionInnards;
199 FunctionInnards << " (" << NameSoFar << ") (";
200 for (FunctionType::param_iterator I = MTy->param_begin(),
201 E = MTy->param_end(); I != E; ++I) {
202 if (I != MTy->param_begin())
203 FunctionInnards << ", ";
204 printType(FunctionInnards, *I, "");
206 if (MTy->isVarArg()) {
207 if (MTy->getNumParams())
208 FunctionInnards << ", ...";
209 } else if (!MTy->getNumParams()) {
210 FunctionInnards << "void";
212 FunctionInnards << ")";
213 std::string tstr = FunctionInnards.str();
214 printType(Out, MTy->getReturnType(), tstr);
217 case Type::StructTyID: {
218 const StructType *STy = cast<StructType>(Ty);
219 Out << NameSoFar + " {\n";
221 for (StructType::element_iterator I = STy->element_begin(),
222 E = STy->element_end(); I != E; ++I) {
224 printType(Out, *I, "field" + utostr(Idx++));
230 case Type::PointerTyID: {
231 const PointerType *PTy = cast<PointerType>(Ty);
232 std::string ptrName = "*" + NameSoFar;
234 if (isa<ArrayType>(PTy->getElementType()))
235 ptrName = "(" + ptrName + ")";
237 return printType(Out, PTy->getElementType(), ptrName);
240 case Type::ArrayTyID: {
241 const ArrayType *ATy = cast<ArrayType>(Ty);
242 unsigned NumElements = ATy->getNumElements();
243 return printType(Out, ATy->getElementType(),
244 NameSoFar + "[" + utostr(NumElements) + "]");
247 case Type::OpaqueTyID: {
248 static int Count = 0;
249 std::string TyName = "struct opaque_" + itostr(Count++);
250 assert(TypeNames.find(Ty) == TypeNames.end());
251 TypeNames[Ty] = TyName;
252 return Out << TyName << " " << NameSoFar;
255 assert(0 && "Unhandled case in getTypeProps!");
262 void CWriter::printConstantArray(ConstantArray *CPA) {
264 // As a special case, print the array as a string if it is an array of
265 // ubytes or an array of sbytes with positive values.
267 const Type *ETy = CPA->getType()->getElementType();
268 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
270 // Make sure the last character is a null char, as automatically added by C
271 if (isString && (CPA->getNumOperands() == 0 ||
272 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
277 // Keep track of whether the last number was a hexadecimal escape
278 bool LastWasHex = false;
280 // Do not include the last character, which we know is null
281 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
282 unsigned char C = cast<ConstantInt>(CPA->getOperand(i))->getRawValue();
284 // Print it out literally if it is a printable character. The only thing
285 // to be careful about is when the last letter output was a hex escape
286 // code, in which case we have to be careful not to print out hex digits
287 // explicitly (the C compiler thinks it is a continuation of the previous
288 // character, sheesh...)
290 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
292 if (C == '"' || C == '\\')
299 case '\n': Out << "\\n"; break;
300 case '\t': Out << "\\t"; break;
301 case '\r': Out << "\\r"; break;
302 case '\v': Out << "\\v"; break;
303 case '\a': Out << "\\a"; break;
304 case '\"': Out << "\\\""; break;
305 case '\'': Out << "\\\'"; break;
308 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
309 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
318 if (CPA->getNumOperands()) {
320 printConstant(cast<Constant>(CPA->getOperand(0)));
321 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
323 printConstant(cast<Constant>(CPA->getOperand(i)));
330 // isFPCSafeToPrint - Returns true if we may assume that CFP may be written out
331 // textually as a double (rather than as a reference to a stack-allocated
332 // variable). We decide this by converting CFP to a string and back into a
333 // double, and then checking whether the conversion results in a bit-equal
334 // double to the original value of CFP. This depends on us and the target C
335 // compiler agreeing on the conversion process (which is pretty likely since we
336 // only deal in IEEE FP).
338 bool isFPCSafeToPrint(const ConstantFP *CFP) {
341 sprintf(Buffer, "%a", CFP->getValue());
343 if (!strncmp(Buffer, "0x", 2) ||
344 !strncmp(Buffer, "-0x", 3) ||
345 !strncmp(Buffer, "+0x", 3))
346 return atof(Buffer) == CFP->getValue();
349 std::string StrVal = ftostr(CFP->getValue());
351 while (StrVal[0] == ' ')
352 StrVal.erase(StrVal.begin());
354 // Check to make sure that the stringized number is not some string like "Inf"
355 // or NaN. Check that the string matches the "[-+]?[0-9]" regex.
356 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
357 ((StrVal[0] == '-' || StrVal[0] == '+') &&
358 (StrVal[1] >= '0' && StrVal[1] <= '9')))
359 // Reparse stringized version!
360 return atof(StrVal.c_str()) == CFP->getValue();
365 // printConstant - The LLVM Constant to C Constant converter.
366 void CWriter::printConstant(Constant *CPV) {
367 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
368 switch (CE->getOpcode()) {
369 case Instruction::Cast:
371 printType(Out, CPV->getType());
373 printConstant(CE->getOperand(0));
377 case Instruction::GetElementPtr:
379 printIndexingExpression(CE->getOperand(0), gep_type_begin(CPV),
383 case Instruction::Add:
384 case Instruction::Sub:
385 case Instruction::Mul:
386 case Instruction::Div:
387 case Instruction::Rem:
388 case Instruction::SetEQ:
389 case Instruction::SetNE:
390 case Instruction::SetLT:
391 case Instruction::SetLE:
392 case Instruction::SetGT:
393 case Instruction::SetGE:
394 case Instruction::Shl:
395 case Instruction::Shr:
397 printConstant(CE->getOperand(0));
398 switch (CE->getOpcode()) {
399 case Instruction::Add: Out << " + "; break;
400 case Instruction::Sub: Out << " - "; break;
401 case Instruction::Mul: Out << " * "; break;
402 case Instruction::Div: Out << " / "; break;
403 case Instruction::Rem: Out << " % "; break;
404 case Instruction::SetEQ: Out << " == "; break;
405 case Instruction::SetNE: Out << " != "; break;
406 case Instruction::SetLT: Out << " < "; break;
407 case Instruction::SetLE: Out << " <= "; break;
408 case Instruction::SetGT: Out << " > "; break;
409 case Instruction::SetGE: Out << " >= "; break;
410 case Instruction::Shl: Out << " << "; break;
411 case Instruction::Shr: Out << " >> "; break;
412 default: assert(0 && "Illegal opcode here!");
414 printConstant(CE->getOperand(1));
419 std::cerr << "CWriter Error: Unhandled constant expression: "
425 switch (CPV->getType()->getPrimitiveID()) {
427 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
428 case Type::SByteTyID:
429 case Type::ShortTyID:
430 Out << cast<ConstantSInt>(CPV)->getValue(); break;
432 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
433 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
435 Out << cast<ConstantSInt>(CPV)->getValue();
439 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
441 case Type::UByteTyID:
442 case Type::UShortTyID:
443 Out << cast<ConstantUInt>(CPV)->getValue(); break;
445 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
446 case Type::ULongTyID:
447 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
449 case Type::FloatTyID:
450 case Type::DoubleTyID: {
451 ConstantFP *FPC = cast<ConstantFP>(CPV);
452 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
453 if (I != FPConstantMap.end()) {
454 // Because of FP precision problems we must load from a stack allocated
455 // value that holds the value in hex.
456 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
457 << "*)&FPConstant" << I->second << ")";
460 // Print out the constant as a floating point number.
462 sprintf(Buffer, "%a", FPC->getValue());
463 Out << Buffer << " /*" << FPC->getValue() << "*/ ";
465 Out << ftostr(FPC->getValue());
471 case Type::ArrayTyID:
472 printConstantArray(cast<ConstantArray>(CPV));
475 case Type::StructTyID: {
477 if (CPV->getNumOperands()) {
479 printConstant(cast<Constant>(CPV->getOperand(0)));
480 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
482 printConstant(cast<Constant>(CPV->getOperand(i)));
489 case Type::PointerTyID:
490 if (isa<ConstantPointerNull>(CPV)) {
492 printType(Out, CPV->getType());
493 Out << ")/*NULL*/0)";
495 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
496 writeOperand(CPR->getValue());
501 std::cerr << "Unknown constant type: " << CPV << "\n";
506 void CWriter::writeOperandInternal(Value *Operand) {
507 if (Instruction *I = dyn_cast<Instruction>(Operand))
508 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
509 // Should we inline this instruction to build a tree?
516 if (Constant *CPV = dyn_cast<Constant>(Operand)) {
519 Out << Mang->getValueName(Operand);
523 void CWriter::writeOperand(Value *Operand) {
524 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
525 Out << "(&"; // Global variables are references as their addresses by llvm
527 writeOperandInternal(Operand);
529 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
533 // nameAllUsedStructureTypes - If there are structure types in the module that
534 // are used but do not have names assigned to them in the symbol table yet then
535 // we assign them names now.
537 bool CWriter::nameAllUsedStructureTypes(Module &M) {
538 // Get a set of types that are used by the program...
539 std::set<const Type *> UT = FUT->getTypes();
541 // Loop over the module symbol table, removing types from UT that are already
544 SymbolTable &MST = M.getSymbolTable();
545 if (MST.find(Type::TypeTy) != MST.end())
546 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
547 E = MST.type_end(Type::TypeTy); I != E; ++I)
548 UT.erase(cast<Type>(I->second));
550 // UT now contains types that are not named. Loop over it, naming structure
553 bool Changed = false;
554 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
556 if (const StructType *ST = dyn_cast<StructType>(*I)) {
557 ((Value*)ST)->setName("unnamed", &MST);
563 // generateCompilerSpecificCode - This is where we add conditional compilation
564 // directives to cater to specific compilers as need be.
566 static void generateCompilerSpecificCode(std::ostream& Out) {
567 // Alloca is hard to get, and we don't want to include stdlib.h here...
568 Out << "/* get a declaration for alloca */\n"
570 << "extern void *__builtin_alloca(unsigned long);\n"
571 << "#define alloca(x) __builtin_alloca(x)\n"
573 << "#ifndef __FreeBSD__\n"
574 << "#include <alloca.h>\n"
578 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
579 // If we aren't being compiled with GCC, just drop these attributes.
580 Out << "#ifndef __GNUC__ /* Can only support \"linkonce\" vars with GCC */\n"
581 << "#define __attribute__(X)\n"
585 // At some point, we should support "external weak" vs. "weak" linkages.
586 // On Mac OS X, "external weak" is spelled "__attribute__((weak_import))".
587 Out << "#if defined(__GNUC__) && defined(__APPLE_CC__)\n"
588 << "#define __EXTERNAL_WEAK__ __attribute__((weak_import))\n"
589 << "#elif defined(__GNUC__)\n"
590 << "#define __EXTERNAL_WEAK__ __attribute__((weak))\n"
592 << "#define __EXTERNAL_WEAK__\n"
596 // For now, turn off the weak linkage attribute on Mac OS X. (See above.)
597 Out << "#if defined(__GNUC__) && defined(__APPLE_CC__)\n"
598 << "#define __ATTRIBUTE_WEAK__\n"
599 << "#elif defined(__GNUC__)\n"
600 << "#define __ATTRIBUTE_WEAK__ __attribute__((weak))\n"
602 << "#define __ATTRIBUTE_WEAK__\n"
606 bool CWriter::doInitialization(Module &M) {
609 FUT = &getAnalysis<FindUsedTypes>();
611 // Ensure that all structure types have names...
612 bool Changed = nameAllUsedStructureTypes(M);
613 Mang = new Mangler(M);
615 // Calculate which global values have names that will collide when we throw
616 // away type information.
617 { // Scope to delete the FoundNames set when we are done with it...
618 std::set<std::string> FoundNames;
619 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
620 if (I->hasName()) // If the global has a name...
621 if (FoundNames.count(I->getName())) // And the name is already used
622 MangledGlobals.insert(I); // Mangle the name
624 FoundNames.insert(I->getName()); // Otherwise, keep track of name
626 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
627 if (I->hasName()) // If the global has a name...
628 if (FoundNames.count(I->getName())) // And the name is already used
629 MangledGlobals.insert(I); // Mangle the name
631 FoundNames.insert(I->getName()); // Otherwise, keep track of name
634 // get declaration for alloca
635 Out << "/* Provide Declarations */\n";
636 Out << "#include <stdarg.h>\n"; // Varargs support
637 Out << "#include <setjmp.h>\n"; // Unwind support
638 generateCompilerSpecificCode(Out);
640 // Provide a definition for `bool' if not compiling with a C++ compiler.
642 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
644 << "\n\n/* Support for floating point constants */\n"
645 << "typedef unsigned long long ConstantDoubleTy;\n"
646 << "typedef unsigned int ConstantFloatTy;\n"
648 << "\n\n/* Global Declarations */\n";
650 // First output all the declarations for the program, because C requires
651 // Functions & globals to be declared before they are used.
654 // Loop over the symbol table, emitting all named constants...
655 printSymbolTable(M.getSymbolTable());
657 // Global variable declarations...
659 Out << "\n/* External Global Variable Declarations */\n";
660 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I) {
661 if (I->hasExternalLinkage()) {
663 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
669 // Function declarations
671 Out << "\n/* Function Declarations */\n";
673 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
674 // If the function is external and the name collides don't print it.
675 // Sometimes the bytecode likes to have multiple "declarations" for
676 // external functions
677 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
678 !I->getIntrinsicID() &&
679 I->getName() != "setjmp" && I->getName() != "longjmp") {
680 printFunctionSignature(I, true);
681 if (I->hasWeakLinkage()) Out << " __ATTRIBUTE_WEAK__";
687 // Print Malloc prototype if needed
689 Out << "\n/* Malloc to make sun happy */\n";
690 Out << "extern void * malloc();\n\n";
693 // Output the global variable declarations
695 Out << "\n\n/* Global Variable Declarations */\n";
696 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
697 if (!I->isExternal()) {
699 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
701 if (I->hasLinkOnceLinkage())
702 Out << " __attribute__((common))";
703 else if (I->hasWeakLinkage())
704 Out << " __ATTRIBUTE_WEAK__";
709 // Output the global variable definitions and contents...
711 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
712 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
713 if (!I->isExternal()) {
714 if (I->hasInternalLinkage())
716 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
717 if (I->hasLinkOnceLinkage())
718 Out << " __attribute__((common))";
719 else if (I->hasWeakLinkage())
720 Out << " __ATTRIBUTE_WEAK__";
722 // If the initializer is not null, emit the initializer. If it is null,
723 // we try to avoid emitting large amounts of zeros. The problem with
724 // this, however, occurs when the variable has weak linkage. In this
725 // case, the assembler will complain about the variable being both weak
726 // and common, so we disable this optimization.
727 if (!I->getInitializer()->isNullValue() ||
728 I->hasWeakLinkage()) {
730 writeOperand(I->getInitializer());
736 // Output all floating point constants that cannot be printed accurately...
737 printFloatingPointConstants(M);
740 Out << "\n\n/* Function Bodies */\n";
745 /// Output all floating point constants that cannot be printed accurately...
746 void CWriter::printFloatingPointConstants(Module &M) {
749 unsigned long long U;
757 // Scan the module for floating point constants. If any FP constant is used
758 // in the function, we want to redirect it here so that we do not depend on
759 // the precision of the printed form, unless the printed form preserves
762 unsigned FPCounter = 0;
763 for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F)
764 for (constant_iterator I = constant_begin(F), E = constant_end(F);
766 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
767 if (!isFPCSafeToPrint(FPC) && // Do not put in FPConstantMap if safe.
768 !FPConstantMap.count(FPC)) {
769 double Val = FPC->getValue();
771 FPConstantMap[FPC] = FPCounter; // Number the FP constants
773 if (FPC->getType() == Type::DoubleTy) {
775 Out << "const ConstantDoubleTy FPConstant" << FPCounter++
776 << " = 0x" << std::hex << DBLUnion.U << std::dec
777 << "ULL; /* " << Val << " */\n";
778 } else if (FPC->getType() == Type::FloatTy) {
780 Out << "const ConstantFloatTy FPConstant" << FPCounter++
781 << " = 0x" << std::hex << FLTUnion.U << std::dec
782 << "U; /* " << Val << " */\n";
784 assert(0 && "Unknown float type!");
791 /// printSymbolTable - Run through symbol table looking for type names. If a
792 /// type name is found, emit it's declaration...
794 void CWriter::printSymbolTable(const SymbolTable &ST) {
795 // If there are no type names, exit early.
796 if (ST.find(Type::TypeTy) == ST.end())
799 // We are only interested in the type plane of the symbol table...
800 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
801 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
803 // Print out forward declarations for structure types before anything else!
804 Out << "/* Structure forward decls */\n";
805 for (; I != End; ++I)
806 if (const Type *STy = dyn_cast<StructType>(I->second))
807 // Only print out used types!
808 if (FUT->getTypes().count(STy)) {
809 std::string Name = "struct l_" + Mangler::makeNameProper(I->first);
810 Out << Name << ";\n";
811 TypeNames.insert(std::make_pair(STy, Name));
816 // Now we can print out typedefs...
817 Out << "/* Typedefs */\n";
818 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
819 // Only print out used types!
820 if (FUT->getTypes().count(cast<Type>(I->second))) {
821 const Type *Ty = cast<Type>(I->second);
822 std::string Name = "l_" + Mangler::makeNameProper(I->first);
824 printType(Out, Ty, Name);
830 // Keep track of which structures have been printed so far...
831 std::set<const StructType *> StructPrinted;
833 // Loop over all structures then push them into the stack so they are
834 // printed in the correct order.
836 Out << "/* Structure contents */\n";
837 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
838 if (const StructType *STy = dyn_cast<StructType>(I->second))
839 // Only print out used types!
840 if (FUT->getTypes().count(STy))
841 printContainedStructs(STy, StructPrinted);
844 // Push the struct onto the stack and recursively push all structs
845 // this one depends on.
846 void CWriter::printContainedStructs(const Type *Ty,
847 std::set<const StructType*> &StructPrinted){
848 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
849 //Check to see if we have already printed this struct
850 if (StructPrinted.count(STy) == 0) {
851 // Print all contained types first...
852 for (StructType::element_iterator I = STy->element_begin(),
853 E = STy->element_end(); I != E; ++I) {
854 const Type *Ty1 = I->get();
855 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
856 printContainedStructs(*I, StructPrinted);
859 //Print structure type out..
860 StructPrinted.insert(STy);
861 std::string Name = TypeNames[STy];
862 printType(Out, STy, Name, true);
866 // If it is an array, check contained types and continue
867 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
868 const Type *Ty1 = ATy->getElementType();
869 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
870 printContainedStructs(Ty1, StructPrinted);
875 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
876 // If the program provides its own malloc prototype we don't need
877 // to include the general one.
878 if (Mang->getValueName(F) == "malloc")
881 if (F->hasInternalLinkage()) Out << "static ";
882 if (F->hasLinkOnceLinkage()) Out << "inline ";
884 // Loop over the arguments, printing them...
885 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
887 std::stringstream FunctionInnards;
889 // Print out the name...
890 FunctionInnards << Mang->getValueName(F) << "(";
892 if (!F->isExternal()) {
895 if (F->abegin()->hasName() || !Prototype)
896 ArgName = Mang->getValueName(F->abegin());
897 printType(FunctionInnards, F->afront().getType(), ArgName);
898 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
900 FunctionInnards << ", ";
901 if (I->hasName() || !Prototype)
902 ArgName = Mang->getValueName(I);
905 printType(FunctionInnards, I->getType(), ArgName);
909 // Loop over the arguments, printing them...
910 for (FunctionType::param_iterator I = FT->param_begin(),
911 E = FT->param_end(); I != E; ++I) {
912 if (I != FT->param_begin()) FunctionInnards << ", ";
913 printType(FunctionInnards, *I);
917 // Finish printing arguments... if this is a vararg function, print the ...,
918 // unless there are no known types, in which case, we just emit ().
920 if (FT->isVarArg() && FT->getNumParams()) {
921 if (FT->getNumParams()) FunctionInnards << ", ";
922 FunctionInnards << "..."; // Output varargs portion of signature!
923 } else if (!FT->isVarArg() && FT->getNumParams() == 0) {
924 FunctionInnards << "void"; // ret() -> ret(void) in C.
926 FunctionInnards << ")";
927 // Print out the return type and the entire signature for that matter
928 printType(Out, F->getReturnType(), FunctionInnards.str());
931 void CWriter::printFunction(Function &F) {
932 printFunctionSignature(&F, false);
935 // print local variable information for the function
936 for (inst_iterator I = inst_begin(&F), E = inst_end(&F); I != E; ++I)
937 if (const AllocaInst *AI = isDirectAlloca(*I)) {
939 printType(Out, AI->getAllocatedType(), Mang->getValueName(AI));
940 Out << "; /* Address exposed local */\n";
941 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
943 printType(Out, (*I)->getType(), Mang->getValueName(*I));
946 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
948 printType(Out, (*I)->getType(),
949 Mang->getValueName(*I)+"__PHI_TEMPORARY");
956 // print the basic blocks
957 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
958 BasicBlock *Prev = BB->getPrev();
960 // Don't print the label for the basic block if there are no uses, or if the
961 // only terminator use is the predecessor basic block's terminator. We have
962 // to scan the use list because PHI nodes use basic blocks too but do not
963 // require a label to be generated.
965 bool NeedsLabel = false;
966 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
968 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
969 if (TI != Prev->getTerminator() ||
970 isa<SwitchInst>(Prev->getTerminator()) ||
971 isa<InvokeInst>(Prev->getTerminator())) {
976 if (NeedsLabel) Out << Mang->getValueName(BB) << ":\n";
978 // Output all of the instructions in the basic block...
979 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
980 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
981 if (II->getType() != Type::VoidTy)
990 // Don't emit prefix or suffix for the terminator...
991 visit(*BB->getTerminator());
997 // Specific Instruction type classes... note that all of the casts are
998 // necessary because we use the instruction classes as opaque types...
1000 void CWriter::visitReturnInst(ReturnInst &I) {
1001 // Don't output a void return if this is the last basic block in the function
1002 if (I.getNumOperands() == 0 &&
1003 &*--I.getParent()->getParent()->end() == I.getParent() &&
1004 !I.getParent()->size() == 1) {
1009 if (I.getNumOperands()) {
1011 writeOperand(I.getOperand(0));
1016 void CWriter::visitSwitchInst(SwitchInst &SI) {
1018 writeOperand(SI.getOperand(0));
1019 Out << ") {\n default:\n";
1020 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
1022 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
1024 writeOperand(SI.getOperand(i));
1026 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
1027 printBranchToBlock(SI.getParent(), Succ, 2);
1028 if (Succ == SI.getParent()->getNext())
1034 void CWriter::visitInvokeInst(InvokeInst &II) {
1035 assert(0 && "Lowerinvoke pass didn't work!");
1039 void CWriter::visitUnwindInst(UnwindInst &I) {
1040 assert(0 && "Lowerinvoke pass didn't work!");
1043 bool isGotoCodeNecessary(BasicBlock *From, BasicBlock *To) {
1044 // If PHI nodes need copies, we need the copy code...
1045 if (isa<PHINode>(To->front()) ||
1046 From->getNext() != To) // Not directly successor, need goto
1049 // Otherwise we don't need the code.
1053 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
1055 for (BasicBlock::iterator I = Succ->begin();
1056 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1057 // now we have to do the printing
1058 Out << std::string(Indent, ' ');
1059 Out << " " << Mang->getValueName(I) << "__PHI_TEMPORARY = ";
1060 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1061 Out << "; /* for PHI node */\n";
1064 if (CurBB->getNext() != Succ ||
1065 isa<InvokeInst>(CurBB->getTerminator()) ||
1066 isa<SwitchInst>(CurBB->getTerminator())) {
1067 Out << std::string(Indent, ' ') << " goto ";
1073 // Branch instruction printing - Avoid printing out a branch to a basic block
1074 // that immediately succeeds the current one.
1076 void CWriter::visitBranchInst(BranchInst &I) {
1077 if (I.isConditional()) {
1078 if (isGotoCodeNecessary(I.getParent(), I.getSuccessor(0))) {
1080 writeOperand(I.getCondition());
1083 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1085 if (isGotoCodeNecessary(I.getParent(), I.getSuccessor(1))) {
1086 Out << " } else {\n";
1087 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1090 // First goto not necessary, assume second one is...
1092 writeOperand(I.getCondition());
1095 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1100 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1105 // PHI nodes get copied into temporary values at the end of predecessor basic
1106 // blocks. We now need to copy these temporary values into the REAL value for
1108 void CWriter::visitPHINode(PHINode &I) {
1110 Out << "__PHI_TEMPORARY";
1114 void CWriter::visitBinaryOperator(Instruction &I) {
1115 // binary instructions, shift instructions, setCond instructions.
1116 assert(!isa<PointerType>(I.getType()));
1118 // We must cast the results of binary operations which might be promoted.
1119 bool needsCast = false;
1120 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1121 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1122 || (I.getType() == Type::FloatTy)) {
1125 printType(Out, I.getType());
1129 writeOperand(I.getOperand(0));
1131 switch (I.getOpcode()) {
1132 case Instruction::Add: Out << " + "; break;
1133 case Instruction::Sub: Out << " - "; break;
1134 case Instruction::Mul: Out << "*"; break;
1135 case Instruction::Div: Out << "/"; break;
1136 case Instruction::Rem: Out << "%"; break;
1137 case Instruction::And: Out << " & "; break;
1138 case Instruction::Or: Out << " | "; break;
1139 case Instruction::Xor: Out << " ^ "; break;
1140 case Instruction::SetEQ: Out << " == "; break;
1141 case Instruction::SetNE: Out << " != "; break;
1142 case Instruction::SetLE: Out << " <= "; break;
1143 case Instruction::SetGE: Out << " >= "; break;
1144 case Instruction::SetLT: Out << " < "; break;
1145 case Instruction::SetGT: Out << " > "; break;
1146 case Instruction::Shl : Out << " << "; break;
1147 case Instruction::Shr : Out << " >> "; break;
1148 default: std::cerr << "Invalid operator type!" << I; abort();
1151 writeOperand(I.getOperand(1));
1158 void CWriter::visitCastInst(CastInst &I) {
1159 if (I.getType() == Type::BoolTy) {
1161 writeOperand(I.getOperand(0));
1166 printType(Out, I.getType());
1168 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1169 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1170 // Avoid "cast to pointer from integer of different size" warnings
1174 writeOperand(I.getOperand(0));
1177 void CWriter::visitCallInst(CallInst &I) {
1178 // Handle intrinsic function calls first...
1179 if (Function *F = I.getCalledFunction())
1180 if (Intrinsic::ID ID = (Intrinsic::ID)F->getIntrinsicID()) {
1182 default: assert(0 && "Unknown LLVM intrinsic!");
1183 case Intrinsic::va_start:
1186 Out << "va_start(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1187 // Output the last argument to the enclosing function...
1188 if (I.getParent()->getParent()->aempty()) {
1189 std::cerr << "The C backend does not currently support zero "
1190 << "argument varargs functions, such as '"
1191 << I.getParent()->getParent()->getName() << "'!\n";
1194 writeOperand(&I.getParent()->getParent()->aback());
1197 case Intrinsic::va_end:
1198 Out << "va_end(*(va_list*)&";
1199 writeOperand(I.getOperand(1));
1202 case Intrinsic::va_copy:
1204 Out << "va_copy(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1205 Out << "*(va_list*)&";
1206 writeOperand(I.getOperand(1));
1209 case Intrinsic::setjmp:
1210 case Intrinsic::sigsetjmp:
1211 // This intrinsic should never exist in the program, but until we get
1212 // setjmp/longjmp transformations going on, we should codegen it to
1213 // something reasonable. This will allow code that never calls longjmp
1217 case Intrinsic::longjmp:
1218 case Intrinsic::siglongjmp:
1219 // Longjmp is not implemented, and never will be. It would cause an
1223 case Intrinsic::memcpy:
1225 writeOperand(I.getOperand(1));
1227 writeOperand(I.getOperand(2));
1229 writeOperand(I.getOperand(3));
1232 case Intrinsic::memmove:
1234 writeOperand(I.getOperand(1));
1236 writeOperand(I.getOperand(2));
1238 writeOperand(I.getOperand(3));
1246 void CWriter::visitCallSite(CallSite CS) {
1247 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1248 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1249 const Type *RetTy = FTy->getReturnType();
1251 writeOperand(CS.getCalledValue());
1254 if (CS.arg_begin() != CS.arg_end()) {
1255 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1258 for (++AI; AI != AE; ++AI) {
1266 void CWriter::visitMallocInst(MallocInst &I) {
1268 printType(Out, I.getType());
1269 Out << ")malloc(sizeof(";
1270 printType(Out, I.getType()->getElementType());
1273 if (I.isArrayAllocation()) {
1275 writeOperand(I.getOperand(0));
1280 void CWriter::visitAllocaInst(AllocaInst &I) {
1282 printType(Out, I.getType());
1283 Out << ") alloca(sizeof(";
1284 printType(Out, I.getType()->getElementType());
1286 if (I.isArrayAllocation()) {
1288 writeOperand(I.getOperand(0));
1293 void CWriter::visitFreeInst(FreeInst &I) {
1294 Out << "free((char*)";
1295 writeOperand(I.getOperand(0));
1299 void CWriter::printIndexingExpression(Value *Ptr, gep_type_iterator I,
1300 gep_type_iterator E) {
1301 bool HasImplicitAddress = false;
1302 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1303 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1304 HasImplicitAddress = true;
1305 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1306 HasImplicitAddress = true;
1307 Ptr = CPR->getValue(); // Get to the global...
1308 } else if (isDirectAlloca(Ptr)) {
1309 HasImplicitAddress = true;
1313 if (!HasImplicitAddress)
1314 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1316 writeOperandInternal(Ptr);
1320 const Constant *CI = dyn_cast<Constant>(I.getOperand());
1321 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1324 writeOperandInternal(Ptr);
1326 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1328 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1331 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1332 "Can only have implicit address with direct accessing");
1334 if (HasImplicitAddress) {
1336 } else if (CI && CI->isNullValue()) {
1337 gep_type_iterator TmpI = I; ++TmpI;
1339 // Print out the -> operator if possible...
1340 if (TmpI != E && isa<StructType>(*TmpI)) {
1341 Out << (HasImplicitAddress ? "." : "->");
1342 Out << "field" << cast<ConstantUInt>(TmpI.getOperand())->getValue();
1348 if (isa<StructType>(*I)) {
1349 Out << ".field" << cast<ConstantUInt>(I.getOperand())->getValue();
1352 writeOperand(I.getOperand());
1357 void CWriter::visitLoadInst(LoadInst &I) {
1359 writeOperand(I.getOperand(0));
1362 void CWriter::visitStoreInst(StoreInst &I) {
1364 writeOperand(I.getPointerOperand());
1366 writeOperand(I.getOperand(0));
1369 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1371 printIndexingExpression(I.getPointerOperand(), gep_type_begin(I),
1375 void CWriter::visitVANextInst(VANextInst &I) {
1376 Out << Mang->getValueName(I.getOperand(0));
1377 Out << "; va_arg(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1378 printType(Out, I.getArgType());
1382 void CWriter::visitVAArgInst(VAArgInst &I) {
1384 Out << "{ va_list Tmp; va_copy(Tmp, *(va_list*)&";
1385 writeOperand(I.getOperand(0));
1386 Out << ");\n " << Mang->getValueName(&I) << " = va_arg(Tmp, ";
1387 printType(Out, I.getType());
1388 Out << ");\n va_end(Tmp); }";
1393 //===----------------------------------------------------------------------===//
1394 // External Interface declaration
1395 //===----------------------------------------------------------------------===//
1397 void llvm::AddPassesToWriteC(PassManager &PM, std::ostream &o) {
1398 PM.add(createLowerInvokePass());
1399 //PM.add(createLowerAllocationsPass());
1400 PM.add(new CWriter(o));