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.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/Assembly/CWriter.h"
15 #include "llvm/Constants.h"
16 #include "llvm/DerivedTypes.h"
17 #include "llvm/Module.h"
18 #include "llvm/Instructions.h"
19 #include "llvm/Pass.h"
20 #include "llvm/SymbolTable.h"
21 #include "llvm/Intrinsics.h"
22 #include "llvm/Analysis/FindUsedTypes.h"
23 #include "llvm/Analysis/ConstantsScanner.h"
24 #include "llvm/Support/InstVisitor.h"
25 #include "llvm/Support/InstIterator.h"
26 #include "llvm/Support/CallSite.h"
27 #include "llvm/Support/Mangler.h"
28 #include "Support/StringExtras.h"
29 #include "Support/STLExtras.h"
30 #include "Config/config.h"
35 class CWriter : public Pass, public InstVisitor<CWriter> {
38 const Module *TheModule;
41 std::map<const Type *, std::string> TypeNames;
42 std::set<const Value*> MangledGlobals;
43 bool needsMalloc, emittedInvoke;
45 std::map<const ConstantFP *, unsigned> FPConstantMap;
47 CWriter(std::ostream &o) : Out(o) {}
49 void getAnalysisUsage(AnalysisUsage &AU) const {
51 AU.addRequired<FindUsedTypes>();
54 virtual bool run(Module &M) {
57 FUT = &getAnalysis<FindUsedTypes>();
59 // Ensure that all structure types have names...
60 bool Changed = nameAllUsedStructureTypes(M);
61 Mang = new Mangler(M);
69 MangledGlobals.clear();
73 std::ostream &printType(std::ostream &Out, const Type *Ty,
74 const std::string &VariableName = "",
75 bool IgnoreName = false);
77 void writeOperand(Value *Operand);
78 void writeOperandInternal(Value *Operand);
81 bool nameAllUsedStructureTypes(Module &M);
82 void printModule(Module *M);
83 void printFloatingPointConstants(Module &M);
84 void printSymbolTable(const SymbolTable &ST);
85 void printContainedStructs(const Type *Ty, std::set<const StructType *> &);
86 void printFunctionSignature(const Function *F, bool Prototype);
88 void printFunction(Function *);
90 void printConstant(Constant *CPV);
91 void printConstantArray(ConstantArray *CPA);
93 // isInlinableInst - Attempt to inline instructions into their uses to build
94 // trees as much as possible. To do this, we have to consistently decide
95 // what is acceptable to inline, so that variable declarations don't get
96 // printed and an extra copy of the expr is not emitted.
98 static bool isInlinableInst(const Instruction &I) {
99 // Must be an expression, must be used exactly once. If it is dead, we
100 // emit it inline where it would go.
101 if (I.getType() == Type::VoidTy || !I.hasOneUse() ||
102 isa<TerminatorInst>(I) || isa<CallInst>(I) || isa<PHINode>(I) ||
103 isa<LoadInst>(I) || isa<VAArgInst>(I) || isa<VANextInst>(I))
104 // Don't inline a load across a store or other bad things!
107 // Only inline instruction it it's use is in the same BB as the inst.
108 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
111 // isDirectAlloca - Define fixed sized allocas in the entry block as direct
112 // variables which are accessed with the & operator. This causes GCC to
113 // generate significantly better code than to emit alloca calls directly.
115 static const AllocaInst *isDirectAlloca(const Value *V) {
116 const AllocaInst *AI = dyn_cast<AllocaInst>(V);
117 if (!AI) return false;
118 if (AI->isArrayAllocation())
119 return 0; // FIXME: we can also inline fixed size array allocas!
120 if (AI->getParent() != &AI->getParent()->getParent()->getEntryBlock())
125 // Instruction visitation functions
126 friend class InstVisitor<CWriter>;
128 void visitReturnInst(ReturnInst &I);
129 void visitBranchInst(BranchInst &I);
130 void visitSwitchInst(SwitchInst &I);
131 void visitInvokeInst(InvokeInst &I);
132 void visitUnwindInst(UnwindInst &I);
134 void visitPHINode(PHINode &I);
135 void visitBinaryOperator(Instruction &I);
137 void visitCastInst (CastInst &I);
138 void visitCallInst (CallInst &I);
139 void visitCallSite (CallSite CS);
140 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
142 void visitMallocInst(MallocInst &I);
143 void visitAllocaInst(AllocaInst &I);
144 void visitFreeInst (FreeInst &I);
145 void visitLoadInst (LoadInst &I);
146 void visitStoreInst (StoreInst &I);
147 void visitGetElementPtrInst(GetElementPtrInst &I);
148 void visitVANextInst(VANextInst &I);
149 void visitVAArgInst (VAArgInst &I);
151 void visitInstruction(Instruction &I) {
152 std::cerr << "C Writer does not know about " << I;
156 void outputLValue(Instruction *I) {
157 Out << " " << Mang->getValueName(I) << " = ";
159 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
161 void printIndexingExpression(Value *Ptr, User::op_iterator I,
162 User::op_iterator E);
166 // Pass the Type* and the variable name and this prints out the variable
169 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
170 const std::string &NameSoFar,
172 if (Ty->isPrimitiveType())
173 switch (Ty->getPrimitiveID()) {
174 case Type::VoidTyID: return Out << "void " << NameSoFar;
175 case Type::BoolTyID: return Out << "bool " << NameSoFar;
176 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
177 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
178 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
179 case Type::ShortTyID: return Out << "short " << NameSoFar;
180 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
181 case Type::IntTyID: return Out << "int " << NameSoFar;
182 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
183 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
184 case Type::FloatTyID: return Out << "float " << NameSoFar;
185 case Type::DoubleTyID: return Out << "double " << NameSoFar;
187 std::cerr << "Unknown primitive type: " << Ty << "\n";
191 // Check to see if the type is named.
192 if (!IgnoreName || isa<OpaqueType>(Ty)) {
193 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
194 if (I != TypeNames.end()) return Out << I->second << " " << NameSoFar;
197 switch (Ty->getPrimitiveID()) {
198 case Type::FunctionTyID: {
199 const FunctionType *MTy = cast<FunctionType>(Ty);
200 std::stringstream FunctionInnards;
201 FunctionInnards << " (" << NameSoFar << ") (";
202 for (FunctionType::ParamTypes::const_iterator
203 I = MTy->getParamTypes().begin(),
204 E = MTy->getParamTypes().end(); I != E; ++I) {
205 if (I != MTy->getParamTypes().begin())
206 FunctionInnards << ", ";
207 printType(FunctionInnards, *I, "");
209 if (MTy->isVarArg()) {
210 if (!MTy->getParamTypes().empty())
211 FunctionInnards << ", ...";
212 } else if (MTy->getParamTypes().empty()) {
213 FunctionInnards << "void";
215 FunctionInnards << ")";
216 std::string tstr = FunctionInnards.str();
217 printType(Out, MTy->getReturnType(), tstr);
220 case Type::StructTyID: {
221 const StructType *STy = cast<StructType>(Ty);
222 Out << NameSoFar + " {\n";
224 for (StructType::ElementTypes::const_iterator
225 I = STy->getElementTypes().begin(),
226 E = STy->getElementTypes().end(); I != E; ++I) {
228 printType(Out, *I, "field" + utostr(Idx++));
234 case Type::PointerTyID: {
235 const PointerType *PTy = cast<PointerType>(Ty);
236 std::string ptrName = "*" + NameSoFar;
238 if (isa<ArrayType>(PTy->getElementType()))
239 ptrName = "(" + ptrName + ")";
241 return printType(Out, PTy->getElementType(), ptrName);
244 case Type::ArrayTyID: {
245 const ArrayType *ATy = cast<ArrayType>(Ty);
246 unsigned NumElements = ATy->getNumElements();
247 return printType(Out, ATy->getElementType(),
248 NameSoFar + "[" + utostr(NumElements) + "]");
251 case Type::OpaqueTyID: {
252 static int Count = 0;
253 std::string TyName = "struct opaque_" + itostr(Count++);
254 assert(TypeNames.find(Ty) == TypeNames.end());
255 TypeNames[Ty] = TyName;
256 return Out << TyName << " " << NameSoFar;
259 assert(0 && "Unhandled case in getTypeProps!");
266 void CWriter::printConstantArray(ConstantArray *CPA) {
268 // As a special case, print the array as a string if it is an array of
269 // ubytes or an array of sbytes with positive values.
271 const Type *ETy = CPA->getType()->getElementType();
272 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
274 // Make sure the last character is a null char, as automatically added by C
275 if (isString && (CPA->getNumOperands() == 0 ||
276 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
281 // Keep track of whether the last number was a hexadecimal escape
282 bool LastWasHex = false;
284 // Do not include the last character, which we know is null
285 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
286 unsigned char C = cast<ConstantInt>(CPA->getOperand(i))->getRawValue();
288 // Print it out literally if it is a printable character. The only thing
289 // to be careful about is when the last letter output was a hex escape
290 // code, in which case we have to be careful not to print out hex digits
291 // explicitly (the C compiler thinks it is a continuation of the previous
292 // character, sheesh...)
294 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
296 if (C == '"' || C == '\\')
303 case '\n': Out << "\\n"; break;
304 case '\t': Out << "\\t"; break;
305 case '\r': Out << "\\r"; break;
306 case '\v': Out << "\\v"; break;
307 case '\a': Out << "\\a"; break;
308 case '\"': Out << "\\\""; break;
309 case '\'': Out << "\\\'"; break;
312 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
313 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
322 if (CPA->getNumOperands()) {
324 printConstant(cast<Constant>(CPA->getOperand(0)));
325 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
327 printConstant(cast<Constant>(CPA->getOperand(i)));
334 // isFPCSafeToPrint - Returns true if we may assume that CFP may be written out
335 // textually as a double (rather than as a reference to a stack-allocated
336 // variable). We decide this by converting CFP to a string and back into a
337 // double, and then checking whether the conversion results in a bit-equal
338 // double to the original value of CFP. This depends on us and the target C
339 // compiler agreeing on the conversion process (which is pretty likely since we
340 // only deal in IEEE FP).
342 static bool isFPCSafeToPrint(const ConstantFP *CFP) {
345 sprintf(Buffer, "%a", CFP->getValue());
347 if (!strncmp(Buffer, "0x", 2) ||
348 !strncmp(Buffer, "-0x", 3) ||
349 !strncmp(Buffer, "+0x", 3))
350 return atof(Buffer) == CFP->getValue();
353 std::string StrVal = ftostr(CFP->getValue());
355 while (StrVal[0] == ' ')
356 StrVal.erase(StrVal.begin());
358 // Check to make sure that the stringized number is not some string like "Inf"
359 // or NaN. Check that the string matches the "[-+]?[0-9]" regex.
360 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
361 ((StrVal[0] == '-' || StrVal[0] == '+') &&
362 (StrVal[1] >= '0' && StrVal[1] <= '9')))
363 // Reparse stringized version!
364 return atof(StrVal.c_str()) == CFP->getValue();
369 // printConstant - The LLVM Constant to C Constant converter.
370 void CWriter::printConstant(Constant *CPV) {
371 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
372 switch (CE->getOpcode()) {
373 case Instruction::Cast:
375 printType(Out, CPV->getType());
377 printConstant(CE->getOperand(0));
381 case Instruction::GetElementPtr:
383 printIndexingExpression(CE->getOperand(0),
384 CPV->op_begin()+1, CPV->op_end());
387 case Instruction::Add:
388 case Instruction::Sub:
389 case Instruction::Mul:
390 case Instruction::Div:
391 case Instruction::Rem:
392 case Instruction::SetEQ:
393 case Instruction::SetNE:
394 case Instruction::SetLT:
395 case Instruction::SetLE:
396 case Instruction::SetGT:
397 case Instruction::SetGE:
399 printConstant(CE->getOperand(0));
400 switch (CE->getOpcode()) {
401 case Instruction::Add: Out << " + "; break;
402 case Instruction::Sub: Out << " - "; break;
403 case Instruction::Mul: Out << " * "; break;
404 case Instruction::Div: Out << " / "; break;
405 case Instruction::Rem: Out << " % "; break;
406 case Instruction::SetEQ: Out << " == "; break;
407 case Instruction::SetNE: Out << " != "; break;
408 case Instruction::SetLT: Out << " < "; break;
409 case Instruction::SetLE: Out << " <= "; break;
410 case Instruction::SetGT: Out << " > "; break;
411 case Instruction::SetGE: 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 void CWriter::printModule(Module *M) {
586 // Calculate which global values have names that will collide when we throw
587 // away type information.
588 { // Scope to delete the FoundNames set when we are done with it...
589 std::set<std::string> FoundNames;
590 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
591 if (I->hasName()) // If the global has a name...
592 if (FoundNames.count(I->getName())) // And the name is already used
593 MangledGlobals.insert(I); // Mangle the name
595 FoundNames.insert(I->getName()); // Otherwise, keep track of name
597 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
598 if (I->hasName()) // If the global has a name...
599 if (FoundNames.count(I->getName())) // And the name is already used
600 MangledGlobals.insert(I); // Mangle the name
602 FoundNames.insert(I->getName()); // Otherwise, keep track of name
605 // get declaration for alloca
606 Out << "/* Provide Declarations */\n";
607 Out << "#include <stdarg.h>\n";
608 Out << "#include <setjmp.h>\n";
609 generateCompilerSpecificCode(Out);
611 // Provide a definition for `bool' if not compiling with a C++ compiler.
613 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
615 << "\n\n/* Support for floating point constants */\n"
616 << "typedef unsigned long long ConstantDoubleTy;\n"
617 << "typedef unsigned int ConstantFloatTy;\n"
619 << "\n\n/* Support for the invoke instruction */\n"
620 << "extern struct __llvm_jmpbuf_list_t {\n"
621 << " jmp_buf buf; struct __llvm_jmpbuf_list_t *next;\n"
622 << "} *__llvm_jmpbuf_list;\n"
624 << "\n\n/* Global Declarations */\n";
626 // First output all the declarations for the program, because C requires
627 // Functions & globals to be declared before they are used.
630 // Loop over the symbol table, emitting all named constants...
631 printSymbolTable(M->getSymbolTable());
633 // Global variable declarations...
635 Out << "\n/* External Global Variable Declarations */\n";
636 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
637 if (I->hasExternalLinkage()) {
639 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
645 // Function declarations
647 Out << "\n/* Function Declarations */\n";
649 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
650 // If the function is external and the name collides don't print it.
651 // Sometimes the bytecode likes to have multiple "declarations" for
652 // external functions
653 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
654 !I->getIntrinsicID()) {
655 printFunctionSignature(I, true);
656 if (I->hasWeakLinkage()) Out << " __attribute__((weak))";
662 // Print Malloc prototype if needed
664 Out << "\n/* Malloc to make sun happy */\n";
665 Out << "extern void * malloc();\n\n";
668 // Output the global variable declarations
670 Out << "\n\n/* Global Variable Declarations */\n";
671 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
672 if (!I->isExternal()) {
674 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
676 if (I->hasLinkOnceLinkage())
677 Out << " __attribute__((common))";
678 else if (I->hasWeakLinkage())
679 Out << " __attribute__((weak))";
684 // Output the global variable definitions and contents...
686 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
687 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
688 if (!I->isExternal()) {
689 if (I->hasInternalLinkage())
691 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
692 if (I->hasLinkOnceLinkage())
693 Out << " __attribute__((common))";
694 else if (I->hasWeakLinkage())
695 Out << " __attribute__((weak))";
696 if (!I->getInitializer()->isNullValue()) {
698 writeOperand(I->getInitializer());
704 // Output all floating point constants that cannot be printed accurately...
705 printFloatingPointConstants(*M);
707 // Output all of the functions...
708 emittedInvoke = false;
710 Out << "\n\n/* Function Bodies */\n";
711 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
715 // If the program included an invoke instruction, we need to output the
716 // support code for it here!
718 Out << "\n/* More support for the invoke instruction */\n"
719 << "struct __llvm_jmpbuf_list_t *__llvm_jmpbuf_list "
720 << "__attribute__((common)) = 0;\n";
723 // Done with global FP constants
724 FPConstantMap.clear();
727 /// Output all floating point constants that cannot be printed accurately...
728 void CWriter::printFloatingPointConstants(Module &M) {
731 unsigned long long U;
739 // Scan the module for floating point constants. If any FP constant is used
740 // in the function, we want to redirect it here so that we do not depend on
741 // the precision of the printed form, unless the printed form preserves
744 unsigned FPCounter = 0;
745 for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F)
746 for (constant_iterator I = constant_begin(F), E = constant_end(F);
748 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
749 if (!isFPCSafeToPrint(FPC) && // Do not put in FPConstantMap if safe.
750 !FPConstantMap.count(FPC)) {
751 double Val = FPC->getValue();
753 FPConstantMap[FPC] = FPCounter; // Number the FP constants
755 if (FPC->getType() == Type::DoubleTy) {
757 Out << "const ConstantDoubleTy FPConstant" << FPCounter++
758 << " = 0x" << std::hex << DBLUnion.U << std::dec
759 << "ULL; /* " << Val << " */\n";
760 } else if (FPC->getType() == Type::FloatTy) {
762 Out << "const ConstantFloatTy FPConstant" << FPCounter++
763 << " = 0x" << std::hex << FLTUnion.U << std::dec
764 << "U; /* " << Val << " */\n";
766 assert(0 && "Unknown float type!");
773 /// printSymbolTable - Run through symbol table looking for type names. If a
774 /// type name is found, emit it's declaration...
776 void CWriter::printSymbolTable(const SymbolTable &ST) {
777 // If there are no type names, exit early.
778 if (ST.find(Type::TypeTy) == ST.end())
781 // We are only interested in the type plane of the symbol table...
782 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
783 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
785 // Print out forward declarations for structure types before anything else!
786 Out << "/* Structure forward decls */\n";
787 for (; I != End; ++I)
788 if (const Type *STy = dyn_cast<StructType>(I->second))
789 // Only print out used types!
790 if (FUT->getTypes().count(STy)) {
791 std::string Name = "struct l_" + Mangler::makeNameProper(I->first);
792 Out << Name << ";\n";
793 TypeNames.insert(std::make_pair(STy, Name));
798 // Now we can print out typedefs...
799 Out << "/* Typedefs */\n";
800 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
801 // Only print out used types!
802 if (FUT->getTypes().count(cast<Type>(I->second))) {
803 const Type *Ty = cast<Type>(I->second);
804 std::string Name = "l_" + Mangler::makeNameProper(I->first);
806 printType(Out, Ty, Name);
812 // Keep track of which structures have been printed so far...
813 std::set<const StructType *> StructPrinted;
815 // Loop over all structures then push them into the stack so they are
816 // printed in the correct order.
818 Out << "/* Structure contents */\n";
819 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
820 if (const StructType *STy = dyn_cast<StructType>(I->second))
821 // Only print out used types!
822 if (FUT->getTypes().count(STy))
823 printContainedStructs(STy, StructPrinted);
826 // Push the struct onto the stack and recursively push all structs
827 // this one depends on.
828 void CWriter::printContainedStructs(const Type *Ty,
829 std::set<const StructType*> &StructPrinted){
830 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
831 //Check to see if we have already printed this struct
832 if (StructPrinted.count(STy) == 0) {
833 // Print all contained types first...
834 for (StructType::ElementTypes::const_iterator
835 I = STy->getElementTypes().begin(),
836 E = STy->getElementTypes().end(); I != E; ++I) {
837 const Type *Ty1 = I->get();
838 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
839 printContainedStructs(*I, StructPrinted);
842 //Print structure type out..
843 StructPrinted.insert(STy);
844 std::string Name = TypeNames[STy];
845 printType(Out, STy, Name, true);
849 // If it is an array, check contained types and continue
850 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
851 const Type *Ty1 = ATy->getElementType();
852 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
853 printContainedStructs(Ty1, StructPrinted);
858 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
859 // If the program provides its own malloc prototype we don't need
860 // to include the general one.
861 if (Mang->getValueName(F) == "malloc")
864 if (F->hasInternalLinkage()) Out << "static ";
865 if (F->hasLinkOnceLinkage()) Out << "inline ";
867 // Loop over the arguments, printing them...
868 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
870 std::stringstream FunctionInnards;
872 // Print out the name...
873 FunctionInnards << Mang->getValueName(F) << "(";
875 if (!F->isExternal()) {
878 if (F->abegin()->hasName() || !Prototype)
879 ArgName = Mang->getValueName(F->abegin());
880 printType(FunctionInnards, F->afront().getType(), ArgName);
881 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
883 FunctionInnards << ", ";
884 if (I->hasName() || !Prototype)
885 ArgName = Mang->getValueName(I);
888 printType(FunctionInnards, I->getType(), ArgName);
892 // Loop over the arguments, printing them...
893 for (FunctionType::ParamTypes::const_iterator I =
894 FT->getParamTypes().begin(),
895 E = FT->getParamTypes().end(); I != E; ++I) {
896 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
897 printType(FunctionInnards, *I);
901 // Finish printing arguments... if this is a vararg function, print the ...,
902 // unless there are no known types, in which case, we just emit ().
904 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
905 if (FT->getParamTypes().size()) FunctionInnards << ", ";
906 FunctionInnards << "..."; // Output varargs portion of signature!
908 FunctionInnards << ")";
909 // Print out the return type and the entire signature for that matter
910 printType(Out, F->getReturnType(), FunctionInnards.str());
913 void CWriter::printFunction(Function *F) {
914 if (F->isExternal()) return;
916 printFunctionSignature(F, false);
919 // print local variable information for the function
920 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
921 if (const AllocaInst *AI = isDirectAlloca(*I)) {
923 printType(Out, AI->getAllocatedType(), Mang->getValueName(AI));
924 Out << "; /* Address exposed local */\n";
925 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
927 printType(Out, (*I)->getType(), Mang->getValueName(*I));
930 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
932 printType(Out, (*I)->getType(),
933 Mang->getValueName(*I)+"__PHI_TEMPORARY");
940 // print the basic blocks
941 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
942 BasicBlock *Prev = BB->getPrev();
944 // Don't print the label for the basic block if there are no uses, or if the
945 // only terminator use is the predecessor basic block's terminator. We have
946 // to scan the use list because PHI nodes use basic blocks too but do not
947 // require a label to be generated.
949 bool NeedsLabel = false;
950 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
952 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
953 if (TI != Prev->getTerminator() ||
954 isa<SwitchInst>(Prev->getTerminator()) ||
955 isa<InvokeInst>(Prev->getTerminator())) {
960 if (NeedsLabel) Out << Mang->getValueName(BB) << ":\n";
962 // Output all of the instructions in the basic block...
963 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
964 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
965 if (II->getType() != Type::VoidTy)
974 // Don't emit prefix or suffix for the terminator...
975 visit(*BB->getTerminator());
981 // Specific Instruction type classes... note that all of the casts are
982 // necessary because we use the instruction classes as opaque types...
984 void CWriter::visitReturnInst(ReturnInst &I) {
985 // Don't output a void return if this is the last basic block in the function
986 if (I.getNumOperands() == 0 &&
987 &*--I.getParent()->getParent()->end() == I.getParent() &&
988 !I.getParent()->size() == 1) {
993 if (I.getNumOperands()) {
995 writeOperand(I.getOperand(0));
1000 void CWriter::visitSwitchInst(SwitchInst &SI) {
1002 writeOperand(SI.getOperand(0));
1003 Out << ") {\n default:\n";
1004 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
1006 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
1008 writeOperand(SI.getOperand(i));
1010 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
1011 printBranchToBlock(SI.getParent(), Succ, 2);
1012 if (Succ == SI.getParent()->getNext())
1018 void CWriter::visitInvokeInst(InvokeInst &II) {
1020 << " struct __llvm_jmpbuf_list_t Entry;\n"
1021 << " Entry.next = __llvm_jmpbuf_list;\n"
1022 << " if (setjmp(Entry.buf)) {\n"
1023 << " __llvm_jmpbuf_list = Entry.next;\n";
1024 printBranchToBlock(II.getParent(), II.getExceptionalDest(), 4);
1026 << " __llvm_jmpbuf_list = &Entry;\n"
1029 if (II.getType() != Type::VoidTy) outputLValue(&II);
1032 << " __llvm_jmpbuf_list = Entry.next;\n"
1034 printBranchToBlock(II.getParent(), II.getNormalDest(), 0);
1035 emittedInvoke = true;
1039 void CWriter::visitUnwindInst(UnwindInst &I) {
1040 // The unwind instructions causes a control flow transfer out of the current
1041 // function, unwinding the stack until a caller who used the invoke
1042 // instruction is found. In this context, we code generated the invoke
1043 // instruction to add an entry to the top of the jmpbuf_list. Thus, here we
1044 // just have to longjmp to the specified handler.
1045 Out << " if (__llvm_jmpbuf_list == 0) { /* unwind */\n"
1046 << " extern write();\n"
1047 << " ((void (*)(int, void*, unsigned))write)(2,\n"
1048 << " \"throw found with no handler!\\n\", 31); abort();\n"
1050 << " longjmp(__llvm_jmpbuf_list->buf, 1);\n";
1051 emittedInvoke = true;
1054 static bool isGotoCodeNecessary(BasicBlock *From, BasicBlock *To) {
1055 // If PHI nodes need copies, we need the copy code...
1056 if (isa<PHINode>(To->front()) ||
1057 From->getNext() != To) // Not directly successor, need goto
1060 // Otherwise we don't need the code.
1064 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
1066 for (BasicBlock::iterator I = Succ->begin();
1067 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1068 // now we have to do the printing
1069 Out << std::string(Indent, ' ');
1070 Out << " " << Mang->getValueName(I) << "__PHI_TEMPORARY = ";
1071 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1072 Out << "; /* for PHI node */\n";
1075 if (CurBB->getNext() != Succ ||
1076 isa<InvokeInst>(CurBB->getTerminator()) ||
1077 isa<SwitchInst>(CurBB->getTerminator())) {
1078 Out << std::string(Indent, ' ') << " goto ";
1084 // Branch instruction printing - Avoid printing out a branch to a basic block
1085 // that immediately succeeds the current one.
1087 void CWriter::visitBranchInst(BranchInst &I) {
1088 if (I.isConditional()) {
1089 if (isGotoCodeNecessary(I.getParent(), I.getSuccessor(0))) {
1091 writeOperand(I.getCondition());
1094 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1096 if (isGotoCodeNecessary(I.getParent(), I.getSuccessor(1))) {
1097 Out << " } else {\n";
1098 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1101 // First goto not necessary, assume second one is...
1103 writeOperand(I.getCondition());
1106 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1111 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1116 // PHI nodes get copied into temporary values at the end of predecessor basic
1117 // blocks. We now need to copy these temporary values into the REAL value for
1119 void CWriter::visitPHINode(PHINode &I) {
1121 Out << "__PHI_TEMPORARY";
1125 void CWriter::visitBinaryOperator(Instruction &I) {
1126 // binary instructions, shift instructions, setCond instructions.
1127 assert(!isa<PointerType>(I.getType()));
1129 // We must cast the results of binary operations which might be promoted.
1130 bool needsCast = false;
1131 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1132 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1133 || (I.getType() == Type::FloatTy)) {
1136 printType(Out, I.getType());
1140 writeOperand(I.getOperand(0));
1142 switch (I.getOpcode()) {
1143 case Instruction::Add: Out << " + "; break;
1144 case Instruction::Sub: Out << " - "; break;
1145 case Instruction::Mul: Out << "*"; break;
1146 case Instruction::Div: Out << "/"; break;
1147 case Instruction::Rem: Out << "%"; break;
1148 case Instruction::And: Out << " & "; break;
1149 case Instruction::Or: Out << " | "; break;
1150 case Instruction::Xor: Out << " ^ "; break;
1151 case Instruction::SetEQ: Out << " == "; break;
1152 case Instruction::SetNE: Out << " != "; break;
1153 case Instruction::SetLE: Out << " <= "; break;
1154 case Instruction::SetGE: Out << " >= "; break;
1155 case Instruction::SetLT: Out << " < "; break;
1156 case Instruction::SetGT: Out << " > "; break;
1157 case Instruction::Shl : Out << " << "; break;
1158 case Instruction::Shr : Out << " >> "; break;
1159 default: std::cerr << "Invalid operator type!" << I; abort();
1162 writeOperand(I.getOperand(1));
1169 void CWriter::visitCastInst(CastInst &I) {
1170 if (I.getType() == Type::BoolTy) {
1172 writeOperand(I.getOperand(0));
1177 printType(Out, I.getType());
1179 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1180 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1181 // Avoid "cast to pointer from integer of different size" warnings
1185 writeOperand(I.getOperand(0));
1188 void CWriter::visitCallInst(CallInst &I) {
1189 // Handle intrinsic function calls first...
1190 if (Function *F = I.getCalledFunction())
1191 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID()) {
1193 default: assert(0 && "Unknown LLVM intrinsic!");
1194 case LLVMIntrinsic::va_start:
1197 Out << "va_start(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1198 // Output the last argument to the enclosing function...
1199 if (I.getParent()->getParent()->aempty()) {
1200 std::cerr << "The C backend does not currently support zero "
1201 << "argument varargs functions, such as '"
1202 << I.getParent()->getParent()->getName() << "'!\n";
1205 writeOperand(&I.getParent()->getParent()->aback());
1208 case LLVMIntrinsic::va_end:
1209 Out << "va_end(*(va_list*)&";
1210 writeOperand(I.getOperand(1));
1213 case LLVMIntrinsic::va_copy:
1215 Out << "va_copy(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1216 Out << "*(va_list*)&";
1217 writeOperand(I.getOperand(1));
1220 case LLVMIntrinsic::setjmp:
1221 case LLVMIntrinsic::sigsetjmp:
1222 // This intrinsic should never exist in the program, but until we get
1223 // setjmp/longjmp transformations going on, we should codegen it to
1224 // something reasonable. This will allow code that never calls longjmp
1228 case LLVMIntrinsic::longjmp:
1229 case LLVMIntrinsic::siglongjmp:
1230 // Longjmp is not implemented, and never will be. It would cause an
1239 void CWriter::visitCallSite(CallSite CS) {
1240 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1241 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1242 const Type *RetTy = FTy->getReturnType();
1244 writeOperand(CS.getCalledValue());
1247 if (CS.arg_begin() != CS.arg_end()) {
1248 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1251 for (++AI; AI != AE; ++AI) {
1259 void CWriter::visitMallocInst(MallocInst &I) {
1261 printType(Out, I.getType());
1262 Out << ")malloc(sizeof(";
1263 printType(Out, I.getType()->getElementType());
1266 if (I.isArrayAllocation()) {
1268 writeOperand(I.getOperand(0));
1273 void CWriter::visitAllocaInst(AllocaInst &I) {
1275 printType(Out, I.getType());
1276 Out << ") alloca(sizeof(";
1277 printType(Out, I.getType()->getElementType());
1279 if (I.isArrayAllocation()) {
1281 writeOperand(I.getOperand(0));
1286 void CWriter::visitFreeInst(FreeInst &I) {
1287 Out << "free((char*)";
1288 writeOperand(I.getOperand(0));
1292 void CWriter::printIndexingExpression(Value *Ptr, User::op_iterator I,
1293 User::op_iterator E) {
1294 bool HasImplicitAddress = false;
1295 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1296 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1297 HasImplicitAddress = true;
1298 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1299 HasImplicitAddress = true;
1300 Ptr = CPR->getValue(); // Get to the global...
1301 } else if (isDirectAlloca(Ptr)) {
1302 HasImplicitAddress = true;
1306 if (!HasImplicitAddress)
1307 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1309 writeOperandInternal(Ptr);
1313 const Constant *CI = dyn_cast<Constant>(I);
1314 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1317 writeOperandInternal(Ptr);
1319 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1321 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1324 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1325 "Can only have implicit address with direct accessing");
1327 if (HasImplicitAddress) {
1329 } else if (CI && CI->isNullValue() && I+1 != E) {
1330 // Print out the -> operator if possible...
1331 if ((*(I+1))->getType() == Type::UByteTy) {
1332 Out << (HasImplicitAddress ? "." : "->");
1333 Out << "field" << cast<ConstantUInt>(*(I+1))->getValue();
1339 if ((*I)->getType() == Type::LongTy) {
1344 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
1348 void CWriter::visitLoadInst(LoadInst &I) {
1350 writeOperand(I.getOperand(0));
1353 void CWriter::visitStoreInst(StoreInst &I) {
1355 writeOperand(I.getPointerOperand());
1357 writeOperand(I.getOperand(0));
1360 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1362 printIndexingExpression(I.getPointerOperand(), I.idx_begin(), I.idx_end());
1365 void CWriter::visitVANextInst(VANextInst &I) {
1366 Out << Mang->getValueName(I.getOperand(0));
1367 Out << "; va_arg(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1368 printType(Out, I.getArgType());
1372 void CWriter::visitVAArgInst(VAArgInst &I) {
1374 Out << "{ va_list Tmp; va_copy(Tmp, *(va_list*)&";
1375 writeOperand(I.getOperand(0));
1376 Out << ");\n " << Mang->getValueName(&I) << " = va_arg(Tmp, ";
1377 printType(Out, I.getType());
1378 Out << ");\n va_end(Tmp); }";
1382 //===----------------------------------------------------------------------===//
1383 // External Interface declaration
1384 //===----------------------------------------------------------------------===//
1386 Pass *createWriteToCPass(std::ostream &o) { return new CWriter(o); }