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"
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;
45 bool needsMalloc, emittedInvoke;
47 std::map<const ConstantFP *, unsigned> FPConstantMap;
49 CWriter(std::ostream &o) : Out(o) {}
51 void getAnalysisUsage(AnalysisUsage &AU) const {
53 AU.addRequired<FindUsedTypes>();
56 virtual bool run(Module &M) {
59 FUT = &getAnalysis<FindUsedTypes>();
61 // Ensure that all structure types have names...
62 bool Changed = nameAllUsedStructureTypes(M);
63 Mang = new Mangler(M);
71 MangledGlobals.clear();
75 std::ostream &printType(std::ostream &Out, const Type *Ty,
76 const std::string &VariableName = "",
77 bool IgnoreName = false);
79 void writeOperand(Value *Operand);
80 void writeOperandInternal(Value *Operand);
83 bool nameAllUsedStructureTypes(Module &M);
84 void printModule(Module *M);
85 void printFloatingPointConstants(Module &M);
86 void printSymbolTable(const SymbolTable &ST);
87 void printContainedStructs(const Type *Ty, std::set<const StructType *> &);
88 void printFunctionSignature(const Function *F, bool Prototype);
90 void printFunction(Function *);
92 void printConstant(Constant *CPV);
93 void printConstantArray(ConstantArray *CPA);
95 // isInlinableInst - Attempt to inline instructions into their uses to build
96 // trees as much as possible. To do this, we have to consistently decide
97 // what is acceptable to inline, so that variable declarations don't get
98 // printed and an extra copy of the expr is not emitted.
100 static bool isInlinableInst(const Instruction &I) {
101 // Must be an expression, must be used exactly once. If it is dead, we
102 // emit it inline where it would go.
103 if (I.getType() == Type::VoidTy || !I.hasOneUse() ||
104 isa<TerminatorInst>(I) || isa<CallInst>(I) || isa<PHINode>(I) ||
105 isa<LoadInst>(I) || isa<VAArgInst>(I) || isa<VANextInst>(I))
106 // Don't inline a load across a store or other bad things!
109 // Only inline instruction it it's use is in the same BB as the inst.
110 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
113 // isDirectAlloca - Define fixed sized allocas in the entry block as direct
114 // variables which are accessed with the & operator. This causes GCC to
115 // generate significantly better code than to emit alloca calls directly.
117 static const AllocaInst *isDirectAlloca(const Value *V) {
118 const AllocaInst *AI = dyn_cast<AllocaInst>(V);
119 if (!AI) return false;
120 if (AI->isArrayAllocation())
121 return 0; // FIXME: we can also inline fixed size array allocas!
122 if (AI->getParent() != &AI->getParent()->getParent()->getEntryBlock())
127 // Instruction visitation functions
128 friend class InstVisitor<CWriter>;
130 void visitReturnInst(ReturnInst &I);
131 void visitBranchInst(BranchInst &I);
132 void visitSwitchInst(SwitchInst &I);
133 void visitInvokeInst(InvokeInst &I);
134 void visitUnwindInst(UnwindInst &I);
136 void visitPHINode(PHINode &I);
137 void visitBinaryOperator(Instruction &I);
139 void visitCastInst (CastInst &I);
140 void visitCallInst (CallInst &I);
141 void visitCallSite (CallSite CS);
142 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
144 void visitMallocInst(MallocInst &I);
145 void visitAllocaInst(AllocaInst &I);
146 void visitFreeInst (FreeInst &I);
147 void visitLoadInst (LoadInst &I);
148 void visitStoreInst (StoreInst &I);
149 void visitGetElementPtrInst(GetElementPtrInst &I);
150 void visitVANextInst(VANextInst &I);
151 void visitVAArgInst (VAArgInst &I);
153 void visitInstruction(Instruction &I) {
154 std::cerr << "C Writer does not know about " << I;
158 void outputLValue(Instruction *I) {
159 Out << " " << Mang->getValueName(I) << " = ";
161 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
163 void printIndexingExpression(Value *Ptr, User::op_iterator I,
164 User::op_iterator E);
167 // Pass the Type* and the variable name and this prints out the variable
170 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
171 const std::string &NameSoFar,
173 if (Ty->isPrimitiveType())
174 switch (Ty->getPrimitiveID()) {
175 case Type::VoidTyID: return Out << "void " << NameSoFar;
176 case Type::BoolTyID: return Out << "bool " << NameSoFar;
177 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
178 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
179 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
180 case Type::ShortTyID: return Out << "short " << NameSoFar;
181 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
182 case Type::IntTyID: return Out << "int " << NameSoFar;
183 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
184 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
185 case Type::FloatTyID: return Out << "float " << NameSoFar;
186 case Type::DoubleTyID: return Out << "double " << NameSoFar;
188 std::cerr << "Unknown primitive type: " << Ty << "\n";
192 // Check to see if the type is named.
193 if (!IgnoreName || isa<OpaqueType>(Ty)) {
194 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
195 if (I != TypeNames.end()) return Out << I->second << " " << NameSoFar;
198 switch (Ty->getPrimitiveID()) {
199 case Type::FunctionTyID: {
200 const FunctionType *MTy = cast<FunctionType>(Ty);
201 std::stringstream FunctionInnards;
202 FunctionInnards << " (" << NameSoFar << ") (";
203 for (FunctionType::ParamTypes::const_iterator
204 I = MTy->getParamTypes().begin(),
205 E = MTy->getParamTypes().end(); I != E; ++I) {
206 if (I != MTy->getParamTypes().begin())
207 FunctionInnards << ", ";
208 printType(FunctionInnards, *I, "");
210 if (MTy->isVarArg()) {
211 if (!MTy->getParamTypes().empty())
212 FunctionInnards << ", ...";
213 } else if (MTy->getParamTypes().empty()) {
214 FunctionInnards << "void";
216 FunctionInnards << ")";
217 std::string tstr = FunctionInnards.str();
218 printType(Out, MTy->getReturnType(), tstr);
221 case Type::StructTyID: {
222 const StructType *STy = cast<StructType>(Ty);
223 Out << NameSoFar + " {\n";
225 for (StructType::ElementTypes::const_iterator
226 I = STy->getElementTypes().begin(),
227 E = STy->getElementTypes().end(); I != E; ++I) {
229 printType(Out, *I, "field" + utostr(Idx++));
235 case Type::PointerTyID: {
236 const PointerType *PTy = cast<PointerType>(Ty);
237 std::string ptrName = "*" + NameSoFar;
239 if (isa<ArrayType>(PTy->getElementType()))
240 ptrName = "(" + ptrName + ")";
242 return printType(Out, PTy->getElementType(), ptrName);
245 case Type::ArrayTyID: {
246 const ArrayType *ATy = cast<ArrayType>(Ty);
247 unsigned NumElements = ATy->getNumElements();
248 return printType(Out, ATy->getElementType(),
249 NameSoFar + "[" + utostr(NumElements) + "]");
252 case Type::OpaqueTyID: {
253 static int Count = 0;
254 std::string TyName = "struct opaque_" + itostr(Count++);
255 assert(TypeNames.find(Ty) == TypeNames.end());
256 TypeNames[Ty] = TyName;
257 return Out << TyName << " " << NameSoFar;
260 assert(0 && "Unhandled case in getTypeProps!");
267 void CWriter::printConstantArray(ConstantArray *CPA) {
269 // As a special case, print the array as a string if it is an array of
270 // ubytes or an array of sbytes with positive values.
272 const Type *ETy = CPA->getType()->getElementType();
273 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
275 // Make sure the last character is a null char, as automatically added by C
276 if (isString && (CPA->getNumOperands() == 0 ||
277 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
282 // Keep track of whether the last number was a hexadecimal escape
283 bool LastWasHex = false;
285 // Do not include the last character, which we know is null
286 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
287 unsigned char C = cast<ConstantInt>(CPA->getOperand(i))->getRawValue();
289 // Print it out literally if it is a printable character. The only thing
290 // to be careful about is when the last letter output was a hex escape
291 // code, in which case we have to be careful not to print out hex digits
292 // explicitly (the C compiler thinks it is a continuation of the previous
293 // character, sheesh...)
295 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
297 if (C == '"' || C == '\\')
304 case '\n': Out << "\\n"; break;
305 case '\t': Out << "\\t"; break;
306 case '\r': Out << "\\r"; break;
307 case '\v': Out << "\\v"; break;
308 case '\a': Out << "\\a"; break;
309 case '\"': Out << "\\\""; break;
310 case '\'': Out << "\\\'"; break;
313 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
314 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
323 if (CPA->getNumOperands()) {
325 printConstant(cast<Constant>(CPA->getOperand(0)));
326 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
328 printConstant(cast<Constant>(CPA->getOperand(i)));
335 // isFPCSafeToPrint - Returns true if we may assume that CFP may be written out
336 // textually as a double (rather than as a reference to a stack-allocated
337 // variable). We decide this by converting CFP to a string and back into a
338 // double, and then checking whether the conversion results in a bit-equal
339 // double to the original value of CFP. This depends on us and the target C
340 // compiler agreeing on the conversion process (which is pretty likely since we
341 // only deal in IEEE FP).
343 bool isFPCSafeToPrint(const ConstantFP *CFP) {
346 sprintf(Buffer, "%a", CFP->getValue());
348 if (!strncmp(Buffer, "0x", 2) ||
349 !strncmp(Buffer, "-0x", 3) ||
350 !strncmp(Buffer, "+0x", 3))
351 return atof(Buffer) == CFP->getValue();
354 std::string StrVal = ftostr(CFP->getValue());
356 while (StrVal[0] == ' ')
357 StrVal.erase(StrVal.begin());
359 // Check to make sure that the stringized number is not some string like "Inf"
360 // or NaN. Check that the string matches the "[-+]?[0-9]" regex.
361 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
362 ((StrVal[0] == '-' || StrVal[0] == '+') &&
363 (StrVal[1] >= '0' && StrVal[1] <= '9')))
364 // Reparse stringized version!
365 return atof(StrVal.c_str()) == CFP->getValue();
370 // printConstant - The LLVM Constant to C Constant converter.
371 void CWriter::printConstant(Constant *CPV) {
372 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
373 switch (CE->getOpcode()) {
374 case Instruction::Cast:
376 printType(Out, CPV->getType());
378 printConstant(CE->getOperand(0));
382 case Instruction::GetElementPtr:
384 printIndexingExpression(CE->getOperand(0),
385 CPV->op_begin()+1, CPV->op_end());
388 case Instruction::Add:
389 case Instruction::Sub:
390 case Instruction::Mul:
391 case Instruction::Div:
392 case Instruction::Rem:
393 case Instruction::SetEQ:
394 case Instruction::SetNE:
395 case Instruction::SetLT:
396 case Instruction::SetLE:
397 case Instruction::SetGT:
398 case Instruction::SetGE:
400 printConstant(CE->getOperand(0));
401 switch (CE->getOpcode()) {
402 case Instruction::Add: Out << " + "; break;
403 case Instruction::Sub: Out << " - "; break;
404 case Instruction::Mul: Out << " * "; break;
405 case Instruction::Div: Out << " / "; break;
406 case Instruction::Rem: Out << " % "; break;
407 case Instruction::SetEQ: Out << " == "; break;
408 case Instruction::SetNE: Out << " != "; break;
409 case Instruction::SetLT: Out << " < "; break;
410 case Instruction::SetLE: Out << " <= "; break;
411 case Instruction::SetGT: Out << " > "; break;
412 case Instruction::SetGE: Out << " >= "; break;
413 default: assert(0 && "Illegal opcode here!");
415 printConstant(CE->getOperand(1));
420 std::cerr << "CWriter Error: Unhandled constant expression: "
426 switch (CPV->getType()->getPrimitiveID()) {
428 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
429 case Type::SByteTyID:
430 case Type::ShortTyID:
431 Out << cast<ConstantSInt>(CPV)->getValue(); break;
433 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
434 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
436 Out << cast<ConstantSInt>(CPV)->getValue();
440 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
442 case Type::UByteTyID:
443 case Type::UShortTyID:
444 Out << cast<ConstantUInt>(CPV)->getValue(); break;
446 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
447 case Type::ULongTyID:
448 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
450 case Type::FloatTyID:
451 case Type::DoubleTyID: {
452 ConstantFP *FPC = cast<ConstantFP>(CPV);
453 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
454 if (I != FPConstantMap.end()) {
455 // Because of FP precision problems we must load from a stack allocated
456 // value that holds the value in hex.
457 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
458 << "*)&FPConstant" << I->second << ")";
461 // Print out the constant as a floating point number.
463 sprintf(Buffer, "%a", FPC->getValue());
464 Out << Buffer << " /*" << FPC->getValue() << "*/ ";
466 Out << ftostr(FPC->getValue());
472 case Type::ArrayTyID:
473 printConstantArray(cast<ConstantArray>(CPV));
476 case Type::StructTyID: {
478 if (CPV->getNumOperands()) {
480 printConstant(cast<Constant>(CPV->getOperand(0)));
481 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
483 printConstant(cast<Constant>(CPV->getOperand(i)));
490 case Type::PointerTyID:
491 if (isa<ConstantPointerNull>(CPV)) {
493 printType(Out, CPV->getType());
494 Out << ")/*NULL*/0)";
496 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
497 writeOperand(CPR->getValue());
502 std::cerr << "Unknown constant type: " << CPV << "\n";
507 void CWriter::writeOperandInternal(Value *Operand) {
508 if (Instruction *I = dyn_cast<Instruction>(Operand))
509 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
510 // Should we inline this instruction to build a tree?
517 if (Constant *CPV = dyn_cast<Constant>(Operand)) {
520 Out << Mang->getValueName(Operand);
524 void CWriter::writeOperand(Value *Operand) {
525 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
526 Out << "(&"; // Global variables are references as their addresses by llvm
528 writeOperandInternal(Operand);
530 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
534 // nameAllUsedStructureTypes - If there are structure types in the module that
535 // are used but do not have names assigned to them in the symbol table yet then
536 // we assign them names now.
538 bool CWriter::nameAllUsedStructureTypes(Module &M) {
539 // Get a set of types that are used by the program...
540 std::set<const Type *> UT = FUT->getTypes();
542 // Loop over the module symbol table, removing types from UT that are already
545 SymbolTable &MST = M.getSymbolTable();
546 if (MST.find(Type::TypeTy) != MST.end())
547 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
548 E = MST.type_end(Type::TypeTy); I != E; ++I)
549 UT.erase(cast<Type>(I->second));
551 // UT now contains types that are not named. Loop over it, naming structure
554 bool Changed = false;
555 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
557 if (const StructType *ST = dyn_cast<StructType>(*I)) {
558 ((Value*)ST)->setName("unnamed", &MST);
564 // generateCompilerSpecificCode - This is where we add conditional compilation
565 // directives to cater to specific compilers as need be.
567 void generateCompilerSpecificCode(std::ostream& Out) {
568 // Alloca is hard to get, and we don't want to include stdlib.h here...
569 Out << "/* get a declaration for alloca */\n"
571 << "extern void *__builtin_alloca(unsigned long);\n"
572 << "#define alloca(x) __builtin_alloca(x)\n"
574 << "#ifndef __FreeBSD__\n"
575 << "#include <alloca.h>\n"
579 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
580 // If we aren't being compiled with GCC, just drop these attributes.
581 Out << "#ifndef __GNUC__ /* Can only support \"linkonce\" vars with GCC */\n"
582 << "#define __attribute__(X)\n"
586 void CWriter::printModule(Module *M) {
587 // Calculate which global values have names that will collide when we throw
588 // away type information.
589 { // Scope to delete the FoundNames set when we are done with it...
590 std::set<std::string> FoundNames;
591 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
592 if (I->hasName()) // If the global has a name...
593 if (FoundNames.count(I->getName())) // And the name is already used
594 MangledGlobals.insert(I); // Mangle the name
596 FoundNames.insert(I->getName()); // Otherwise, keep track of name
598 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
599 if (I->hasName()) // If the global has a name...
600 if (FoundNames.count(I->getName())) // And the name is already used
601 MangledGlobals.insert(I); // Mangle the name
603 FoundNames.insert(I->getName()); // Otherwise, keep track of name
606 // get declaration for alloca
607 Out << "/* Provide Declarations */\n";
608 Out << "#include <stdarg.h>\n";
609 Out << "#include <setjmp.h>\n";
610 generateCompilerSpecificCode(Out);
612 // Provide a definition for `bool' if not compiling with a C++ compiler.
614 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
616 << "\n\n/* Support for floating point constants */\n"
617 << "typedef unsigned long long ConstantDoubleTy;\n"
618 << "typedef unsigned int ConstantFloatTy;\n"
620 << "\n\n/* Support for the invoke instruction */\n"
621 << "extern struct __llvm_jmpbuf_list_t {\n"
622 << " jmp_buf buf; struct __llvm_jmpbuf_list_t *next;\n"
623 << "} *__llvm_jmpbuf_list;\n"
625 << "\n\n/* Global Declarations */\n";
627 // First output all the declarations for the program, because C requires
628 // Functions & globals to be declared before they are used.
631 // Loop over the symbol table, emitting all named constants...
632 printSymbolTable(M->getSymbolTable());
634 // Global variable declarations...
636 Out << "\n/* External Global Variable Declarations */\n";
637 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I) {
638 if (I->hasExternalLinkage()) {
640 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
646 // Function declarations
648 Out << "\n/* Function Declarations */\n";
650 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
651 // If the function is external and the name collides don't print it.
652 // Sometimes the bytecode likes to have multiple "declarations" for
653 // external functions
654 if ((I->hasInternalLinkage() || !MangledGlobals.count(I)) &&
655 !I->getIntrinsicID()) {
656 printFunctionSignature(I, true);
657 if (I->hasWeakLinkage()) Out << " __attribute__((weak))";
663 // Print Malloc prototype if needed
665 Out << "\n/* Malloc to make sun happy */\n";
666 Out << "extern void * malloc();\n\n";
669 // Output the global variable declarations
671 Out << "\n\n/* Global Variable Declarations */\n";
672 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
673 if (!I->isExternal()) {
675 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
677 if (I->hasLinkOnceLinkage())
678 Out << " __attribute__((common))";
679 else if (I->hasWeakLinkage())
680 Out << " __attribute__((weak))";
685 // Output the global variable definitions and contents...
687 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
688 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
689 if (!I->isExternal()) {
690 if (I->hasInternalLinkage())
692 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
693 if (I->hasLinkOnceLinkage())
694 Out << " __attribute__((common))";
695 else if (I->hasWeakLinkage())
696 Out << " __attribute__((weak))";
698 // If the initializer is not null, emit the initializer. If it is null,
699 // we try to avoid emitting large amounts of zeros. The problem with
700 // this, however, occurs when the variable has weak linkage. In this
701 // case, the assembler will complain about the variable being both weak
702 // and common, so we disable this optimization.
703 if (!I->getInitializer()->isNullValue() ||
704 I->hasWeakLinkage()) {
706 writeOperand(I->getInitializer());
712 // Output all floating point constants that cannot be printed accurately...
713 printFloatingPointConstants(*M);
715 // Output all of the functions...
716 emittedInvoke = false;
718 Out << "\n\n/* Function Bodies */\n";
719 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
723 // If the program included an invoke instruction, we need to output the
724 // support code for it here!
726 Out << "\n/* More support for the invoke instruction */\n"
727 << "struct __llvm_jmpbuf_list_t *__llvm_jmpbuf_list "
728 << "__attribute__((common)) = 0;\n";
731 // Done with global FP constants
732 FPConstantMap.clear();
735 /// Output all floating point constants that cannot be printed accurately...
736 void CWriter::printFloatingPointConstants(Module &M) {
739 unsigned long long U;
747 // Scan the module for floating point constants. If any FP constant is used
748 // in the function, we want to redirect it here so that we do not depend on
749 // the precision of the printed form, unless the printed form preserves
752 unsigned FPCounter = 0;
753 for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F)
754 for (constant_iterator I = constant_begin(F), E = constant_end(F);
756 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
757 if (!isFPCSafeToPrint(FPC) && // Do not put in FPConstantMap if safe.
758 !FPConstantMap.count(FPC)) {
759 double Val = FPC->getValue();
761 FPConstantMap[FPC] = FPCounter; // Number the FP constants
763 if (FPC->getType() == Type::DoubleTy) {
765 Out << "const ConstantDoubleTy FPConstant" << FPCounter++
766 << " = 0x" << std::hex << DBLUnion.U << std::dec
767 << "ULL; /* " << Val << " */\n";
768 } else if (FPC->getType() == Type::FloatTy) {
770 Out << "const ConstantFloatTy FPConstant" << FPCounter++
771 << " = 0x" << std::hex << FLTUnion.U << std::dec
772 << "U; /* " << Val << " */\n";
774 assert(0 && "Unknown float type!");
781 /// printSymbolTable - Run through symbol table looking for type names. If a
782 /// type name is found, emit it's declaration...
784 void CWriter::printSymbolTable(const SymbolTable &ST) {
785 // If there are no type names, exit early.
786 if (ST.find(Type::TypeTy) == ST.end())
789 // We are only interested in the type plane of the symbol table...
790 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
791 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
793 // Print out forward declarations for structure types before anything else!
794 Out << "/* Structure forward decls */\n";
795 for (; I != End; ++I)
796 if (const Type *STy = dyn_cast<StructType>(I->second))
797 // Only print out used types!
798 if (FUT->getTypes().count(STy)) {
799 std::string Name = "struct l_" + Mangler::makeNameProper(I->first);
800 Out << Name << ";\n";
801 TypeNames.insert(std::make_pair(STy, Name));
806 // Now we can print out typedefs...
807 Out << "/* Typedefs */\n";
808 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
809 // Only print out used types!
810 if (FUT->getTypes().count(cast<Type>(I->second))) {
811 const Type *Ty = cast<Type>(I->second);
812 std::string Name = "l_" + Mangler::makeNameProper(I->first);
814 printType(Out, Ty, Name);
820 // Keep track of which structures have been printed so far...
821 std::set<const StructType *> StructPrinted;
823 // Loop over all structures then push them into the stack so they are
824 // printed in the correct order.
826 Out << "/* Structure contents */\n";
827 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
828 if (const StructType *STy = dyn_cast<StructType>(I->second))
829 // Only print out used types!
830 if (FUT->getTypes().count(STy))
831 printContainedStructs(STy, StructPrinted);
834 // Push the struct onto the stack and recursively push all structs
835 // this one depends on.
836 void CWriter::printContainedStructs(const Type *Ty,
837 std::set<const StructType*> &StructPrinted){
838 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
839 //Check to see if we have already printed this struct
840 if (StructPrinted.count(STy) == 0) {
841 // Print all contained types first...
842 for (StructType::ElementTypes::const_iterator
843 I = STy->getElementTypes().begin(),
844 E = STy->getElementTypes().end(); I != E; ++I) {
845 const Type *Ty1 = I->get();
846 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
847 printContainedStructs(*I, StructPrinted);
850 //Print structure type out..
851 StructPrinted.insert(STy);
852 std::string Name = TypeNames[STy];
853 printType(Out, STy, Name, true);
857 // If it is an array, check contained types and continue
858 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
859 const Type *Ty1 = ATy->getElementType();
860 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
861 printContainedStructs(Ty1, StructPrinted);
866 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
867 // If the program provides its own malloc prototype we don't need
868 // to include the general one.
869 if (Mang->getValueName(F) == "malloc")
872 if (F->hasInternalLinkage()) Out << "static ";
873 if (F->hasLinkOnceLinkage()) Out << "inline ";
875 // Loop over the arguments, printing them...
876 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
878 std::stringstream FunctionInnards;
880 // Print out the name...
881 FunctionInnards << Mang->getValueName(F) << "(";
883 if (!F->isExternal()) {
886 if (F->abegin()->hasName() || !Prototype)
887 ArgName = Mang->getValueName(F->abegin());
888 printType(FunctionInnards, F->afront().getType(), ArgName);
889 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
891 FunctionInnards << ", ";
892 if (I->hasName() || !Prototype)
893 ArgName = Mang->getValueName(I);
896 printType(FunctionInnards, I->getType(), ArgName);
900 // Loop over the arguments, printing them...
901 for (FunctionType::ParamTypes::const_iterator I =
902 FT->getParamTypes().begin(),
903 E = FT->getParamTypes().end(); I != E; ++I) {
904 if (I != FT->getParamTypes().begin()) FunctionInnards << ", ";
905 printType(FunctionInnards, *I);
909 // Finish printing arguments... if this is a vararg function, print the ...,
910 // unless there are no known types, in which case, we just emit ().
912 if (FT->isVarArg() && !FT->getParamTypes().empty()) {
913 if (FT->getParamTypes().size()) FunctionInnards << ", ";
914 FunctionInnards << "..."; // Output varargs portion of signature!
916 FunctionInnards << ")";
917 // Print out the return type and the entire signature for that matter
918 printType(Out, F->getReturnType(), FunctionInnards.str());
921 void CWriter::printFunction(Function *F) {
922 if (F->isExternal()) return;
924 printFunctionSignature(F, false);
927 // print local variable information for the function
928 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
929 if (const AllocaInst *AI = isDirectAlloca(*I)) {
931 printType(Out, AI->getAllocatedType(), Mang->getValueName(AI));
932 Out << "; /* Address exposed local */\n";
933 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
935 printType(Out, (*I)->getType(), Mang->getValueName(*I));
938 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
940 printType(Out, (*I)->getType(),
941 Mang->getValueName(*I)+"__PHI_TEMPORARY");
948 // print the basic blocks
949 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
950 BasicBlock *Prev = BB->getPrev();
952 // Don't print the label for the basic block if there are no uses, or if the
953 // only terminator use is the predecessor basic block's terminator. We have
954 // to scan the use list because PHI nodes use basic blocks too but do not
955 // require a label to be generated.
957 bool NeedsLabel = false;
958 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
960 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
961 if (TI != Prev->getTerminator() ||
962 isa<SwitchInst>(Prev->getTerminator()) ||
963 isa<InvokeInst>(Prev->getTerminator())) {
968 if (NeedsLabel) Out << Mang->getValueName(BB) << ":\n";
970 // Output all of the instructions in the basic block...
971 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
972 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
973 if (II->getType() != Type::VoidTy)
982 // Don't emit prefix or suffix for the terminator...
983 visit(*BB->getTerminator());
989 // Specific Instruction type classes... note that all of the casts are
990 // necessary because we use the instruction classes as opaque types...
992 void CWriter::visitReturnInst(ReturnInst &I) {
993 // Don't output a void return if this is the last basic block in the function
994 if (I.getNumOperands() == 0 &&
995 &*--I.getParent()->getParent()->end() == I.getParent() &&
996 !I.getParent()->size() == 1) {
1001 if (I.getNumOperands()) {
1003 writeOperand(I.getOperand(0));
1008 void CWriter::visitSwitchInst(SwitchInst &SI) {
1010 writeOperand(SI.getOperand(0));
1011 Out << ") {\n default:\n";
1012 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
1014 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
1016 writeOperand(SI.getOperand(i));
1018 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
1019 printBranchToBlock(SI.getParent(), Succ, 2);
1020 if (Succ == SI.getParent()->getNext())
1026 void CWriter::visitInvokeInst(InvokeInst &II) {
1028 << " struct __llvm_jmpbuf_list_t Entry;\n"
1029 << " Entry.next = __llvm_jmpbuf_list;\n"
1030 << " if (setjmp(Entry.buf)) {\n"
1031 << " __llvm_jmpbuf_list = Entry.next;\n";
1032 printBranchToBlock(II.getParent(), II.getExceptionalDest(), 4);
1034 << " __llvm_jmpbuf_list = &Entry;\n"
1037 if (II.getType() != Type::VoidTy) outputLValue(&II);
1040 << " __llvm_jmpbuf_list = Entry.next;\n"
1042 printBranchToBlock(II.getParent(), II.getNormalDest(), 0);
1043 emittedInvoke = true;
1047 void CWriter::visitUnwindInst(UnwindInst &I) {
1048 // The unwind instructions causes a control flow transfer out of the current
1049 // function, unwinding the stack until a caller who used the invoke
1050 // instruction is found. In this context, we code generated the invoke
1051 // instruction to add an entry to the top of the jmpbuf_list. Thus, here we
1052 // just have to longjmp to the specified handler.
1053 Out << " if (__llvm_jmpbuf_list == 0) { /* unwind */\n"
1054 << " extern write();\n"
1055 << " ((void (*)(int, void*, unsigned))write)(2,\n"
1056 << " \"throw found with no handler!\\n\", 31); abort();\n"
1058 << " longjmp(__llvm_jmpbuf_list->buf, 1);\n";
1059 emittedInvoke = true;
1062 bool isGotoCodeNecessary(BasicBlock *From, BasicBlock *To) {
1063 // If PHI nodes need copies, we need the copy code...
1064 if (isa<PHINode>(To->front()) ||
1065 From->getNext() != To) // Not directly successor, need goto
1068 // Otherwise we don't need the code.
1072 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
1074 for (BasicBlock::iterator I = Succ->begin();
1075 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1076 // now we have to do the printing
1077 Out << std::string(Indent, ' ');
1078 Out << " " << Mang->getValueName(I) << "__PHI_TEMPORARY = ";
1079 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1080 Out << "; /* for PHI node */\n";
1083 if (CurBB->getNext() != Succ ||
1084 isa<InvokeInst>(CurBB->getTerminator()) ||
1085 isa<SwitchInst>(CurBB->getTerminator())) {
1086 Out << std::string(Indent, ' ') << " goto ";
1092 // Branch instruction printing - Avoid printing out a branch to a basic block
1093 // that immediately succeeds the current one.
1095 void CWriter::visitBranchInst(BranchInst &I) {
1096 if (I.isConditional()) {
1097 if (isGotoCodeNecessary(I.getParent(), I.getSuccessor(0))) {
1099 writeOperand(I.getCondition());
1102 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1104 if (isGotoCodeNecessary(I.getParent(), I.getSuccessor(1))) {
1105 Out << " } else {\n";
1106 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1109 // First goto not necessary, assume second one is...
1111 writeOperand(I.getCondition());
1114 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1119 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1124 // PHI nodes get copied into temporary values at the end of predecessor basic
1125 // blocks. We now need to copy these temporary values into the REAL value for
1127 void CWriter::visitPHINode(PHINode &I) {
1129 Out << "__PHI_TEMPORARY";
1133 void CWriter::visitBinaryOperator(Instruction &I) {
1134 // binary instructions, shift instructions, setCond instructions.
1135 assert(!isa<PointerType>(I.getType()));
1137 // We must cast the results of binary operations which might be promoted.
1138 bool needsCast = false;
1139 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1140 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1141 || (I.getType() == Type::FloatTy)) {
1144 printType(Out, I.getType());
1148 writeOperand(I.getOperand(0));
1150 switch (I.getOpcode()) {
1151 case Instruction::Add: Out << " + "; break;
1152 case Instruction::Sub: Out << " - "; break;
1153 case Instruction::Mul: Out << "*"; break;
1154 case Instruction::Div: Out << "/"; break;
1155 case Instruction::Rem: Out << "%"; break;
1156 case Instruction::And: Out << " & "; break;
1157 case Instruction::Or: Out << " | "; break;
1158 case Instruction::Xor: Out << " ^ "; break;
1159 case Instruction::SetEQ: Out << " == "; break;
1160 case Instruction::SetNE: Out << " != "; break;
1161 case Instruction::SetLE: Out << " <= "; break;
1162 case Instruction::SetGE: Out << " >= "; break;
1163 case Instruction::SetLT: Out << " < "; break;
1164 case Instruction::SetGT: Out << " > "; break;
1165 case Instruction::Shl : Out << " << "; break;
1166 case Instruction::Shr : Out << " >> "; break;
1167 default: std::cerr << "Invalid operator type!" << I; abort();
1170 writeOperand(I.getOperand(1));
1177 void CWriter::visitCastInst(CastInst &I) {
1178 if (I.getType() == Type::BoolTy) {
1180 writeOperand(I.getOperand(0));
1185 printType(Out, I.getType());
1187 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1188 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1189 // Avoid "cast to pointer from integer of different size" warnings
1193 writeOperand(I.getOperand(0));
1196 void CWriter::visitCallInst(CallInst &I) {
1197 // Handle intrinsic function calls first...
1198 if (Function *F = I.getCalledFunction())
1199 if (Intrinsic::ID ID = (Intrinsic::ID)F->getIntrinsicID()) {
1201 default: assert(0 && "Unknown LLVM intrinsic!");
1202 case Intrinsic::va_start:
1205 Out << "va_start(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1206 // Output the last argument to the enclosing function...
1207 if (I.getParent()->getParent()->aempty()) {
1208 std::cerr << "The C backend does not currently support zero "
1209 << "argument varargs functions, such as '"
1210 << I.getParent()->getParent()->getName() << "'!\n";
1213 writeOperand(&I.getParent()->getParent()->aback());
1216 case Intrinsic::va_end:
1217 Out << "va_end(*(va_list*)&";
1218 writeOperand(I.getOperand(1));
1221 case Intrinsic::va_copy:
1223 Out << "va_copy(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1224 Out << "*(va_list*)&";
1225 writeOperand(I.getOperand(1));
1228 case Intrinsic::setjmp:
1229 case Intrinsic::sigsetjmp:
1230 // This intrinsic should never exist in the program, but until we get
1231 // setjmp/longjmp transformations going on, we should codegen it to
1232 // something reasonable. This will allow code that never calls longjmp
1236 case Intrinsic::longjmp:
1237 case Intrinsic::siglongjmp:
1238 // Longjmp is not implemented, and never will be. It would cause an
1247 void CWriter::visitCallSite(CallSite CS) {
1248 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1249 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1250 const Type *RetTy = FTy->getReturnType();
1252 writeOperand(CS.getCalledValue());
1255 if (CS.arg_begin() != CS.arg_end()) {
1256 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1259 for (++AI; AI != AE; ++AI) {
1267 void CWriter::visitMallocInst(MallocInst &I) {
1269 printType(Out, I.getType());
1270 Out << ")malloc(sizeof(";
1271 printType(Out, I.getType()->getElementType());
1274 if (I.isArrayAllocation()) {
1276 writeOperand(I.getOperand(0));
1281 void CWriter::visitAllocaInst(AllocaInst &I) {
1283 printType(Out, I.getType());
1284 Out << ") alloca(sizeof(";
1285 printType(Out, I.getType()->getElementType());
1287 if (I.isArrayAllocation()) {
1289 writeOperand(I.getOperand(0));
1294 void CWriter::visitFreeInst(FreeInst &I) {
1295 Out << "free((char*)";
1296 writeOperand(I.getOperand(0));
1300 void CWriter::printIndexingExpression(Value *Ptr, User::op_iterator I,
1301 User::op_iterator E) {
1302 bool HasImplicitAddress = false;
1303 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1304 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1305 HasImplicitAddress = true;
1306 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1307 HasImplicitAddress = true;
1308 Ptr = CPR->getValue(); // Get to the global...
1309 } else if (isDirectAlloca(Ptr)) {
1310 HasImplicitAddress = true;
1314 if (!HasImplicitAddress)
1315 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1317 writeOperandInternal(Ptr);
1321 const Constant *CI = dyn_cast<Constant>(I);
1322 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1325 writeOperandInternal(Ptr);
1327 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1329 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1332 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1333 "Can only have implicit address with direct accessing");
1335 if (HasImplicitAddress) {
1337 } else if (CI && CI->isNullValue() && I+1 != E) {
1338 // Print out the -> operator if possible...
1339 if ((*(I+1))->getType() == Type::UByteTy) {
1340 Out << (HasImplicitAddress ? "." : "->");
1341 Out << "field" << cast<ConstantUInt>(*(I+1))->getValue();
1347 if ((*I)->getType() == Type::LongTy) {
1352 Out << ".field" << cast<ConstantUInt>(*I)->getValue();
1356 void CWriter::visitLoadInst(LoadInst &I) {
1358 writeOperand(I.getOperand(0));
1361 void CWriter::visitStoreInst(StoreInst &I) {
1363 writeOperand(I.getPointerOperand());
1365 writeOperand(I.getOperand(0));
1368 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1370 printIndexingExpression(I.getPointerOperand(), I.idx_begin(), I.idx_end());
1373 void CWriter::visitVANextInst(VANextInst &I) {
1374 Out << Mang->getValueName(I.getOperand(0));
1375 Out << "; va_arg(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1376 printType(Out, I.getArgType());
1380 void CWriter::visitVAArgInst(VAArgInst &I) {
1382 Out << "{ va_list Tmp; va_copy(Tmp, *(va_list*)&";
1383 writeOperand(I.getOperand(0));
1384 Out << ");\n " << Mang->getValueName(&I) << " = va_arg(Tmp, ";
1385 printType(Out, I.getType());
1386 Out << ");\n va_end(Tmp); }";
1391 //===----------------------------------------------------------------------===//
1392 // External Interface declaration
1393 //===----------------------------------------------------------------------===//
1395 Pass *createWriteToCPass(std::ostream &o) { return new CWriter(o); }
1397 } // End llvm namespace