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 "CTargetMachine.h"
16 #include "llvm/Target/TargetMachineImpls.h"
17 #include "llvm/Constants.h"
18 #include "llvm/DerivedTypes.h"
19 #include "llvm/Module.h"
20 #include "llvm/Instructions.h"
21 #include "llvm/Pass.h"
22 #include "llvm/PassManager.h"
23 #include "llvm/SymbolTable.h"
24 #include "llvm/Intrinsics.h"
25 #include "llvm/IntrinsicLowering.h"
26 #include "llvm/Analysis/FindUsedTypes.h"
27 #include "llvm/Analysis/ConstantsScanner.h"
28 #include "llvm/Transforms/Scalar.h"
29 #include "llvm/Support/CallSite.h"
30 #include "llvm/Support/GetElementPtrTypeIterator.h"
31 #include "llvm/Support/InstVisitor.h"
32 #include "llvm/Support/Mangler.h"
33 #include "Support/StringExtras.h"
34 #include "Config/config.h"
40 class CWriter : public Pass, public InstVisitor<CWriter> {
42 IntrinsicLowering &IL;
44 const Module *TheModule;
47 std::map<const Type *, std::string> TypeNames;
49 std::map<const ConstantFP *, unsigned> FPConstantMap;
51 CWriter(std::ostream &o, IntrinsicLowering &il) : Out(o), IL(il) {}
53 void getAnalysisUsage(AnalysisUsage &AU) const {
54 AU.addRequired<FindUsedTypes>();
57 virtual const char *getPassName() const { return "C backend"; }
59 bool doInitialization(Module &M);
61 // First pass, lower all unhandled intrinsics.
66 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
76 std::ostream &printType(std::ostream &Out, const Type *Ty,
77 const std::string &VariableName = "",
78 bool IgnoreName = false);
80 void writeOperand(Value *Operand);
81 void writeOperandInternal(Value *Operand);
84 void lowerIntrinsics(Module &M);
86 bool nameAllUsedStructureTypes(Module &M);
87 void printModule(Module *M);
88 void printFloatingPointConstants(Module &M);
89 void printSymbolTable(const SymbolTable &ST);
90 void printContainedStructs(const Type *Ty, std::set<const StructType *> &);
91 void printFunctionSignature(const Function *F, bool Prototype);
93 void printFunction(Function &);
95 void printConstant(Constant *CPV);
96 void printConstantArray(ConstantArray *CPA);
98 // isInlinableInst - Attempt to inline instructions into their uses to build
99 // trees as much as possible. To do this, we have to consistently decide
100 // what is acceptable to inline, so that variable declarations don't get
101 // printed and an extra copy of the expr is not emitted.
103 static bool isInlinableInst(const Instruction &I) {
104 // Must be an expression, must be used exactly once. If it is dead, we
105 // emit it inline where it would go.
106 if (I.getType() == Type::VoidTy || !I.hasOneUse() ||
107 isa<TerminatorInst>(I) || isa<CallInst>(I) || isa<PHINode>(I) ||
108 isa<LoadInst>(I) || isa<VAArgInst>(I) || isa<VANextInst>(I))
109 // Don't inline a load across a store or other bad things!
112 // Only inline instruction it it's use is in the same BB as the inst.
113 return I.getParent() == cast<Instruction>(I.use_back())->getParent();
116 // isDirectAlloca - Define fixed sized allocas in the entry block as direct
117 // variables which are accessed with the & operator. This causes GCC to
118 // generate significantly better code than to emit alloca calls directly.
120 static const AllocaInst *isDirectAlloca(const Value *V) {
121 const AllocaInst *AI = dyn_cast<AllocaInst>(V);
122 if (!AI) return false;
123 if (AI->isArrayAllocation())
124 return 0; // FIXME: we can also inline fixed size array allocas!
125 if (AI->getParent() != &AI->getParent()->getParent()->getEntryBlock())
130 // Instruction visitation functions
131 friend class InstVisitor<CWriter>;
133 void visitReturnInst(ReturnInst &I);
134 void visitBranchInst(BranchInst &I);
135 void visitSwitchInst(SwitchInst &I);
136 void visitInvokeInst(InvokeInst &I);
137 void visitUnwindInst(UnwindInst &I);
139 void visitPHINode(PHINode &I);
140 void visitBinaryOperator(Instruction &I);
142 void visitCastInst (CastInst &I);
143 void visitSelectInst(SelectInst &I);
144 void visitCallInst (CallInst &I);
145 void visitCallSite (CallSite CS);
146 void visitShiftInst(ShiftInst &I) { visitBinaryOperator(I); }
148 void visitMallocInst(MallocInst &I);
149 void visitAllocaInst(AllocaInst &I);
150 void visitFreeInst (FreeInst &I);
151 void visitLoadInst (LoadInst &I);
152 void visitStoreInst (StoreInst &I);
153 void visitGetElementPtrInst(GetElementPtrInst &I);
154 void visitVANextInst(VANextInst &I);
155 void visitVAArgInst (VAArgInst &I);
157 void visitInstruction(Instruction &I) {
158 std::cerr << "C Writer does not know about " << I;
162 void outputLValue(Instruction *I) {
163 Out << " " << Mang->getValueName(I) << " = ";
165 void printBranchToBlock(BasicBlock *CurBlock, BasicBlock *SuccBlock,
167 void printIndexingExpression(Value *Ptr, gep_type_iterator I,
168 gep_type_iterator E);
172 // Pass the Type* and the variable name and this prints out the variable
175 std::ostream &CWriter::printType(std::ostream &Out, const Type *Ty,
176 const std::string &NameSoFar,
178 if (Ty->isPrimitiveType())
179 switch (Ty->getPrimitiveID()) {
180 case Type::VoidTyID: return Out << "void " << NameSoFar;
181 case Type::BoolTyID: return Out << "bool " << NameSoFar;
182 case Type::UByteTyID: return Out << "unsigned char " << NameSoFar;
183 case Type::SByteTyID: return Out << "signed char " << NameSoFar;
184 case Type::UShortTyID: return Out << "unsigned short " << NameSoFar;
185 case Type::ShortTyID: return Out << "short " << NameSoFar;
186 case Type::UIntTyID: return Out << "unsigned " << NameSoFar;
187 case Type::IntTyID: return Out << "int " << NameSoFar;
188 case Type::ULongTyID: return Out << "unsigned long long " << NameSoFar;
189 case Type::LongTyID: return Out << "signed long long " << NameSoFar;
190 case Type::FloatTyID: return Out << "float " << NameSoFar;
191 case Type::DoubleTyID: return Out << "double " << NameSoFar;
193 std::cerr << "Unknown primitive type: " << Ty << "\n";
197 // Check to see if the type is named.
198 if (!IgnoreName || isa<OpaqueType>(Ty)) {
199 std::map<const Type *, std::string>::iterator I = TypeNames.find(Ty);
200 if (I != TypeNames.end()) return Out << I->second << " " << NameSoFar;
203 switch (Ty->getPrimitiveID()) {
204 case Type::FunctionTyID: {
205 const FunctionType *MTy = cast<FunctionType>(Ty);
206 std::stringstream FunctionInnards;
207 FunctionInnards << " (" << NameSoFar << ") (";
208 for (FunctionType::param_iterator I = MTy->param_begin(),
209 E = MTy->param_end(); I != E; ++I) {
210 if (I != MTy->param_begin())
211 FunctionInnards << ", ";
212 printType(FunctionInnards, *I, "");
214 if (MTy->isVarArg()) {
215 if (MTy->getNumParams())
216 FunctionInnards << ", ...";
217 } else if (!MTy->getNumParams()) {
218 FunctionInnards << "void";
220 FunctionInnards << ")";
221 std::string tstr = FunctionInnards.str();
222 printType(Out, MTy->getReturnType(), tstr);
225 case Type::StructTyID: {
226 const StructType *STy = cast<StructType>(Ty);
227 Out << NameSoFar + " {\n";
229 for (StructType::element_iterator I = STy->element_begin(),
230 E = STy->element_end(); I != E; ++I) {
232 printType(Out, *I, "field" + utostr(Idx++));
238 case Type::PointerTyID: {
239 const PointerType *PTy = cast<PointerType>(Ty);
240 std::string ptrName = "*" + NameSoFar;
242 if (isa<ArrayType>(PTy->getElementType()))
243 ptrName = "(" + ptrName + ")";
245 return printType(Out, PTy->getElementType(), ptrName);
248 case Type::ArrayTyID: {
249 const ArrayType *ATy = cast<ArrayType>(Ty);
250 unsigned NumElements = ATy->getNumElements();
251 return printType(Out, ATy->getElementType(),
252 NameSoFar + "[" + utostr(NumElements) + "]");
255 case Type::OpaqueTyID: {
256 static int Count = 0;
257 std::string TyName = "struct opaque_" + itostr(Count++);
258 assert(TypeNames.find(Ty) == TypeNames.end());
259 TypeNames[Ty] = TyName;
260 return Out << TyName << " " << NameSoFar;
263 assert(0 && "Unhandled case in getTypeProps!");
270 void CWriter::printConstantArray(ConstantArray *CPA) {
272 // As a special case, print the array as a string if it is an array of
273 // ubytes or an array of sbytes with positive values.
275 const Type *ETy = CPA->getType()->getElementType();
276 bool isString = (ETy == Type::SByteTy || ETy == Type::UByteTy);
278 // Make sure the last character is a null char, as automatically added by C
279 if (isString && (CPA->getNumOperands() == 0 ||
280 !cast<Constant>(*(CPA->op_end()-1))->isNullValue()))
285 // Keep track of whether the last number was a hexadecimal escape
286 bool LastWasHex = false;
288 // Do not include the last character, which we know is null
289 for (unsigned i = 0, e = CPA->getNumOperands()-1; i != e; ++i) {
290 unsigned char C = cast<ConstantInt>(CPA->getOperand(i))->getRawValue();
292 // Print it out literally if it is a printable character. The only thing
293 // to be careful about is when the last letter output was a hex escape
294 // code, in which case we have to be careful not to print out hex digits
295 // explicitly (the C compiler thinks it is a continuation of the previous
296 // character, sheesh...)
298 if (isprint(C) && (!LastWasHex || !isxdigit(C))) {
300 if (C == '"' || C == '\\')
307 case '\n': Out << "\\n"; break;
308 case '\t': Out << "\\t"; break;
309 case '\r': Out << "\\r"; break;
310 case '\v': Out << "\\v"; break;
311 case '\a': Out << "\\a"; break;
312 case '\"': Out << "\\\""; break;
313 case '\'': Out << "\\\'"; break;
316 Out << (char)(( C/16 < 10) ? ( C/16 +'0') : ( C/16 -10+'A'));
317 Out << (char)(((C&15) < 10) ? ((C&15)+'0') : ((C&15)-10+'A'));
326 if (CPA->getNumOperands()) {
328 printConstant(cast<Constant>(CPA->getOperand(0)));
329 for (unsigned i = 1, e = CPA->getNumOperands(); i != e; ++i) {
331 printConstant(cast<Constant>(CPA->getOperand(i)));
338 // isFPCSafeToPrint - Returns true if we may assume that CFP may be written out
339 // textually as a double (rather than as a reference to a stack-allocated
340 // variable). We decide this by converting CFP to a string and back into a
341 // double, and then checking whether the conversion results in a bit-equal
342 // double to the original value of CFP. This depends on us and the target C
343 // compiler agreeing on the conversion process (which is pretty likely since we
344 // only deal in IEEE FP).
346 bool isFPCSafeToPrint(const ConstantFP *CFP) {
349 sprintf(Buffer, "%a", CFP->getValue());
351 if (!strncmp(Buffer, "0x", 2) ||
352 !strncmp(Buffer, "-0x", 3) ||
353 !strncmp(Buffer, "+0x", 3))
354 return atof(Buffer) == CFP->getValue();
357 std::string StrVal = ftostr(CFP->getValue());
359 while (StrVal[0] == ' ')
360 StrVal.erase(StrVal.begin());
362 // Check to make sure that the stringized number is not some string like "Inf"
363 // or NaN. Check that the string matches the "[-+]?[0-9]" regex.
364 if ((StrVal[0] >= '0' && StrVal[0] <= '9') ||
365 ((StrVal[0] == '-' || StrVal[0] == '+') &&
366 (StrVal[1] >= '0' && StrVal[1] <= '9')))
367 // Reparse stringized version!
368 return atof(StrVal.c_str()) == CFP->getValue();
373 // printConstant - The LLVM Constant to C Constant converter.
374 void CWriter::printConstant(Constant *CPV) {
375 if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CPV)) {
376 switch (CE->getOpcode()) {
377 case Instruction::Cast:
379 printType(Out, CPV->getType());
381 printConstant(CE->getOperand(0));
385 case Instruction::GetElementPtr:
387 printIndexingExpression(CE->getOperand(0), gep_type_begin(CPV),
391 case Instruction::Add:
392 case Instruction::Sub:
393 case Instruction::Mul:
394 case Instruction::Div:
395 case Instruction::Rem:
396 case Instruction::SetEQ:
397 case Instruction::SetNE:
398 case Instruction::SetLT:
399 case Instruction::SetLE:
400 case Instruction::SetGT:
401 case Instruction::SetGE:
402 case Instruction::Shl:
403 case Instruction::Shr:
405 printConstant(CE->getOperand(0));
406 switch (CE->getOpcode()) {
407 case Instruction::Add: Out << " + "; break;
408 case Instruction::Sub: Out << " - "; break;
409 case Instruction::Mul: Out << " * "; break;
410 case Instruction::Div: Out << " / "; break;
411 case Instruction::Rem: Out << " % "; break;
412 case Instruction::SetEQ: Out << " == "; break;
413 case Instruction::SetNE: Out << " != "; break;
414 case Instruction::SetLT: Out << " < "; break;
415 case Instruction::SetLE: Out << " <= "; break;
416 case Instruction::SetGT: Out << " > "; break;
417 case Instruction::SetGE: Out << " >= "; break;
418 case Instruction::Shl: Out << " << "; break;
419 case Instruction::Shr: Out << " >> "; break;
420 default: assert(0 && "Illegal opcode here!");
422 printConstant(CE->getOperand(1));
427 std::cerr << "CWriter Error: Unhandled constant expression: "
433 switch (CPV->getType()->getPrimitiveID()) {
435 Out << (CPV == ConstantBool::False ? "0" : "1"); break;
436 case Type::SByteTyID:
437 case Type::ShortTyID:
438 Out << cast<ConstantSInt>(CPV)->getValue(); break;
440 if ((int)cast<ConstantSInt>(CPV)->getValue() == (int)0x80000000)
441 Out << "((int)0x80000000)"; // Handle MININT specially to avoid warning
443 Out << cast<ConstantSInt>(CPV)->getValue();
447 Out << cast<ConstantSInt>(CPV)->getValue() << "ll"; break;
449 case Type::UByteTyID:
450 case Type::UShortTyID:
451 Out << cast<ConstantUInt>(CPV)->getValue(); break;
453 Out << cast<ConstantUInt>(CPV)->getValue() << "u"; break;
454 case Type::ULongTyID:
455 Out << cast<ConstantUInt>(CPV)->getValue() << "ull"; break;
457 case Type::FloatTyID:
458 case Type::DoubleTyID: {
459 ConstantFP *FPC = cast<ConstantFP>(CPV);
460 std::map<const ConstantFP*, unsigned>::iterator I = FPConstantMap.find(FPC);
461 if (I != FPConstantMap.end()) {
462 // Because of FP precision problems we must load from a stack allocated
463 // value that holds the value in hex.
464 Out << "(*(" << (FPC->getType() == Type::FloatTy ? "float" : "double")
465 << "*)&FPConstant" << I->second << ")";
468 // Print out the constant as a floating point number.
470 sprintf(Buffer, "%a", FPC->getValue());
471 Out << Buffer << " /*" << FPC->getValue() << "*/ ";
473 Out << ftostr(FPC->getValue());
479 case Type::ArrayTyID:
480 if (isa<ConstantAggregateZero>(CPV)) {
481 const ArrayType *AT = cast<ArrayType>(CPV->getType());
483 if (AT->getNumElements()) {
485 Constant *CZ = Constant::getNullValue(AT->getElementType());
487 for (unsigned i = 1, e = AT->getNumElements(); i != e; ++i) {
494 printConstantArray(cast<ConstantArray>(CPV));
498 case Type::StructTyID:
499 if (isa<ConstantAggregateZero>(CPV)) {
500 const StructType *ST = cast<StructType>(CPV->getType());
502 if (ST->getNumElements()) {
504 printConstant(Constant::getNullValue(ST->getElementType(0)));
505 for (unsigned i = 1, e = ST->getNumElements(); i != e; ++i) {
507 printConstant(Constant::getNullValue(ST->getElementType(i)));
513 if (CPV->getNumOperands()) {
515 printConstant(cast<Constant>(CPV->getOperand(0)));
516 for (unsigned i = 1, e = CPV->getNumOperands(); i != e; ++i) {
518 printConstant(cast<Constant>(CPV->getOperand(i)));
525 case Type::PointerTyID:
526 if (isa<ConstantPointerNull>(CPV)) {
528 printType(Out, CPV->getType());
529 Out << ")/*NULL*/0)";
531 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(CPV)) {
532 writeOperand(CPR->getValue());
537 std::cerr << "Unknown constant type: " << CPV << "\n";
542 void CWriter::writeOperandInternal(Value *Operand) {
543 if (Instruction *I = dyn_cast<Instruction>(Operand))
544 if (isInlinableInst(*I) && !isDirectAlloca(I)) {
545 // Should we inline this instruction to build a tree?
552 if (Constant *CPV = dyn_cast<Constant>(Operand)) {
555 Out << Mang->getValueName(Operand);
559 void CWriter::writeOperand(Value *Operand) {
560 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
561 Out << "(&"; // Global variables are references as their addresses by llvm
563 writeOperandInternal(Operand);
565 if (isa<GlobalVariable>(Operand) || isDirectAlloca(Operand))
569 // nameAllUsedStructureTypes - If there are structure types in the module that
570 // are used but do not have names assigned to them in the symbol table yet then
571 // we assign them names now.
573 bool CWriter::nameAllUsedStructureTypes(Module &M) {
574 // Get a set of types that are used by the program...
575 std::set<const Type *> UT = FUT->getTypes();
577 // Loop over the module symbol table, removing types from UT that are already
580 SymbolTable &MST = M.getSymbolTable();
581 if (MST.find(Type::TypeTy) != MST.end())
582 for (SymbolTable::type_iterator I = MST.type_begin(Type::TypeTy),
583 E = MST.type_end(Type::TypeTy); I != E; ++I)
584 UT.erase(cast<Type>(I->second));
586 // UT now contains types that are not named. Loop over it, naming structure
589 bool Changed = false;
590 for (std::set<const Type *>::const_iterator I = UT.begin(), E = UT.end();
592 if (const StructType *ST = dyn_cast<StructType>(*I)) {
593 ((Value*)ST)->setName("unnamed", &MST);
599 // generateCompilerSpecificCode - This is where we add conditional compilation
600 // directives to cater to specific compilers as need be.
602 static void generateCompilerSpecificCode(std::ostream& Out) {
603 // Alloca is hard to get, and we don't want to include stdlib.h here...
604 Out << "/* get a declaration for alloca */\n"
606 << "extern void *__builtin_alloca(unsigned long);\n"
607 << "#define alloca(x) __builtin_alloca(x)\n"
609 << "#ifndef __FreeBSD__\n"
610 << "#include <alloca.h>\n"
614 // We output GCC specific attributes to preserve 'linkonce'ness on globals.
615 // If we aren't being compiled with GCC, just drop these attributes.
616 Out << "#ifndef __GNUC__ /* Can only support \"linkonce\" vars with GCC */\n"
617 << "#define __attribute__(X)\n"
621 // At some point, we should support "external weak" vs. "weak" linkages.
622 // On Mac OS X, "external weak" is spelled "__attribute__((weak_import))".
623 Out << "#if defined(__GNUC__) && defined(__APPLE_CC__)\n"
624 << "#define __EXTERNAL_WEAK__ __attribute__((weak_import))\n"
625 << "#elif defined(__GNUC__)\n"
626 << "#define __EXTERNAL_WEAK__ __attribute__((weak))\n"
628 << "#define __EXTERNAL_WEAK__\n"
632 // For now, turn off the weak linkage attribute on Mac OS X. (See above.)
633 Out << "#if defined(__GNUC__) && defined(__APPLE_CC__)\n"
634 << "#define __ATTRIBUTE_WEAK__\n"
635 << "#elif defined(__GNUC__)\n"
636 << "#define __ATTRIBUTE_WEAK__ __attribute__((weak))\n"
638 << "#define __ATTRIBUTE_WEAK__\n"
642 bool CWriter::doInitialization(Module &M) {
645 FUT = &getAnalysis<FindUsedTypes>();
647 // Ensure that all structure types have names...
648 bool Changed = nameAllUsedStructureTypes(M);
649 Mang = new Mangler(M);
651 // get declaration for alloca
652 Out << "/* Provide Declarations */\n";
653 Out << "#include <stdarg.h>\n"; // Varargs support
654 Out << "#include <setjmp.h>\n"; // Unwind support
655 generateCompilerSpecificCode(Out);
657 // Provide a definition for `bool' if not compiling with a C++ compiler.
659 << "#ifndef __cplusplus\ntypedef unsigned char bool;\n#endif\n"
661 << "\n\n/* Support for floating point constants */\n"
662 << "typedef unsigned long long ConstantDoubleTy;\n"
663 << "typedef unsigned int ConstantFloatTy;\n"
665 << "\n\n/* Global Declarations */\n";
667 // First output all the declarations for the program, because C requires
668 // Functions & globals to be declared before they are used.
671 // Loop over the symbol table, emitting all named constants...
672 printSymbolTable(M.getSymbolTable());
674 // Global variable declarations...
676 Out << "\n/* External Global Variable Declarations */\n";
677 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I) {
678 if (I->hasExternalLinkage()) {
680 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
686 // Function declarations
688 Out << "\n/* Function Declarations */\n";
689 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
690 // Don't print declarations for intrinsic functions.
691 if (!I->getIntrinsicID()) {
692 printFunctionSignature(I, true);
693 if (I->hasWeakLinkage()) Out << " __ATTRIBUTE_WEAK__";
694 if (I->hasLinkOnceLinkage()) Out << " __ATTRIBUTE_WEAK__";
700 // Output the global variable declarations
702 Out << "\n\n/* Global Variable Declarations */\n";
703 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
704 if (!I->isExternal()) {
706 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
708 if (I->hasLinkOnceLinkage())
709 Out << " __attribute__((common))";
710 else if (I->hasWeakLinkage())
711 Out << " __ATTRIBUTE_WEAK__";
716 // Output the global variable definitions and contents...
718 Out << "\n\n/* Global Variable Definitions and Initialization */\n";
719 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
720 if (!I->isExternal()) {
721 if (I->hasInternalLinkage())
723 printType(Out, I->getType()->getElementType(), Mang->getValueName(I));
724 if (I->hasLinkOnceLinkage())
725 Out << " __attribute__((common))";
726 else if (I->hasWeakLinkage())
727 Out << " __ATTRIBUTE_WEAK__";
729 // If the initializer is not null, emit the initializer. If it is null,
730 // we try to avoid emitting large amounts of zeros. The problem with
731 // this, however, occurs when the variable has weak linkage. In this
732 // case, the assembler will complain about the variable being both weak
733 // and common, so we disable this optimization.
734 if (!I->getInitializer()->isNullValue()) {
736 writeOperand(I->getInitializer());
737 } else if (I->hasWeakLinkage()) {
738 // We have to specify an initializer, but it doesn't have to be
739 // complete. If the value is an aggregate, print out { 0 }, and let
740 // the compiler figure out the rest of the zeros.
742 if (isa<StructType>(I->getInitializer()->getType()) ||
743 isa<ArrayType>(I->getInitializer()->getType())) {
746 // Just print it out normally.
747 writeOperand(I->getInitializer());
754 // Output all floating point constants that cannot be printed accurately...
755 printFloatingPointConstants(M);
758 Out << "\n\n/* Function Bodies */\n";
763 /// Output all floating point constants that cannot be printed accurately...
764 void CWriter::printFloatingPointConstants(Module &M) {
767 unsigned long long U;
775 // Scan the module for floating point constants. If any FP constant is used
776 // in the function, we want to redirect it here so that we do not depend on
777 // the precision of the printed form, unless the printed form preserves
780 unsigned FPCounter = 0;
781 for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F)
782 for (constant_iterator I = constant_begin(F), E = constant_end(F);
784 if (const ConstantFP *FPC = dyn_cast<ConstantFP>(*I))
785 if (!isFPCSafeToPrint(FPC) && // Do not put in FPConstantMap if safe.
786 !FPConstantMap.count(FPC)) {
787 double Val = FPC->getValue();
789 FPConstantMap[FPC] = FPCounter; // Number the FP constants
791 if (FPC->getType() == Type::DoubleTy) {
793 Out << "static const ConstantDoubleTy FPConstant" << FPCounter++
794 << " = 0x" << std::hex << DBLUnion.U << std::dec
795 << "ULL; /* " << Val << " */\n";
796 } else if (FPC->getType() == Type::FloatTy) {
798 Out << "static const ConstantFloatTy FPConstant" << FPCounter++
799 << " = 0x" << std::hex << FLTUnion.U << std::dec
800 << "U; /* " << Val << " */\n";
802 assert(0 && "Unknown float type!");
809 /// printSymbolTable - Run through symbol table looking for type names. If a
810 /// type name is found, emit it's declaration...
812 void CWriter::printSymbolTable(const SymbolTable &ST) {
813 // If there are no type names, exit early.
814 if (ST.find(Type::TypeTy) == ST.end())
817 // We are only interested in the type plane of the symbol table...
818 SymbolTable::type_const_iterator I = ST.type_begin(Type::TypeTy);
819 SymbolTable::type_const_iterator End = ST.type_end(Type::TypeTy);
821 // Print out forward declarations for structure types before anything else!
822 Out << "/* Structure forward decls */\n";
823 for (; I != End; ++I)
824 if (const Type *STy = dyn_cast<StructType>(I->second))
825 // Only print out used types!
826 if (FUT->getTypes().count(STy)) {
827 std::string Name = "struct l_" + Mangler::makeNameProper(I->first);
828 Out << Name << ";\n";
829 TypeNames.insert(std::make_pair(STy, Name));
834 // Now we can print out typedefs...
835 Out << "/* Typedefs */\n";
836 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
837 // Only print out used types!
838 if (FUT->getTypes().count(cast<Type>(I->second))) {
839 const Type *Ty = cast<Type>(I->second);
840 std::string Name = "l_" + Mangler::makeNameProper(I->first);
842 printType(Out, Ty, Name);
848 // Keep track of which structures have been printed so far...
849 std::set<const StructType *> StructPrinted;
851 // Loop over all structures then push them into the stack so they are
852 // printed in the correct order.
854 Out << "/* Structure contents */\n";
855 for (I = ST.type_begin(Type::TypeTy); I != End; ++I)
856 if (const StructType *STy = dyn_cast<StructType>(I->second))
857 // Only print out used types!
858 if (FUT->getTypes().count(STy))
859 printContainedStructs(STy, StructPrinted);
862 // Push the struct onto the stack and recursively push all structs
863 // this one depends on.
864 void CWriter::printContainedStructs(const Type *Ty,
865 std::set<const StructType*> &StructPrinted){
866 if (const StructType *STy = dyn_cast<StructType>(Ty)) {
867 //Check to see if we have already printed this struct
868 if (StructPrinted.count(STy) == 0) {
869 // Print all contained types first...
870 for (StructType::element_iterator I = STy->element_begin(),
871 E = STy->element_end(); I != E; ++I) {
872 const Type *Ty1 = I->get();
873 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
874 printContainedStructs(*I, StructPrinted);
877 //Print structure type out..
878 StructPrinted.insert(STy);
879 std::string Name = TypeNames[STy];
880 printType(Out, STy, Name, true);
884 // If it is an array, check contained types and continue
885 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)){
886 const Type *Ty1 = ATy->getElementType();
887 if (isa<StructType>(Ty1) || isa<ArrayType>(Ty1))
888 printContainedStructs(Ty1, StructPrinted);
893 void CWriter::printFunctionSignature(const Function *F, bool Prototype) {
894 if (F->hasInternalLinkage()) Out << "static ";
896 // Loop over the arguments, printing them...
897 const FunctionType *FT = cast<FunctionType>(F->getFunctionType());
899 std::stringstream FunctionInnards;
901 // Print out the name...
902 FunctionInnards << Mang->getValueName(F) << "(";
904 if (!F->isExternal()) {
907 if (F->abegin()->hasName() || !Prototype)
908 ArgName = Mang->getValueName(F->abegin());
909 printType(FunctionInnards, F->afront().getType(), ArgName);
910 for (Function::const_aiterator I = ++F->abegin(), E = F->aend();
912 FunctionInnards << ", ";
913 if (I->hasName() || !Prototype)
914 ArgName = Mang->getValueName(I);
917 printType(FunctionInnards, I->getType(), ArgName);
921 // Loop over the arguments, printing them...
922 for (FunctionType::param_iterator I = FT->param_begin(),
923 E = FT->param_end(); I != E; ++I) {
924 if (I != FT->param_begin()) FunctionInnards << ", ";
925 printType(FunctionInnards, *I);
929 // Finish printing arguments... if this is a vararg function, print the ...,
930 // unless there are no known types, in which case, we just emit ().
932 if (FT->isVarArg() && FT->getNumParams()) {
933 if (FT->getNumParams()) FunctionInnards << ", ";
934 FunctionInnards << "..."; // Output varargs portion of signature!
935 } else if (!FT->isVarArg() && FT->getNumParams() == 0) {
936 FunctionInnards << "void"; // ret() -> ret(void) in C.
938 FunctionInnards << ")";
939 // Print out the return type and the entire signature for that matter
940 printType(Out, F->getReturnType(), FunctionInnards.str());
943 void CWriter::printFunction(Function &F) {
944 printFunctionSignature(&F, false);
947 // print local variable information for the function
948 for (inst_iterator I = inst_begin(&F), E = inst_end(&F); I != E; ++I)
949 if (const AllocaInst *AI = isDirectAlloca(*I)) {
951 printType(Out, AI->getAllocatedType(), Mang->getValueName(AI));
952 Out << "; /* Address exposed local */\n";
953 } else if ((*I)->getType() != Type::VoidTy && !isInlinableInst(**I)) {
955 printType(Out, (*I)->getType(), Mang->getValueName(*I));
958 if (isa<PHINode>(*I)) { // Print out PHI node temporaries as well...
960 printType(Out, (*I)->getType(),
961 Mang->getValueName(*I)+"__PHI_TEMPORARY");
968 // print the basic blocks
969 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
970 BasicBlock *Prev = BB->getPrev();
972 // Don't print the label for the basic block if there are no uses, or if the
973 // only terminator use is the predecessor basic block's terminator. We have
974 // to scan the use list because PHI nodes use basic blocks too but do not
975 // require a label to be generated.
977 bool NeedsLabel = false;
978 for (Value::use_iterator UI = BB->use_begin(), UE = BB->use_end();
980 if (TerminatorInst *TI = dyn_cast<TerminatorInst>(*UI))
981 if (TI != Prev->getTerminator() ||
982 isa<SwitchInst>(Prev->getTerminator()) ||
983 isa<InvokeInst>(Prev->getTerminator())) {
988 if (NeedsLabel) Out << Mang->getValueName(BB) << ":\n";
990 // Output all of the instructions in the basic block...
991 for (BasicBlock::iterator II = BB->begin(), E = --BB->end(); II != E; ++II){
992 if (!isInlinableInst(*II) && !isDirectAlloca(II)) {
993 if (II->getType() != Type::VoidTy)
1002 // Don't emit prefix or suffix for the terminator...
1003 visit(*BB->getTerminator());
1009 // Specific Instruction type classes... note that all of the casts are
1010 // necessary because we use the instruction classes as opaque types...
1012 void CWriter::visitReturnInst(ReturnInst &I) {
1013 // Don't output a void return if this is the last basic block in the function
1014 if (I.getNumOperands() == 0 &&
1015 &*--I.getParent()->getParent()->end() == I.getParent() &&
1016 !I.getParent()->size() == 1) {
1021 if (I.getNumOperands()) {
1023 writeOperand(I.getOperand(0));
1028 void CWriter::visitSwitchInst(SwitchInst &SI) {
1030 writeOperand(SI.getOperand(0));
1031 Out << ") {\n default:\n";
1032 printBranchToBlock(SI.getParent(), SI.getDefaultDest(), 2);
1034 for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2) {
1036 writeOperand(SI.getOperand(i));
1038 BasicBlock *Succ = cast<BasicBlock>(SI.getOperand(i+1));
1039 printBranchToBlock(SI.getParent(), Succ, 2);
1040 if (Succ == SI.getParent()->getNext())
1046 void CWriter::visitInvokeInst(InvokeInst &II) {
1047 assert(0 && "Lowerinvoke pass didn't work!");
1051 void CWriter::visitUnwindInst(UnwindInst &I) {
1052 assert(0 && "Lowerinvoke pass didn't work!");
1055 bool isGotoCodeNecessary(BasicBlock *From, BasicBlock *To) {
1056 // If PHI nodes need copies, we need the copy code...
1057 if (isa<PHINode>(To->front()) ||
1058 From->getNext() != To) // Not directly successor, need goto
1061 // Otherwise we don't need the code.
1065 void CWriter::printBranchToBlock(BasicBlock *CurBB, BasicBlock *Succ,
1067 for (BasicBlock::iterator I = Succ->begin();
1068 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
1069 // now we have to do the printing
1070 Out << std::string(Indent, ' ');
1071 Out << " " << Mang->getValueName(I) << "__PHI_TEMPORARY = ";
1072 writeOperand(PN->getIncomingValue(PN->getBasicBlockIndex(CurBB)));
1073 Out << "; /* for PHI node */\n";
1076 if (CurBB->getNext() != Succ ||
1077 isa<InvokeInst>(CurBB->getTerminator()) ||
1078 isa<SwitchInst>(CurBB->getTerminator())) {
1079 Out << std::string(Indent, ' ') << " goto ";
1085 // Branch instruction printing - Avoid printing out a branch to a basic block
1086 // that immediately succeeds the current one.
1088 void CWriter::visitBranchInst(BranchInst &I) {
1089 if (I.isConditional()) {
1090 if (isGotoCodeNecessary(I.getParent(), I.getSuccessor(0))) {
1092 writeOperand(I.getCondition());
1095 printBranchToBlock(I.getParent(), I.getSuccessor(0), 2);
1097 if (isGotoCodeNecessary(I.getParent(), I.getSuccessor(1))) {
1098 Out << " } else {\n";
1099 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1102 // First goto not necessary, assume second one is...
1104 writeOperand(I.getCondition());
1107 printBranchToBlock(I.getParent(), I.getSuccessor(1), 2);
1112 printBranchToBlock(I.getParent(), I.getSuccessor(0), 0);
1117 // PHI nodes get copied into temporary values at the end of predecessor basic
1118 // blocks. We now need to copy these temporary values into the REAL value for
1120 void CWriter::visitPHINode(PHINode &I) {
1122 Out << "__PHI_TEMPORARY";
1126 void CWriter::visitBinaryOperator(Instruction &I) {
1127 // binary instructions, shift instructions, setCond instructions.
1128 assert(!isa<PointerType>(I.getType()));
1130 // We must cast the results of binary operations which might be promoted.
1131 bool needsCast = false;
1132 if ((I.getType() == Type::UByteTy) || (I.getType() == Type::SByteTy)
1133 || (I.getType() == Type::UShortTy) || (I.getType() == Type::ShortTy)
1134 || (I.getType() == Type::FloatTy)) {
1137 printType(Out, I.getType());
1141 writeOperand(I.getOperand(0));
1143 switch (I.getOpcode()) {
1144 case Instruction::Add: Out << " + "; break;
1145 case Instruction::Sub: Out << " - "; break;
1146 case Instruction::Mul: Out << "*"; break;
1147 case Instruction::Div: Out << "/"; break;
1148 case Instruction::Rem: Out << "%"; break;
1149 case Instruction::And: Out << " & "; break;
1150 case Instruction::Or: Out << " | "; break;
1151 case Instruction::Xor: Out << " ^ "; break;
1152 case Instruction::SetEQ: Out << " == "; break;
1153 case Instruction::SetNE: Out << " != "; break;
1154 case Instruction::SetLE: Out << " <= "; break;
1155 case Instruction::SetGE: Out << " >= "; break;
1156 case Instruction::SetLT: Out << " < "; break;
1157 case Instruction::SetGT: Out << " > "; break;
1158 case Instruction::Shl : Out << " << "; break;
1159 case Instruction::Shr : Out << " >> "; break;
1160 default: std::cerr << "Invalid operator type!" << I; abort();
1163 writeOperand(I.getOperand(1));
1170 void CWriter::visitCastInst(CastInst &I) {
1171 if (I.getType() == Type::BoolTy) {
1173 writeOperand(I.getOperand(0));
1178 printType(Out, I.getType());
1180 if (isa<PointerType>(I.getType())&&I.getOperand(0)->getType()->isIntegral() ||
1181 isa<PointerType>(I.getOperand(0)->getType())&&I.getType()->isIntegral()) {
1182 // Avoid "cast to pointer from integer of different size" warnings
1186 writeOperand(I.getOperand(0));
1189 void CWriter::visitSelectInst(SelectInst &I) {
1191 writeOperand(I.getCondition());
1193 writeOperand(I.getTrueValue());
1195 writeOperand(I.getFalseValue());
1200 void CWriter::lowerIntrinsics(Module &M) {
1201 for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F)
1202 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
1203 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; )
1204 if (CallInst *CI = dyn_cast<CallInst>(I++))
1205 if (Function *F = CI->getCalledFunction())
1206 switch (F->getIntrinsicID()) {
1207 case Intrinsic::not_intrinsic:
1208 case Intrinsic::vastart:
1209 case Intrinsic::vacopy:
1210 case Intrinsic::vaend:
1211 case Intrinsic::returnaddress:
1212 case Intrinsic::frameaddress:
1213 case Intrinsic::setjmp:
1214 case Intrinsic::longjmp:
1215 // We directly implement these intrinsics
1218 // All other intrinsic calls we must lower.
1219 Instruction *Before = CI->getPrev();
1220 IL.LowerIntrinsicCall(CI);
1221 if (Before) { // Move iterator to instruction after call
1231 void CWriter::visitCallInst(CallInst &I) {
1232 // Handle intrinsic function calls first...
1233 if (Function *F = I.getCalledFunction())
1234 if (Intrinsic::ID ID = (Intrinsic::ID)F->getIntrinsicID()) {
1236 default: assert(0 && "Unknown LLVM intrinsic!");
1237 case Intrinsic::vastart:
1240 Out << "va_start(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1241 // Output the last argument to the enclosing function...
1242 if (I.getParent()->getParent()->aempty()) {
1243 std::cerr << "The C backend does not currently support zero "
1244 << "argument varargs functions, such as '"
1245 << I.getParent()->getParent()->getName() << "'!\n";
1248 writeOperand(&I.getParent()->getParent()->aback());
1251 case Intrinsic::vaend:
1252 Out << "va_end(*(va_list*)&";
1253 writeOperand(I.getOperand(1));
1256 case Intrinsic::vacopy:
1258 Out << "va_copy(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1259 Out << "*(va_list*)&";
1260 writeOperand(I.getOperand(1));
1263 case Intrinsic::returnaddress:
1264 Out << "__builtin_return_address(";
1265 writeOperand(I.getOperand(1));
1268 case Intrinsic::frameaddress:
1269 Out << "__builtin_frame_address(";
1270 writeOperand(I.getOperand(1));
1273 case Intrinsic::setjmp:
1274 Out << "setjmp(*(jmp_buf*)";
1275 writeOperand(I.getOperand(1));
1278 case Intrinsic::longjmp:
1279 Out << "longjmp(*(jmp_buf*)";
1280 writeOperand(I.getOperand(1));
1282 writeOperand(I.getOperand(2));
1290 void CWriter::visitCallSite(CallSite CS) {
1291 const PointerType *PTy = cast<PointerType>(CS.getCalledValue()->getType());
1292 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType());
1293 const Type *RetTy = FTy->getReturnType();
1295 writeOperand(CS.getCalledValue());
1298 if (CS.arg_begin() != CS.arg_end()) {
1299 CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
1302 for (++AI; AI != AE; ++AI) {
1310 void CWriter::visitMallocInst(MallocInst &I) {
1311 assert(0 && "lowerallocations pass didn't work!");
1314 void CWriter::visitAllocaInst(AllocaInst &I) {
1316 printType(Out, I.getType());
1317 Out << ") alloca(sizeof(";
1318 printType(Out, I.getType()->getElementType());
1320 if (I.isArrayAllocation()) {
1322 writeOperand(I.getOperand(0));
1327 void CWriter::visitFreeInst(FreeInst &I) {
1328 assert(0 && "lowerallocations pass didn't work!");
1331 void CWriter::printIndexingExpression(Value *Ptr, gep_type_iterator I,
1332 gep_type_iterator E) {
1333 bool HasImplicitAddress = false;
1334 // If accessing a global value with no indexing, avoid *(&GV) syndrome
1335 if (GlobalValue *V = dyn_cast<GlobalValue>(Ptr)) {
1336 HasImplicitAddress = true;
1337 } else if (ConstantPointerRef *CPR = dyn_cast<ConstantPointerRef>(Ptr)) {
1338 HasImplicitAddress = true;
1339 Ptr = CPR->getValue(); // Get to the global...
1340 } else if (isDirectAlloca(Ptr)) {
1341 HasImplicitAddress = true;
1345 if (!HasImplicitAddress)
1346 Out << "*"; // Implicit zero first argument: '*x' is equivalent to 'x[0]'
1348 writeOperandInternal(Ptr);
1352 const Constant *CI = dyn_cast<Constant>(I.getOperand());
1353 if (HasImplicitAddress && (!CI || !CI->isNullValue()))
1356 writeOperandInternal(Ptr);
1358 if (HasImplicitAddress && (!CI || !CI->isNullValue())) {
1360 HasImplicitAddress = false; // HIA is only true if we haven't addressed yet
1363 assert(!HasImplicitAddress || (CI && CI->isNullValue()) &&
1364 "Can only have implicit address with direct accessing");
1366 if (HasImplicitAddress) {
1368 } else if (CI && CI->isNullValue()) {
1369 gep_type_iterator TmpI = I; ++TmpI;
1371 // Print out the -> operator if possible...
1372 if (TmpI != E && isa<StructType>(*TmpI)) {
1373 Out << (HasImplicitAddress ? "." : "->");
1374 Out << "field" << cast<ConstantUInt>(TmpI.getOperand())->getValue();
1380 if (isa<StructType>(*I)) {
1381 Out << ".field" << cast<ConstantUInt>(I.getOperand())->getValue();
1384 writeOperand(I.getOperand());
1389 void CWriter::visitLoadInst(LoadInst &I) {
1391 writeOperand(I.getOperand(0));
1394 void CWriter::visitStoreInst(StoreInst &I) {
1396 writeOperand(I.getPointerOperand());
1398 writeOperand(I.getOperand(0));
1401 void CWriter::visitGetElementPtrInst(GetElementPtrInst &I) {
1403 printIndexingExpression(I.getPointerOperand(), gep_type_begin(I),
1407 void CWriter::visitVANextInst(VANextInst &I) {
1408 Out << Mang->getValueName(I.getOperand(0));
1409 Out << "; va_arg(*(va_list*)&" << Mang->getValueName(&I) << ", ";
1410 printType(Out, I.getArgType());
1414 void CWriter::visitVAArgInst(VAArgInst &I) {
1416 Out << "{ va_list Tmp; va_copy(Tmp, *(va_list*)&";
1417 writeOperand(I.getOperand(0));
1418 Out << ");\n " << Mang->getValueName(&I) << " = va_arg(Tmp, ";
1419 printType(Out, I.getType());
1420 Out << ");\n va_end(Tmp); }";
1423 //===----------------------------------------------------------------------===//
1424 // External Interface declaration
1425 //===----------------------------------------------------------------------===//
1427 bool CTargetMachine::addPassesToEmitAssembly(PassManager &PM, std::ostream &o) {
1428 PM.add(createLowerAllocationsPass());
1429 PM.add(createLowerInvokePass());
1430 PM.add(new CWriter(o, getIntrinsicLowering()));
1434 TargetMachine *llvm::allocateCTargetMachine(const Module &M,
1435 IntrinsicLowering *IL) {
1436 return new CTargetMachine(M, IL);