X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=tools%2Flto%2FLTOModule.cpp;h=5a388cbad6f777e0282b9bdc845aa3e4eb34a64d;hb=f010c464a11444733ec67e31aace8bcebeaf2588;hp=d386bacdfc9326c8bb99755569d0276d9d2feede;hpb=cbb170d057aa6692b19f577b1e09a6c1c7a26969;p=oota-llvm.git diff --git a/tools/lto/LTOModule.cpp b/tools/lto/LTOModule.cpp index d386bacdfc9..5a388cbad6f 100644 --- a/tools/lto/LTOModule.cpp +++ b/tools/lto/LTOModule.cpp @@ -4,527 +4,899 @@ // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. -// +// //===----------------------------------------------------------------------===// // -// This file implements the Link Time Optimization library. This library is +// This file implements the Link Time Optimization library. This library is // intended to be used by linker to optimize code at link time. // //===----------------------------------------------------------------------===// #include "LTOModule.h" - -#include "llvm/Constants.h" -#include "llvm/LLVMContext.h" -#include "llvm/Module.h" #include "llvm/ADT/OwningPtr.h" #include "llvm/ADT/Triple.h" #include "llvm/Bitcode/ReaderWriter.h" -#include "llvm/Support/SystemUtils.h" +#include "llvm/Constants.h" +#include "llvm/LLVMContext.h" +#include "llvm/MC/MCExpr.h" +#include "llvm/MC/MCInst.h" +#include "llvm/MC/MCParser/MCAsmParser.h" +#include "llvm/MC/MCStreamer.h" +#include "llvm/MC/MCSubtargetInfo.h" +#include "llvm/MC/MCSymbol.h" +#include "llvm/MC/MCTargetAsmParser.h" +#include "llvm/MC/SubtargetFeature.h" +#include "llvm/Module.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Support/Host.h" #include "llvm/Support/MemoryBuffer.h" -#include "llvm/Support/MathExtras.h" -#include "llvm/System/Host.h" -#include "llvm/System/Path.h" -#include "llvm/System/Process.h" -#include "llvm/Target/Mangler.h" -#include "llvm/Target/SubtargetFeature.h" -#include "llvm/MC/MCAsmInfo.h" -#include "llvm/MC/MCContext.h" -#include "llvm/Target/TargetMachine.h" -#include "llvm/Target/TargetRegistry.h" -#include "llvm/Target/TargetSelect.h" - +#include "llvm/Support/Path.h" +#include "llvm/Support/SourceMgr.h" +#include "llvm/Support/TargetRegistry.h" +#include "llvm/Support/TargetSelect.h" +#include "llvm/Support/system_error.h" +#include "llvm/Target/TargetRegisterInfo.h" using namespace llvm; -bool LTOModule::isBitcodeFile(const void* mem, size_t length) -{ - return llvm::sys::IdentifyFileType((char*)mem, length) - == llvm::sys::Bitcode_FileType; +static cl::opt +EnableFPMAD("enable-fp-mad", + cl::desc("Enable less precise MAD instructions to be generated"), + cl::init(false)); + +static cl::opt +DisableFPElim("disable-fp-elim", + cl::desc("Disable frame pointer elimination optimization"), + cl::init(false)); + +static cl::opt +DisableFPElimNonLeaf("disable-non-leaf-fp-elim", + cl::desc("Disable frame pointer elimination optimization for non-leaf funcs"), + cl::init(false)); + +static cl::opt +EnableUnsafeFPMath("enable-unsafe-fp-math", + cl::desc("Enable optimizations that may decrease FP precision"), + cl::init(false)); + +static cl::opt +EnableNoInfsFPMath("enable-no-infs-fp-math", + cl::desc("Enable FP math optimizations that assume no +-Infs"), + cl::init(false)); + +static cl::opt +EnableNoNaNsFPMath("enable-no-nans-fp-math", + cl::desc("Enable FP math optimizations that assume no NaNs"), + cl::init(false)); + +static cl::opt +EnableHonorSignDependentRoundingFPMath("enable-sign-dependent-rounding-fp-math", + cl::Hidden, + cl::desc("Force codegen to assume rounding mode can change dynamically"), + cl::init(false)); + +static cl::opt +GenerateSoftFloatCalls("soft-float", + cl::desc("Generate software floating point library calls"), + cl::init(false)); + +static cl::opt +FloatABIForCalls("float-abi", + cl::desc("Choose float ABI type"), + cl::init(FloatABI::Default), + cl::values( + clEnumValN(FloatABI::Default, "default", + "Target default float ABI type"), + clEnumValN(FloatABI::Soft, "soft", + "Soft float ABI (implied by -soft-float)"), + clEnumValN(FloatABI::Hard, "hard", + "Hard float ABI (uses FP registers)"), + clEnumValEnd)); + +static cl::opt +FuseFPOps("fp-contract", + cl::desc("Enable aggresive formation of fused FP ops"), + cl::init(FPOpFusion::Standard), + cl::values( + clEnumValN(FPOpFusion::Fast, "fast", + "Fuse FP ops whenever profitable"), + clEnumValN(FPOpFusion::Standard, "on", + "Only fuse 'blessed' FP ops."), + clEnumValN(FPOpFusion::Strict, "off", + "Only fuse FP ops when the result won't be effected."), + clEnumValEnd)); + +static cl::opt +DontPlaceZerosInBSS("nozero-initialized-in-bss", + cl::desc("Don't place zero-initialized symbols into bss section"), + cl::init(false)); + +static cl::opt +EnableGuaranteedTailCallOpt("tailcallopt", + cl::desc("Turn fastcc calls into tail calls by (potentially) changing ABI."), + cl::init(false)); + +static cl::opt +DisableTailCalls("disable-tail-calls", + cl::desc("Never emit tail calls"), + cl::init(false)); + +static cl::opt +OverrideStackAlignment("stack-alignment", + cl::desc("Override default stack alignment"), + cl::init(0)); + +static cl::opt +EnableRealignStack("realign-stack", + cl::desc("Realign stack if needed"), + cl::init(true)); + +static cl::opt +TrapFuncName("trap-func", cl::Hidden, + cl::desc("Emit a call to trap function rather than a trap instruction"), + cl::init("")); + +static cl::opt +EnablePIE("enable-pie", + cl::desc("Assume the creation of a position independent executable."), + cl::init(false)); + +static cl::opt +SegmentedStacks("segmented-stacks", + cl::desc("Use segmented stacks if possible."), + cl::init(false)); + +static cl::opt +UseInitArray("use-init-array", + cl::desc("Use .init_array instead of .ctors."), + cl::init(false)); + +static cl::opt +SSPBufferSize("stack-protector-buffer-size", cl::init(8), + cl::desc("Lower bound for a buffer to be considered for " + "stack protection")); + +LTOModule::LTOModule(llvm::Module *m, llvm::TargetMachine *t) + : _module(m), _target(t), + _context(*_target->getMCAsmInfo(), *_target->getRegisterInfo(), NULL), + _mangler(_context, *_target->getDataLayout()) {} + +/// isBitcodeFile - Returns 'true' if the file (or memory contents) is LLVM +/// bitcode. +bool LTOModule::isBitcodeFile(const void *mem, size_t length) { + return llvm::sys::IdentifyFileType((const char*)mem, length) + == llvm::sys::Bitcode_FileType; } -bool LTOModule::isBitcodeFile(const char* path) -{ - return llvm::sys::Path(path).isBitcodeFile(); +bool LTOModule::isBitcodeFile(const char *path) { + return llvm::sys::Path(path).isBitcodeFile(); } -bool LTOModule::isBitcodeFileForTarget(const void* mem, size_t length, - const char* triplePrefix) -{ - MemoryBuffer* buffer = makeBuffer(mem, length); - if (!buffer) - return false; - return isTargetMatch(buffer, triplePrefix); +/// isBitcodeFileForTarget - Returns 'true' if the file (or memory contents) is +/// LLVM bitcode for the specified triple. +bool LTOModule::isBitcodeFileForTarget(const void *mem, size_t length, + const char *triplePrefix) { + MemoryBuffer *buffer = makeBuffer(mem, length); + if (!buffer) + return false; + return isTargetMatch(buffer, triplePrefix); } - -bool LTOModule::isBitcodeFileForTarget(const char* path, - const char* triplePrefix) -{ - MemoryBuffer *buffer = MemoryBuffer::getFile(path); - if (buffer == NULL) - return false; - return isTargetMatch(buffer, triplePrefix); +bool LTOModule::isBitcodeFileForTarget(const char *path, + const char *triplePrefix) { + OwningPtr buffer; + if (MemoryBuffer::getFile(path, buffer)) + return false; + return isTargetMatch(buffer.take(), triplePrefix); } -// takes ownership of buffer -bool LTOModule::isTargetMatch(MemoryBuffer* buffer, const char* triplePrefix) -{ - OwningPtr m(getLazyBitcodeModule(buffer, getGlobalContext())); - // on success, m owns buffer and both are deleted at end of this method - if (!m) { - delete buffer; - return false; - } - std::string actualTarget = m->getTargetTriple(); - return (strncmp(actualTarget.c_str(), triplePrefix, - strlen(triplePrefix)) == 0); +/// isTargetMatch - Returns 'true' if the memory buffer is for the specified +/// target triple. +bool LTOModule::isTargetMatch(MemoryBuffer *buffer, const char *triplePrefix) { + std::string Triple = getBitcodeTargetTriple(buffer, getGlobalContext()); + delete buffer; + return strncmp(Triple.c_str(), triplePrefix, strlen(triplePrefix)) == 0; } - -LTOModule::LTOModule(Module* m, TargetMachine* t) - : _module(m), _target(t), _symbolsParsed(false) -{ +/// makeLTOModule - Create an LTOModule. N.B. These methods take ownership of +/// the buffer. +LTOModule *LTOModule::makeLTOModule(const char *path, std::string &errMsg) { + OwningPtr buffer; + if (error_code ec = MemoryBuffer::getFile(path, buffer)) { + errMsg = ec.message(); + return NULL; + } + return makeLTOModule(buffer.take(), errMsg); } -LTOModule* LTOModule::makeLTOModule(const char* path, - std::string& errMsg) -{ - OwningPtr buffer(MemoryBuffer::getFile(path, &errMsg)); - if (!buffer) - return NULL; - return makeLTOModule(buffer.get(), errMsg); +LTOModule *LTOModule::makeLTOModule(int fd, const char *path, + size_t size, std::string &errMsg) { + return makeLTOModule(fd, path, size, size, 0, errMsg); } -/// makeBuffer - create a MemoryBuffer from a memory range. -/// MemoryBuffer requires the byte past end of the buffer to be a zero. -/// We might get lucky and already be that way, otherwise make a copy. -/// Also if next byte is on a different page, don't assume it is readable. -MemoryBuffer* LTOModule::makeBuffer(const void* mem, size_t length) -{ - const char *startPtr = (char*)mem; - const char *endPtr = startPtr+length; - if (((uintptr_t)endPtr & (sys::Process::GetPageSize()-1)) == 0 || - *endPtr != 0) - return MemoryBuffer::getMemBufferCopy(StringRef(startPtr, length)); - - return MemoryBuffer::getMemBuffer(StringRef(startPtr, length)); +LTOModule *LTOModule::makeLTOModule(int fd, const char *path, + size_t file_size, + size_t map_size, + off_t offset, + std::string &errMsg) { + OwningPtr buffer; + if (error_code ec = MemoryBuffer::getOpenFile(fd, path, buffer, file_size, + map_size, offset, false)) { + errMsg = ec.message(); + return NULL; + } + return makeLTOModule(buffer.take(), errMsg); } - -LTOModule* LTOModule::makeLTOModule(const void* mem, size_t length, - std::string& errMsg) -{ - OwningPtr buffer(makeBuffer(mem, length)); - if (!buffer) - return NULL; - return makeLTOModule(buffer.get(), errMsg); +LTOModule *LTOModule::makeLTOModule(const void *mem, size_t length, + std::string &errMsg) { + OwningPtr buffer(makeBuffer(mem, length)); + if (!buffer) + return NULL; + return makeLTOModule(buffer.take(), errMsg); } -LTOModule* LTOModule::makeLTOModule(MemoryBuffer* buffer, - std::string& errMsg) -{ - InitializeAllTargets(); - - // parse bitcode buffer - OwningPtr m(ParseBitcodeFile(buffer, getGlobalContext(), &errMsg)); - if (!m) - return NULL; - - std::string Triple = m->getTargetTriple(); - if (Triple.empty()) - Triple = sys::getHostTriple(); - - // find machine architecture for this module - const Target* march = TargetRegistry::lookupTarget(Triple, errMsg); - if (!march) - return NULL; - - // construct LTModule, hand over ownership of module and target - SubtargetFeatures Features; - Features.getDefaultSubtargetFeatures("" /* cpu */, llvm::Triple(Triple)); - std::string FeatureStr = Features.getString(); - TargetMachine* target = march->createTargetMachine(Triple, FeatureStr); - return new LTOModule(m.take(), target); +void LTOModule::getTargetOptions(TargetOptions &Options) { + Options.LessPreciseFPMADOption = EnableFPMAD; + Options.NoFramePointerElim = DisableFPElim; + Options.NoFramePointerElimNonLeaf = DisableFPElimNonLeaf; + Options.AllowFPOpFusion = FuseFPOps; + Options.UnsafeFPMath = EnableUnsafeFPMath; + Options.NoInfsFPMath = EnableNoInfsFPMath; + Options.NoNaNsFPMath = EnableNoNaNsFPMath; + Options.HonorSignDependentRoundingFPMathOption = + EnableHonorSignDependentRoundingFPMath; + Options.UseSoftFloat = GenerateSoftFloatCalls; + if (FloatABIForCalls != FloatABI::Default) + Options.FloatABIType = FloatABIForCalls; + Options.NoZerosInBSS = DontPlaceZerosInBSS; + Options.GuaranteedTailCallOpt = EnableGuaranteedTailCallOpt; + Options.DisableTailCalls = DisableTailCalls; + Options.StackAlignmentOverride = OverrideStackAlignment; + Options.RealignStack = EnableRealignStack; + Options.TrapFuncName = TrapFuncName; + Options.PositionIndependentExecutable = EnablePIE; + Options.EnableSegmentedStacks = SegmentedStacks; + Options.UseInitArray = UseInitArray; + Options.SSPBufferSize = SSPBufferSize; } - -const char* LTOModule::getTargetTriple() -{ - return _module->getTargetTriple().c_str(); +LTOModule *LTOModule::makeLTOModule(MemoryBuffer *buffer, + std::string &errMsg) { + static bool Initialized = false; + if (!Initialized) { + InitializeAllTargets(); + InitializeAllTargetMCs(); + InitializeAllAsmParsers(); + Initialized = true; + } + + // parse bitcode buffer + OwningPtr m(getLazyBitcodeModule(buffer, getGlobalContext(), + &errMsg)); + if (!m) { + delete buffer; + return NULL; + } + + std::string TripleStr = m->getTargetTriple(); + if (TripleStr.empty()) + TripleStr = sys::getDefaultTargetTriple(); + llvm::Triple Triple(TripleStr); + + // find machine architecture for this module + const Target *march = TargetRegistry::lookupTarget(TripleStr, errMsg); + if (!march) + return NULL; + + // construct LTOModule, hand over ownership of module and target + SubtargetFeatures Features; + Features.getDefaultSubtargetFeatures(Triple); + std::string FeatureStr = Features.getString(); + // Set a default CPU for Darwin triples. + std::string CPU; + if (Triple.isOSDarwin()) { + if (Triple.getArch() == llvm::Triple::x86_64) + CPU = "core2"; + else if (Triple.getArch() == llvm::Triple::x86) + CPU = "yonah"; + } + TargetOptions Options; + getTargetOptions(Options); + TargetMachine *target = march->createTargetMachine(TripleStr, CPU, FeatureStr, + Options); + LTOModule *Ret = new LTOModule(m.take(), target); + if (Ret->parseSymbols(errMsg)) { + delete Ret; + return NULL; + } + + return Ret; } -void LTOModule::setTargetTriple(const char *triple) -{ - _module->setTargetTriple(triple); +/// makeBuffer - Create a MemoryBuffer from a memory range. +MemoryBuffer *LTOModule::makeBuffer(const void *mem, size_t length) { + const char *startPtr = (const char*)mem; + return MemoryBuffer::getMemBuffer(StringRef(startPtr, length), "", false); } -void LTOModule::addDefinedFunctionSymbol(Function* f, Mangler &mangler) -{ - // add to list of defined symbols - addDefinedSymbol(f, mangler, true); - - // add external symbols referenced by this function. - for (Function::iterator b = f->begin(); b != f->end(); ++b) { - for (BasicBlock::iterator i = b->begin(); i != b->end(); ++i) { - for (unsigned count = 0, total = i->getNumOperands(); - count != total; ++count) { - findExternalRefs(i->getOperand(count), mangler); - } +/// objcClassNameFromExpression - Get string that the data pointer points to. +bool LTOModule::objcClassNameFromExpression(Constant *c, std::string &name) { + if (ConstantExpr *ce = dyn_cast(c)) { + Constant *op = ce->getOperand(0); + if (GlobalVariable *gvn = dyn_cast(op)) { + Constant *cn = gvn->getInitializer(); + if (ConstantDataArray *ca = dyn_cast(cn)) { + if (ca->isCString()) { + name = ".objc_class_name_" + ca->getAsCString().str(); + return true; } + } } + } + return false; } -// get string that data pointer points to -bool LTOModule::objcClassNameFromExpression(Constant* c, std::string& name) -{ - if (ConstantExpr* ce = dyn_cast(c)) { - Constant* op = ce->getOperand(0); - if (GlobalVariable* gvn = dyn_cast(op)) { - Constant* cn = gvn->getInitializer(); - if (ConstantArray* ca = dyn_cast(cn)) { - if (ca->isCString()) { - name = ".objc_class_name_" + ca->getAsString(); - return true; - } - } - } - } - return false; -} +/// addObjCClass - Parse i386/ppc ObjC class data structure. +void LTOModule::addObjCClass(GlobalVariable *clgv) { + ConstantStruct *c = dyn_cast(clgv->getInitializer()); + if (!c) return; -// parse i386/ppc ObjC class data structure -void LTOModule::addObjCClass(GlobalVariable* clgv) -{ - if (ConstantStruct* c = dyn_cast(clgv->getInitializer())) { - // second slot in __OBJC,__class is pointer to superclass name - std::string superclassName; - if (objcClassNameFromExpression(c->getOperand(1), superclassName)) { - NameAndAttributes info; - if (_undefines.find(superclassName.c_str()) == _undefines.end()) { - const char* symbolName = ::strdup(superclassName.c_str()); - info.name = ::strdup(symbolName); - info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; - // string is owned by _undefines - _undefines[info.name] = info; - } - } - // third slot in __OBJC,__class is pointer to class name - std::string className; - if (objcClassNameFromExpression(c->getOperand(2), className)) { - const char* symbolName = ::strdup(className.c_str()); - NameAndAttributes info; - info.name = symbolName; - info.attributes = (lto_symbol_attributes) - (LTO_SYMBOL_PERMISSIONS_DATA | - LTO_SYMBOL_DEFINITION_REGULAR | - LTO_SYMBOL_SCOPE_DEFAULT); - _symbols.push_back(info); - _defines[info.name] = 1; - } + // second slot in __OBJC,__class is pointer to superclass name + std::string superclassName; + if (objcClassNameFromExpression(c->getOperand(1), superclassName)) { + NameAndAttributes info; + StringMap::value_type &entry = + _undefines.GetOrCreateValue(superclassName); + if (!entry.getValue().name) { + const char *symbolName = entry.getKey().data(); + info.name = symbolName; + info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; + info.isFunction = false; + info.symbol = clgv; + entry.setValue(info); } -} + } + // third slot in __OBJC,__class is pointer to class name + std::string className; + if (objcClassNameFromExpression(c->getOperand(2), className)) { + StringSet::value_type &entry = _defines.GetOrCreateValue(className); + entry.setValue(1); -// parse i386/ppc ObjC category data structure -void LTOModule::addObjCCategory(GlobalVariable* clgv) -{ - if (ConstantStruct* c = dyn_cast(clgv->getInitializer())) { - // second slot in __OBJC,__category is pointer to target class name - std::string targetclassName; - if (objcClassNameFromExpression(c->getOperand(1), targetclassName)) { - NameAndAttributes info; - if (_undefines.find(targetclassName.c_str()) == _undefines.end()) { - const char* symbolName = ::strdup(targetclassName.c_str()); - info.name = ::strdup(symbolName); - info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; - // string is owned by _undefines - _undefines[info.name] = info; - } - } - } + NameAndAttributes info; + info.name = entry.getKey().data(); + info.attributes = LTO_SYMBOL_PERMISSIONS_DATA | + LTO_SYMBOL_DEFINITION_REGULAR | LTO_SYMBOL_SCOPE_DEFAULT; + info.isFunction = false; + info.symbol = clgv; + _symbols.push_back(info); + } } +/// addObjCCategory - Parse i386/ppc ObjC category data structure. +void LTOModule::addObjCCategory(GlobalVariable *clgv) { + ConstantStruct *c = dyn_cast(clgv->getInitializer()); + if (!c) return; + + // second slot in __OBJC,__category is pointer to target class name + std::string targetclassName; + if (!objcClassNameFromExpression(c->getOperand(1), targetclassName)) + return; + + NameAndAttributes info; + StringMap::value_type &entry = + _undefines.GetOrCreateValue(targetclassName); + + if (entry.getValue().name) + return; + + const char *symbolName = entry.getKey().data(); + info.name = symbolName; + info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; + info.isFunction = false; + info.symbol = clgv; + entry.setValue(info); +} -// parse i386/ppc ObjC class list data structure -void LTOModule::addObjCClassRef(GlobalVariable* clgv) -{ - std::string targetclassName; - if (objcClassNameFromExpression(clgv->getInitializer(), targetclassName)) { - NameAndAttributes info; - if (_undefines.find(targetclassName.c_str()) == _undefines.end()) { - const char* symbolName = ::strdup(targetclassName.c_str()); - info.name = ::strdup(symbolName); - info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; - // string is owned by _undefines - _undefines[info.name] = info; - } - } +/// addObjCClassRef - Parse i386/ppc ObjC class list data structure. +void LTOModule::addObjCClassRef(GlobalVariable *clgv) { + std::string targetclassName; + if (!objcClassNameFromExpression(clgv->getInitializer(), targetclassName)) + return; + + NameAndAttributes info; + StringMap::value_type &entry = + _undefines.GetOrCreateValue(targetclassName); + if (entry.getValue().name) + return; + + const char *symbolName = entry.getKey().data(); + info.name = symbolName; + info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; + info.isFunction = false; + info.symbol = clgv; + entry.setValue(info); } +/// addDefinedDataSymbol - Add a data symbol as defined to the list. +void LTOModule::addDefinedDataSymbol(GlobalValue *v) { + // Add to list of defined symbols. + addDefinedSymbol(v, false); + + if (!v->hasSection() /* || !isTargetDarwin */) + return; + + // Special case i386/ppc ObjC data structures in magic sections: + // The issue is that the old ObjC object format did some strange + // contortions to avoid real linker symbols. For instance, the + // ObjC class data structure is allocated statically in the executable + // that defines that class. That data structures contains a pointer to + // its superclass. But instead of just initializing that part of the + // struct to the address of its superclass, and letting the static and + // dynamic linkers do the rest, the runtime works by having that field + // instead point to a C-string that is the name of the superclass. + // At runtime the objc initialization updates that pointer and sets + // it to point to the actual super class. As far as the linker + // knows it is just a pointer to a string. But then someone wanted the + // linker to issue errors at build time if the superclass was not found. + // So they figured out a way in mach-o object format to use an absolute + // symbols (.objc_class_name_Foo = 0) and a floating reference + // (.reference .objc_class_name_Bar) to cause the linker into erroring when + // a class was missing. + // The following synthesizes the implicit .objc_* symbols for the linker + // from the ObjC data structures generated by the front end. + + // special case if this data blob is an ObjC class definition + if (v->getSection().compare(0, 15, "__OBJC,__class,") == 0) { + if (GlobalVariable *gv = dyn_cast(v)) { + addObjCClass(gv); + } + } -void LTOModule::addDefinedDataSymbol(GlobalValue* v, Mangler& mangler) -{ - // add to list of defined symbols - addDefinedSymbol(v, mangler, false); - - // Special case i386/ppc ObjC data structures in magic sections: - // The issue is that the old ObjC object format did some strange - // contortions to avoid real linker symbols. For instance, the - // ObjC class data structure is allocated statically in the executable - // that defines that class. That data structures contains a pointer to - // its superclass. But instead of just initializing that part of the - // struct to the address of its superclass, and letting the static and - // dynamic linkers do the rest, the runtime works by having that field - // instead point to a C-string that is the name of the superclass. - // At runtime the objc initialization updates that pointer and sets - // it to point to the actual super class. As far as the linker - // knows it is just a pointer to a string. But then someone wanted the - // linker to issue errors at build time if the superclass was not found. - // So they figured out a way in mach-o object format to use an absolute - // symbols (.objc_class_name_Foo = 0) and a floating reference - // (.reference .objc_class_name_Bar) to cause the linker into erroring when - // a class was missing. - // The following synthesizes the implicit .objc_* symbols for the linker - // from the ObjC data structures generated by the front end. - if (v->hasSection() /* && isTargetDarwin */) { - // special case if this data blob is an ObjC class definition - if (v->getSection().compare(0, 15, "__OBJC,__class,") == 0) { - if (GlobalVariable* gv = dyn_cast(v)) { - addObjCClass(gv); - } - } - - // special case if this data blob is an ObjC category definition - else if (v->getSection().compare(0, 18, "__OBJC,__category,") == 0) { - if (GlobalVariable* gv = dyn_cast(v)) { - addObjCCategory(gv); - } - } - - // special case if this data blob is the list of referenced classes - else if (v->getSection().compare(0, 18, "__OBJC,__cls_refs,") == 0) { - if (GlobalVariable* gv = dyn_cast(v)) { - addObjCClassRef(gv); - } - } + // special case if this data blob is an ObjC category definition + else if (v->getSection().compare(0, 18, "__OBJC,__category,") == 0) { + if (GlobalVariable *gv = dyn_cast(v)) { + addObjCCategory(gv); } + } - // add external symbols referenced by this data. - for (unsigned count = 0, total = v->getNumOperands(); - count != total; ++count) { - findExternalRefs(v->getOperand(count), mangler); + // special case if this data blob is the list of referenced classes + else if (v->getSection().compare(0, 18, "__OBJC,__cls_refs,") == 0) { + if (GlobalVariable *gv = dyn_cast(v)) { + addObjCClassRef(gv); } + } } +/// addDefinedFunctionSymbol - Add a function symbol as defined to the list. +void LTOModule::addDefinedFunctionSymbol(Function *f) { + // add to list of defined symbols + addDefinedSymbol(f, true); +} -void LTOModule::addDefinedSymbol(GlobalValue* def, Mangler &mangler, - bool isFunction) -{ - // ignore all llvm.* symbols - if (def->getName().startswith("llvm.")) - return; - - // string is owned by _defines - const char* symbolName = ::strdup(mangler.getNameWithPrefix(def).c_str()); - - // set alignment part log2() can have rounding errors - uint32_t align = def->getAlignment(); - uint32_t attr = align ? CountTrailingZeros_32(def->getAlignment()) : 0; - - // set permissions part - if (isFunction) - attr |= LTO_SYMBOL_PERMISSIONS_CODE; - else { - GlobalVariable* gv = dyn_cast(def); - if (gv && gv->isConstant()) - attr |= LTO_SYMBOL_PERMISSIONS_RODATA; - else - attr |= LTO_SYMBOL_PERMISSIONS_DATA; - } - - // set definition part - if (def->hasWeakLinkage() || def->hasLinkOnceLinkage()) { - attr |= LTO_SYMBOL_DEFINITION_WEAK; - } - else if (def->hasCommonLinkage()) { - attr |= LTO_SYMBOL_DEFINITION_TENTATIVE; - } - else { - attr |= LTO_SYMBOL_DEFINITION_REGULAR; - } - - // set scope part - if (def->hasHiddenVisibility()) - attr |= LTO_SYMBOL_SCOPE_HIDDEN; - else if (def->hasProtectedVisibility()) - attr |= LTO_SYMBOL_SCOPE_PROTECTED; - else if (def->hasExternalLinkage() || def->hasWeakLinkage() - || def->hasLinkOnceLinkage() || def->hasCommonLinkage()) - attr |= LTO_SYMBOL_SCOPE_DEFAULT; +/// addDefinedSymbol - Add a defined symbol to the list. +void LTOModule::addDefinedSymbol(GlobalValue *def, bool isFunction) { + // ignore all llvm.* symbols + if (def->getName().startswith("llvm.")) + return; + + // string is owned by _defines + SmallString<64> Buffer; + _mangler.getNameWithPrefix(Buffer, def, false); + + // set alignment part log2() can have rounding errors + uint32_t align = def->getAlignment(); + uint32_t attr = align ? CountTrailingZeros_32(def->getAlignment()) : 0; + + // set permissions part + if (isFunction) { + attr |= LTO_SYMBOL_PERMISSIONS_CODE; + } else { + GlobalVariable *gv = dyn_cast(def); + if (gv && gv->isConstant()) + attr |= LTO_SYMBOL_PERMISSIONS_RODATA; else - attr |= LTO_SYMBOL_SCOPE_INTERNAL; + attr |= LTO_SYMBOL_PERMISSIONS_DATA; + } + + // set definition part + if (def->hasWeakLinkage() || def->hasLinkOnceLinkage() || + def->hasLinkerPrivateWeakLinkage()) + attr |= LTO_SYMBOL_DEFINITION_WEAK; + else if (def->hasCommonLinkage()) + attr |= LTO_SYMBOL_DEFINITION_TENTATIVE; + else + attr |= LTO_SYMBOL_DEFINITION_REGULAR; + + // set scope part + if (def->hasHiddenVisibility()) + attr |= LTO_SYMBOL_SCOPE_HIDDEN; + else if (def->hasProtectedVisibility()) + attr |= LTO_SYMBOL_SCOPE_PROTECTED; + else if (def->hasExternalLinkage() || def->hasWeakLinkage() || + def->hasLinkOnceLinkage() || def->hasCommonLinkage() || + def->hasLinkerPrivateWeakLinkage()) + attr |= LTO_SYMBOL_SCOPE_DEFAULT; + else if (def->hasLinkOnceODRAutoHideLinkage()) + attr |= LTO_SYMBOL_SCOPE_DEFAULT_CAN_BE_HIDDEN; + else + attr |= LTO_SYMBOL_SCOPE_INTERNAL; + + StringSet::value_type &entry = _defines.GetOrCreateValue(Buffer); + entry.setValue(1); + + // fill information structure + NameAndAttributes info; + StringRef Name = entry.getKey(); + info.name = Name.data(); + assert(info.name[Name.size()] == '\0'); + info.attributes = attr; + info.isFunction = isFunction; + info.symbol = def; + + // add to table of symbols + _symbols.push_back(info); +} + +/// addAsmGlobalSymbol - Add a global symbol from module-level ASM to the +/// defined list. +void LTOModule::addAsmGlobalSymbol(const char *name, + lto_symbol_attributes scope) { + StringSet::value_type &entry = _defines.GetOrCreateValue(name); + + // only add new define if not already defined + if (entry.getValue()) + return; + + entry.setValue(1); + + NameAndAttributes &info = _undefines[entry.getKey().data()]; + + if (info.symbol == 0) { + // FIXME: This is trying to take care of module ASM like this: + // + // module asm ".zerofill __FOO, __foo, _bar_baz_qux, 0" + // + // but is gross and its mother dresses it funny. Have the ASM parser give us + // more details for this type of situation so that we're not guessing so + // much. + + // fill information structure + info.name = entry.getKey().data(); + info.attributes = + LTO_SYMBOL_PERMISSIONS_DATA | LTO_SYMBOL_DEFINITION_REGULAR | scope; + info.isFunction = false; + info.symbol = 0; // add to table of symbols - NameAndAttributes info; - info.name = symbolName; - info.attributes = (lto_symbol_attributes)attr; _symbols.push_back(info); - _defines[info.name] = 1; -} + return; + } -void LTOModule::addAsmGlobalSymbol(const char *name) { - // only add new define if not already defined - if (_defines.count(name) == 0) - return; - - // string is owned by _defines - const char *symbolName = ::strdup(name); - uint32_t attr = LTO_SYMBOL_DEFINITION_REGULAR; - attr |= LTO_SYMBOL_SCOPE_DEFAULT; - NameAndAttributes info; - info.name = symbolName; - info.attributes = (lto_symbol_attributes)attr; - _symbols.push_back(info); - _defines[info.name] = 1; + if (info.isFunction) + addDefinedFunctionSymbol(cast(info.symbol)); + else + addDefinedDataSymbol(info.symbol); + + _symbols.back().attributes &= ~LTO_SYMBOL_SCOPE_MASK; + _symbols.back().attributes |= scope; } -void LTOModule::addPotentialUndefinedSymbol(GlobalValue* decl, Mangler &mangler) -{ - // ignore all llvm.* symbols - if (decl->getName().startswith("llvm.")) - return; +/// addAsmGlobalSymbolUndef - Add a global symbol from module-level ASM to the +/// undefined list. +void LTOModule::addAsmGlobalSymbolUndef(const char *name) { + StringMap::value_type &entry = + _undefines.GetOrCreateValue(name); - // ignore all aliases - if (isa(decl)) - return; + _asm_undefines.push_back(entry.getKey().data()); - std::string name = mangler.getNameWithPrefix(decl); + // we already have the symbol + if (entry.getValue().name) + return; - // we already have the symbol - if (_undefines.find(name) != _undefines.end()) - return; + uint32_t attr = LTO_SYMBOL_DEFINITION_UNDEFINED;; + attr |= LTO_SYMBOL_SCOPE_DEFAULT; + NameAndAttributes info; + info.name = entry.getKey().data(); + info.attributes = attr; + info.isFunction = false; + info.symbol = 0; - NameAndAttributes info; - // string is owned by _undefines - info.name = ::strdup(name.c_str()); - if (decl->hasExternalWeakLinkage()) - info.attributes = LTO_SYMBOL_DEFINITION_WEAKUNDEF; - else - info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; - _undefines[name] = info; + entry.setValue(info); } +/// addPotentialUndefinedSymbol - Add a symbol which isn't defined just yet to a +/// list to be resolved later. +void LTOModule::addPotentialUndefinedSymbol(GlobalValue *decl, bool isFunc) { + // ignore all llvm.* symbols + if (decl->getName().startswith("llvm.")) + return; + + // ignore all aliases + if (isa(decl)) + return; + SmallString<64> name; + _mangler.getNameWithPrefix(name, decl, false); -// Find external symbols referenced by VALUE. This is a recursive function. -void LTOModule::findExternalRefs(Value* value, Mangler &mangler) { + StringMap::value_type &entry = + _undefines.GetOrCreateValue(name); - if (GlobalValue* gv = dyn_cast(value)) { - if (!gv->hasExternalLinkage()) - addPotentialUndefinedSymbol(gv, mangler); - // If this is a variable definition, do not recursively process - // initializer. It might contain a reference to this variable - // and cause an infinite loop. The initializer will be - // processed in addDefinedDataSymbol(). - return; + // we already have the symbol + if (entry.getValue().name) + return; + + NameAndAttributes info; + + info.name = entry.getKey().data(); + + if (decl->hasExternalWeakLinkage()) + info.attributes = LTO_SYMBOL_DEFINITION_WEAKUNDEF; + else + info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; + + info.isFunction = isFunc; + info.symbol = decl; + + entry.setValue(info); +} + +namespace { + class RecordStreamer : public MCStreamer { + public: + enum State { NeverSeen, Global, Defined, DefinedGlobal, Used }; + + private: + StringMap Symbols; + + void markDefined(const MCSymbol &Symbol) { + State &S = Symbols[Symbol.getName()]; + switch (S) { + case DefinedGlobal: + case Global: + S = DefinedGlobal; + break; + case NeverSeen: + case Defined: + case Used: + S = Defined; + break; + } + } + void markGlobal(const MCSymbol &Symbol) { + State &S = Symbols[Symbol.getName()]; + switch (S) { + case DefinedGlobal: + case Defined: + S = DefinedGlobal; + break; + + case NeverSeen: + case Global: + case Used: + S = Global; + break; + } + } + void markUsed(const MCSymbol &Symbol) { + State &S = Symbols[Symbol.getName()]; + switch (S) { + case DefinedGlobal: + case Defined: + case Global: + break; + + case NeverSeen: + case Used: + S = Used; + break; + } } - // GlobalValue, even with InternalLinkage type, may have operands with - // ExternalLinkage type. Do not ignore these operands. - if (Constant* c = dyn_cast(value)) { - // Handle ConstantExpr, ConstantStruct, ConstantArry etc. - for (unsigned i = 0, e = c->getNumOperands(); i != e; ++i) - findExternalRefs(c->getOperand(i), mangler); + // FIXME: mostly copied for the obj streamer. + void AddValueSymbols(const MCExpr *Value) { + switch (Value->getKind()) { + case MCExpr::Target: + // FIXME: What should we do in here? + break; + + case MCExpr::Constant: + break; + + case MCExpr::Binary: { + const MCBinaryExpr *BE = cast(Value); + AddValueSymbols(BE->getLHS()); + AddValueSymbols(BE->getRHS()); + break; + } + + case MCExpr::SymbolRef: + markUsed(cast(Value)->getSymbol()); + break; + + case MCExpr::Unary: + AddValueSymbols(cast(Value)->getSubExpr()); + break; + } } -} -void LTOModule::lazyParseSymbols() -{ - if (!_symbolsParsed) { - _symbolsParsed = true; - - // Use mangler to add GlobalPrefix to names to match linker names. - MCContext Context(*_target->getMCAsmInfo()); - Mangler mangler(Context, *_target->getTargetData()); - - // add functions - for (Module::iterator f = _module->begin(); f != _module->end(); ++f) { - if (f->isDeclaration()) - addPotentialUndefinedSymbol(f, mangler); - else - addDefinedFunctionSymbol(f, mangler); - } - - // add data - for (Module::global_iterator v = _module->global_begin(), - e = _module->global_end(); v != e; ++v) { - if (v->isDeclaration()) - addPotentialUndefinedSymbol(v, mangler); - else - addDefinedDataSymbol(v, mangler); - } + public: + typedef StringMap::const_iterator const_iterator; - // add asm globals - const std::string &inlineAsm = _module->getModuleInlineAsm(); - const std::string glbl = ".globl"; - std::string asmSymbolName; - std::string::size_type pos = inlineAsm.find(glbl, 0); - while (pos != std::string::npos) { - // eat .globl - pos = pos + 6; - - // skip white space between .globl and symbol name - std::string::size_type pbegin = inlineAsm.find_first_not_of(' ', pos); - if (pbegin == std::string::npos) - break; - - // find end-of-line - std::string::size_type pend = inlineAsm.find_first_of('\n', pbegin); - if (pend == std::string::npos) - break; - - asmSymbolName.assign(inlineAsm, pbegin, pend - pbegin); - addAsmGlobalSymbol(asmSymbolName.c_str()); - - // search next .globl - pos = inlineAsm.find(glbl, pend); - } + const_iterator begin() { + return Symbols.begin(); + } - // make symbols for all undefines - for (StringMap::iterator it=_undefines.begin(); - it != _undefines.end(); ++it) { - // if this symbol also has a definition, then don't make an undefine - // because it is a tentative definition - if (_defines.count(it->getKey()) == 0) { - NameAndAttributes info = it->getValue(); - _symbols.push_back(info); - } - } - } -} + const_iterator end() { + return Symbols.end(); + } + + RecordStreamer(MCContext &Context) : MCStreamer(Context) {} + + virtual void EmitInstruction(const MCInst &Inst) { + // Scan for values. + for (unsigned i = Inst.getNumOperands(); i--; ) + if (Inst.getOperand(i).isExpr()) + AddValueSymbols(Inst.getOperand(i).getExpr()); + } + virtual void EmitLabel(MCSymbol *Symbol) { + Symbol->setSection(*getCurrentSection()); + markDefined(*Symbol); + } + virtual void EmitAssignment(MCSymbol *Symbol, const MCExpr *Value) { + // FIXME: should we handle aliases? + markDefined(*Symbol); + } + virtual void EmitSymbolAttribute(MCSymbol *Symbol, MCSymbolAttr Attribute) { + if (Attribute == MCSA_Global) + markGlobal(*Symbol); + } + virtual void EmitZerofill(const MCSection *Section, MCSymbol *Symbol, + uint64_t Size , unsigned ByteAlignment) { + markDefined(*Symbol); + } + virtual void EmitCommonSymbol(MCSymbol *Symbol, uint64_t Size, + unsigned ByteAlignment) { + markDefined(*Symbol); + } + // Noop calls. + virtual void ChangeSection(const MCSection *Section) {} + virtual void InitSections() {} + virtual void EmitAssemblerFlag(MCAssemblerFlag Flag) {} + virtual void EmitThumbFunc(MCSymbol *Func) {} + virtual void EmitSymbolDesc(MCSymbol *Symbol, unsigned DescValue) {} + virtual void EmitWeakReference(MCSymbol *Alias, const MCSymbol *Symbol) {} + virtual void BeginCOFFSymbolDef(const MCSymbol *Symbol) {} + virtual void EmitCOFFSymbolStorageClass(int StorageClass) {} + virtual void EmitCOFFSymbolType(int Type) {} + virtual void EndCOFFSymbolDef() {} + virtual void EmitELFSize(MCSymbol *Symbol, const MCExpr *Value) {} + virtual void EmitLocalCommonSymbol(MCSymbol *Symbol, uint64_t Size, + unsigned ByteAlignment) {} + virtual void EmitTBSSSymbol(const MCSection *Section, MCSymbol *Symbol, + uint64_t Size, unsigned ByteAlignment) {} + virtual void EmitBytes(StringRef Data, unsigned AddrSpace) {} + virtual void EmitValueImpl(const MCExpr *Value, unsigned Size, + unsigned AddrSpace) {} + virtual void EmitULEB128Value(const MCExpr *Value) {} + virtual void EmitSLEB128Value(const MCExpr *Value) {} + virtual void EmitValueToAlignment(unsigned ByteAlignment, int64_t Value, + unsigned ValueSize, + unsigned MaxBytesToEmit) {} + virtual void EmitCodeAlignment(unsigned ByteAlignment, + unsigned MaxBytesToEmit) {} + virtual bool EmitValueToOffset(const MCExpr *Offset, + unsigned char Value ) { return false; } + virtual void EmitFileDirective(StringRef Filename) {} + virtual void EmitDwarfAdvanceLineAddr(int64_t LineDelta, + const MCSymbol *LastLabel, + const MCSymbol *Label, + unsigned PointerSize) {} + virtual void FinishImpl() {} + }; +} // end anonymous namespace + +/// addAsmGlobalSymbols - Add global symbols from module-level ASM to the +/// defined or undefined lists. +bool LTOModule::addAsmGlobalSymbols(std::string &errMsg) { + const std::string &inlineAsm = _module->getModuleInlineAsm(); + if (inlineAsm.empty()) + return false; -uint32_t LTOModule::getSymbolCount() -{ - lazyParseSymbols(); - return _symbols.size(); + OwningPtr Streamer(new RecordStreamer(_context)); + MemoryBuffer *Buffer = MemoryBuffer::getMemBuffer(inlineAsm); + SourceMgr SrcMgr; + SrcMgr.AddNewSourceBuffer(Buffer, SMLoc()); + OwningPtr Parser(createMCAsmParser(SrcMgr, + _context, *Streamer, + *_target->getMCAsmInfo())); + const Target &T = _target->getTarget(); + OwningPtr + STI(T.createMCSubtargetInfo(_target->getTargetTriple(), + _target->getTargetCPU(), + _target->getTargetFeatureString())); + OwningPtr TAP(T.createMCAsmParser(*STI, *Parser.get())); + if (!TAP) { + errMsg = "target " + std::string(T.getName()) + + " does not define AsmParser."; + return true; + } + + Parser->setTargetParser(*TAP); + if (Parser->Run(false)) + return true; + + for (RecordStreamer::const_iterator i = Streamer->begin(), + e = Streamer->end(); i != e; ++i) { + StringRef Key = i->first(); + RecordStreamer::State Value = i->second; + if (Value == RecordStreamer::DefinedGlobal) + addAsmGlobalSymbol(Key.data(), LTO_SYMBOL_SCOPE_DEFAULT); + else if (Value == RecordStreamer::Defined) + addAsmGlobalSymbol(Key.data(), LTO_SYMBOL_SCOPE_INTERNAL); + else if (Value == RecordStreamer::Global || + Value == RecordStreamer::Used) + addAsmGlobalSymbolUndef(Key.data()); + } + + return false; } +/// isDeclaration - Return 'true' if the global value is a declaration. +static bool isDeclaration(const GlobalValue &V) { + if (V.hasAvailableExternallyLinkage()) + return true; -lto_symbol_attributes LTOModule::getSymbolAttributes(uint32_t index) -{ - lazyParseSymbols(); - if (index < _symbols.size()) - return _symbols[index].attributes; - else - return lto_symbol_attributes(0); + if (V.isMaterializable()) + return false; + + return V.isDeclaration(); } -const char* LTOModule::getSymbolName(uint32_t index) -{ - lazyParseSymbols(); - if (index < _symbols.size()) - return _symbols[index].name; +/// parseSymbols - Parse the symbols from the module and model-level ASM and add +/// them to either the defined or undefined lists. +bool LTOModule::parseSymbols(std::string &errMsg) { + // add functions + for (Module::iterator f = _module->begin(), e = _module->end(); f != e; ++f) { + if (isDeclaration(*f)) + addPotentialUndefinedSymbol(f, true); + else + addDefinedFunctionSymbol(f); + } + + // add data + for (Module::global_iterator v = _module->global_begin(), + e = _module->global_end(); v != e; ++v) { + if (isDeclaration(*v)) + addPotentialUndefinedSymbol(v, false); + else + addDefinedDataSymbol(v); + } + + // add asm globals + if (addAsmGlobalSymbols(errMsg)) + return true; + + // add aliases + for (Module::alias_iterator a = _module->alias_begin(), + e = _module->alias_end(); a != e; ++a) { + if (isDeclaration(*a->getAliasedGlobal())) + // Is an alias to a declaration. + addPotentialUndefinedSymbol(a, false); else - return NULL; + addDefinedDataSymbol(a); + } + + // make symbols for all undefines + for (StringMap::iterator u =_undefines.begin(), + e = _undefines.end(); u != e; ++u) { + // If this symbol also has a definition, then don't make an undefine because + // it is a tentative definition. + if (_defines.count(u->getKey())) continue; + NameAndAttributes info = u->getValue(); + _symbols.push_back(info); + } + + return false; }