//
// The LLVM Compiler Infrastructure
//
-// This file was developed by the LLVM research group and is distributed under
-// the University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
-// This pass loops over all of the functions in the input module, looking for a
-// main function. If a main function is found, all other functions and all
-// global variables with initializers are marked as internal.
+// This pass loops over all of the functions and variables in the input module.
+// If the function or variable is not in the list of external names given to
+// the pass it is marked as internal.
+//
+// This transformation would not be legal in a regular compilation, but it gets
+// extra information from the linker about what is safe.
+//
+// For example: Internalizing a function with external linkage. Only if we are
+// told it is only used from within this module, it is safe to do it.
//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "internalize"
#include "llvm/Transforms/IPO.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/Analysis/CallGraph.h"
+#include "llvm/IR/Module.h"
#include "llvm/Pass.h"
-#include "llvm/Module.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
-#include "llvm/ADT/Statistic.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Transforms/Utils/GlobalStatus.h"
+#include "llvm/Transforms/Utils/ModuleUtils.h"
#include <fstream>
#include <set>
using namespace llvm;
+#define DEBUG_TYPE "internalize"
+
+STATISTIC(NumAliases , "Number of aliases internalized");
STATISTIC(NumFunctions, "Number of functions internalized");
STATISTIC(NumGlobals , "Number of global vars internalized");
-namespace {
-
- // APIFile - A file which contains a list of symbols that should not be marked
- // external.
- cl::opt<std::string>
- APIFile("internalize-public-api-file", cl::value_desc("filename"),
- cl::desc("A file containing list of symbol names to preserve"));
+// APIFile - A file which contains a list of symbols that should not be marked
+// external.
+static cl::opt<std::string>
+APIFile("internalize-public-api-file", cl::value_desc("filename"),
+ cl::desc("A file containing list of symbol names to preserve"));
- // APIList - A list of symbols that should not be marked internal.
- cl::list<std::string>
- APIList("internalize-public-api-list", cl::value_desc("list"),
- cl::desc("A list of symbol names to preserve"),
- cl::CommaSeparated);
+// APIList - A list of symbols that should not be marked internal.
+static cl::list<std::string>
+APIList("internalize-public-api-list", cl::value_desc("list"),
+ cl::desc("A list of symbol names to preserve"),
+ cl::CommaSeparated);
+namespace {
class InternalizePass : public ModulePass {
std::set<std::string> ExternalNames;
- bool DontInternalize;
public:
- InternalizePass(bool InternalizeEverything = true);
- InternalizePass(const std::vector <const char *>& exportList);
+ static char ID; // Pass identification, replacement for typeid
+ explicit InternalizePass();
+ explicit InternalizePass(ArrayRef<const char *> ExportList);
void LoadFile(const char *Filename);
- virtual bool runOnModule(Module &M);
+ bool maybeInternalize(GlobalValue &GV,
+ const std::set<const Comdat *> &ExternalComdats);
+ void checkComdatVisibility(GlobalValue &GV,
+ std::set<const Comdat *> &ExternalComdats);
+ bool runOnModule(Module &M) override;
+
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.setPreservesCFG();
+ AU.addPreserved<CallGraphWrapperPass>();
+ }
};
- RegisterPass<InternalizePass> X("internalize", "Internalize Global Symbols");
} // end anonymous namespace
-InternalizePass::InternalizePass(bool InternalizeEverything)
- : DontInternalize(false){
- if (!APIFile.empty()) // If a filename is specified, use it
+char InternalizePass::ID = 0;
+INITIALIZE_PASS(InternalizePass, "internalize",
+ "Internalize Global Symbols", false, false)
+
+InternalizePass::InternalizePass() : ModulePass(ID) {
+ initializeInternalizePassPass(*PassRegistry::getPassRegistry());
+ if (!APIFile.empty()) // If a filename is specified, use it.
LoadFile(APIFile.c_str());
- else if (!APIList.empty()) // Else, if a list is specified, use it.
- ExternalNames.insert(APIList.begin(), APIList.end());
- else if (!InternalizeEverything)
- // Finally, if we're allowed to, internalize all but main.
- DontInternalize = true;
+ ExternalNames.insert(APIList.begin(), APIList.end());
}
-InternalizePass::InternalizePass(const std::vector<const char *>&exportList)
- : DontInternalize(false){
- for(std::vector<const char *>::const_iterator itr = exportList.begin();
- itr != exportList.end(); itr++) {
+InternalizePass::InternalizePass(ArrayRef<const char *> ExportList)
+ : ModulePass(ID) {
+ initializeInternalizePassPass(*PassRegistry::getPassRegistry());
+ for(ArrayRef<const char *>::const_iterator itr = ExportList.begin();
+ itr != ExportList.end(); itr++) {
ExternalNames.insert(*itr);
}
}
// Load the APIFile...
std::ifstream In(Filename);
if (!In.good()) {
- cerr << "WARNING: Internalize couldn't load file '" << Filename << "'!\n";
- return; // Do not internalize anything...
+ errs() << "WARNING: Internalize couldn't load file '" << Filename
+ << "'! Continuing as if it's empty.\n";
+ return; // Just continue as if the file were empty
}
while (In) {
std::string Symbol;
}
}
+static bool isExternallyVisible(const GlobalValue &GV,
+ const std::set<std::string> &ExternalNames) {
+ // Function must be defined here
+ if (GV.isDeclaration())
+ return true;
+
+ // Available externally is really just a "declaration with a body".
+ if (GV.hasAvailableExternallyLinkage())
+ return true;
+
+ // Assume that dllexported symbols are referenced elsewhere
+ if (GV.hasDLLExportStorageClass())
+ return true;
+
+ // Marked to keep external?
+ if (!GV.hasLocalLinkage() && ExternalNames.count(GV.getName()))
+ return true;
+
+ return false;
+}
+
+// Internalize GV if it is possible to do so, i.e. it is not externally visible
+// and is not a member of an externally visible comdat.
+bool InternalizePass::maybeInternalize(
+ GlobalValue &GV, const std::set<const Comdat *> &ExternalComdats) {
+ if (Comdat *C = GV.getComdat()) {
+ if (ExternalComdats.count(C))
+ return false;
+
+ // If a comdat is not externally visible we can drop it.
+ if (auto GO = dyn_cast<GlobalObject>(&GV))
+ GO->setComdat(nullptr);
+
+ if (GV.hasLocalLinkage())
+ return false;
+ } else {
+ if (GV.hasLocalLinkage())
+ return false;
+
+ if (isExternallyVisible(GV, ExternalNames))
+ return false;
+ }
+
+ GV.setVisibility(GlobalValue::DefaultVisibility);
+ GV.setLinkage(GlobalValue::InternalLinkage);
+ return true;
+}
+
+// If GV is part of a comdat and is externally visible, keep track of its
+// comdat so that we don't internalize any of its members.
+void InternalizePass::checkComdatVisibility(
+ GlobalValue &GV, std::set<const Comdat *> &ExternalComdats) {
+ Comdat *C = GV.getComdat();
+ if (!C)
+ return;
+
+ if (isExternallyVisible(GV, ExternalNames))
+ ExternalComdats.insert(C);
+}
+
bool InternalizePass::runOnModule(Module &M) {
- if (DontInternalize) return false;
-
- // If no list or file of symbols was specified, check to see if there is a
- // "main" symbol defined in the module. If so, use it, otherwise do not
- // internalize the module, it must be a library or something.
- //
- if (ExternalNames.empty()) {
- Function *MainFunc = M.getMainFunction();
- if (MainFunc == 0 || MainFunc->isExternal())
- return false; // No main found, must be a library...
-
- // Preserve main, internalize all else.
- ExternalNames.insert(MainFunc->getName());
+ CallGraphWrapperPass *CGPass = getAnalysisIfAvailable<CallGraphWrapperPass>();
+ CallGraph *CG = CGPass ? &CGPass->getCallGraph() : nullptr;
+ CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : nullptr;
+
+ SmallPtrSet<GlobalValue *, 8> Used;
+ collectUsedGlobalVariables(M, Used, false);
+
+ // Collect comdat visiblity information for the module.
+ std::set<const Comdat *> ExternalComdats;
+ if (!M.getComdatSymbolTable().empty()) {
+ for (Function &F : M)
+ checkComdatVisibility(F, ExternalComdats);
+ for (GlobalVariable &GV : M.globals())
+ checkComdatVisibility(GV, ExternalComdats);
+ for (GlobalAlias &GA : M.aliases())
+ checkComdatVisibility(GA, ExternalComdats);
}
-
- bool Changed = false;
-
- // Found a main function, mark all functions not named main as internal.
- for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
- if (!I->isExternal() && // Function must be defined here
- !I->hasInternalLinkage() && // Can't already have internal linkage
- !ExternalNames.count(I->getName())) {// Not marked to keep external?
- I->setLinkage(GlobalValue::InternalLinkage);
- Changed = true;
- ++NumFunctions;
- DOUT << "Internalizing func " << I->getName() << "\n";
- }
-
+
+ // We must assume that globals in llvm.used have a reference that not even
+ // the linker can see, so we don't internalize them.
+ // For llvm.compiler.used the situation is a bit fuzzy. The assembler and
+ // linker can drop those symbols. If this pass is running as part of LTO,
+ // one might think that it could just drop llvm.compiler.used. The problem
+ // is that even in LTO llvm doesn't see every reference. For example,
+ // we don't see references from function local inline assembly. To be
+ // conservative, we internalize symbols in llvm.compiler.used, but we
+ // keep llvm.compiler.used so that the symbol is not deleted by llvm.
+ for (GlobalValue *V : Used) {
+ ExternalNames.insert(V->getName());
+ }
+
+ // Mark all functions not in the api as internal.
+ for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
+ if (!maybeInternalize(*I, ExternalComdats))
+ continue;
+
+ if (ExternalNode)
+ // Remove a callgraph edge from the external node to this function.
+ ExternalNode->removeOneAbstractEdgeTo((*CG)[I]);
+
+ ++NumFunctions;
+ DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n");
+ }
+
// Never internalize the llvm.used symbol. It is used to implement
// attribute((used)).
+ // FIXME: Shouldn't this just filter on llvm.metadata section??
ExternalNames.insert("llvm.used");
-
- // Never internalize anchors used by the debugger, else the debugger won't
- // find them. (see MachineDebugInfo.)
- ExternalNames.insert("llvm.dbg.compile_units");
- ExternalNames.insert("llvm.dbg.global_variables");
- ExternalNames.insert("llvm.dbg.subprograms");
-
- // Mark all global variables with initializers as internal as well.
+ ExternalNames.insert("llvm.compiler.used");
+
+ // Never internalize anchors used by the machine module info, else the info
+ // won't find them. (see MachineModuleInfo.)
+ ExternalNames.insert("llvm.global_ctors");
+ ExternalNames.insert("llvm.global_dtors");
+ ExternalNames.insert("llvm.global.annotations");
+
+ // Never internalize symbols code-gen inserts.
+ // FIXME: We should probably add this (and the __stack_chk_guard) via some
+ // type of call-back in CodeGen.
+ ExternalNames.insert("__stack_chk_fail");
+ ExternalNames.insert("__stack_chk_guard");
+
+ // Mark all global variables with initializers that are not in the api as
+ // internal as well.
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
- I != E; ++I)
- if (!I->isExternal() && !I->hasInternalLinkage() &&
- !ExternalNames.count(I->getName())) {
- // Special case handling of the global ctor and dtor list. When we
- // internalize it, we mark it constant, which allows elimination of
- // the list if it's empty.
- //
- if (I->hasAppendingLinkage() && (I->getName() == "llvm.global_ctors" ||
- I->getName() == "llvm.global_dtors")) {
- // If the global ctors/dtors list has no uses, do not internalize it, as
- // there is no __main in this program, so the asmprinter should handle
- // it.
- if (I->use_empty()) continue;
-
- // Otherwise, also mark the list constant, as we know that it will not
- // be mutated any longer, and the makes simple IPO xforms automatically
- // better.
- I->setConstant(true);
- }
-
- I->setLinkage(GlobalValue::InternalLinkage);
- Changed = true;
- ++NumGlobals;
- DOUT << "Internalized gvar " << I->getName() << "\n";
- }
-
- return Changed;
-}
+ I != E; ++I) {
+ if (!maybeInternalize(*I, ExternalComdats))
+ continue;
-ModulePass *llvm::createInternalizePass(bool InternalizeEverything) {
- return new InternalizePass(InternalizeEverything);
+ ++NumGlobals;
+ DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n");
+ }
+
+ // Mark all aliases that are not in the api as internal as well.
+ for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
+ I != E; ++I) {
+ if (!maybeInternalize(*I, ExternalComdats))
+ continue;
+
+ ++NumAliases;
+ DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n");
+ }
+
+ // We do not keep track of whether this pass changed the module because
+ // it adds unnecessary complexity:
+ // 1) This pass will generally be near the start of the pass pipeline, so
+ // there will be no analyses to invalidate.
+ // 2) This pass will most likely end up changing the module and it isn't worth
+ // worrying about optimizing the case where the module is unchanged.
+ return true;
}
-ModulePass *llvm::createInternalizePass(const std::vector <const char *> &el) {
- return new InternalizePass(el);
+ModulePass *llvm::createInternalizePass() { return new InternalizePass(); }
+
+ModulePass *llvm::createInternalizePass(ArrayRef<const char *> ExportList) {
+ return new InternalizePass(ExportList);
}