//
// 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 file implements stripping symbols out of symbol tables.
+// The StripSymbols transformation implements code stripping. Specifically, it
+// can delete:
//
-// Specifically, this allows you to strip all of the symbols out of:
-// * All functions in a module
-// * All non-essential symbols in a module (all function symbols + all module
-// scope symbols)
-// * Debug information.
+// * names for virtual registers
+// * symbols for internal globals and functions
+// * debug information
//
-// Notice that:
-// * This pass makes code much less readable, so it should only be used in
-// situations where the 'strip' utility would be used (such as reducing
-// code size, and making it harder to reverse engineer code).
+// Note that this transformation makes code much less readable, so it should
+// only be used in situations where the 'strip' utility would be used, such as
+// reducing code size or making it harder to reverse engineer code.
//
//===----------------------------------------------------------------------===//
#include "llvm/Transforms/IPO.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Instructions.h"
-#include "llvm/Module.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DebugInfo.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/TypeFinder.h"
+#include "llvm/IR/ValueSymbolTable.h"
#include "llvm/Pass.h"
-#include "llvm/ValueSymbolTable.h"
-#include "llvm/TypeSymbolTable.h"
-#include "llvm/Support/Compiler.h"
+#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
namespace {
- class VISIBILITY_HIDDEN StripSymbols : public ModulePass {
+ class StripSymbols : public ModulePass {
bool OnlyDebugInfo;
public:
- static char ID; // Pass identifcation, replacement for typeid
- StripSymbols(bool ODI = false)
- : ModulePass((intptr_t)&ID), OnlyDebugInfo(ODI) {}
+ static char ID; // Pass identification, replacement for typeid
+ explicit StripSymbols(bool ODI = false)
+ : ModulePass(ID), OnlyDebugInfo(ODI) {
+ initializeStripSymbolsPass(*PassRegistry::getPassRegistry());
+ }
- virtual bool runOnModule(Module &M);
+ bool runOnModule(Module &M) override;
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesAll();
}
};
- char StripSymbols::ID = 0;
- RegisterPass<StripSymbols> X("strip", "Strip all symbols from a module");
+ class StripNonDebugSymbols : public ModulePass {
+ public:
+ static char ID; // Pass identification, replacement for typeid
+ explicit StripNonDebugSymbols()
+ : ModulePass(ID) {
+ initializeStripNonDebugSymbolsPass(*PassRegistry::getPassRegistry());
+ }
+
+ bool runOnModule(Module &M) override;
+
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.setPreservesAll();
+ }
+ };
+
+ class StripDebugDeclare : public ModulePass {
+ public:
+ static char ID; // Pass identification, replacement for typeid
+ explicit StripDebugDeclare()
+ : ModulePass(ID) {
+ initializeStripDebugDeclarePass(*PassRegistry::getPassRegistry());
+ }
+
+ bool runOnModule(Module &M) override;
+
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.setPreservesAll();
+ }
+ };
+
+ class StripDeadDebugInfo : public ModulePass {
+ public:
+ static char ID; // Pass identification, replacement for typeid
+ explicit StripDeadDebugInfo()
+ : ModulePass(ID) {
+ initializeStripDeadDebugInfoPass(*PassRegistry::getPassRegistry());
+ }
+
+ bool runOnModule(Module &M) override;
+
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.setPreservesAll();
+ }
+ };
}
+char StripSymbols::ID = 0;
+INITIALIZE_PASS(StripSymbols, "strip",
+ "Strip all symbols from a module", false, false)
+
ModulePass *llvm::createStripSymbolsPass(bool OnlyDebugInfo) {
return new StripSymbols(OnlyDebugInfo);
}
+char StripNonDebugSymbols::ID = 0;
+INITIALIZE_PASS(StripNonDebugSymbols, "strip-nondebug",
+ "Strip all symbols, except dbg symbols, from a module",
+ false, false)
+
+ModulePass *llvm::createStripNonDebugSymbolsPass() {
+ return new StripNonDebugSymbols();
+}
+
+char StripDebugDeclare::ID = 0;
+INITIALIZE_PASS(StripDebugDeclare, "strip-debug-declare",
+ "Strip all llvm.dbg.declare intrinsics", false, false)
+
+ModulePass *llvm::createStripDebugDeclarePass() {
+ return new StripDebugDeclare();
+}
+
+char StripDeadDebugInfo::ID = 0;
+INITIALIZE_PASS(StripDeadDebugInfo, "strip-dead-debug-info",
+ "Strip debug info for unused symbols", false, false)
+
+ModulePass *llvm::createStripDeadDebugInfoPass() {
+ return new StripDeadDebugInfo();
+}
+
+/// OnlyUsedBy - Return true if V is only used by Usr.
+static bool OnlyUsedBy(Value *V, Value *Usr) {
+ for (User *U : V->users())
+ if (U != Usr)
+ return false;
+
+ return true;
+}
+
static void RemoveDeadConstant(Constant *C) {
assert(C->use_empty() && "Constant is not dead!");
- std::vector<Constant*> Operands;
+ SmallPtrSet<Constant*, 4> Operands;
for (unsigned i = 0, e = C->getNumOperands(); i != e; ++i)
- if (isa<DerivedType>(C->getOperand(i)->getType()) &&
- C->getOperand(i)->hasOneUse())
- Operands.push_back(C->getOperand(i));
+ if (OnlyUsedBy(C->getOperand(i), C))
+ Operands.insert(cast<Constant>(C->getOperand(i)));
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) {
- if (!GV->hasInternalLinkage()) return; // Don't delete non static globals.
+ if (!GV->hasLocalLinkage()) return; // Don't delete non-static globals.
GV->eraseFromParent();
}
else if (!isa<Function>(C))
- C->destroyConstant();
+ if (isa<CompositeType>(C->getType()))
+ C->destroyConstant();
// If the constant referenced anything, see if we can delete it as well.
- while (!Operands.empty()) {
- RemoveDeadConstant(Operands.back());
- Operands.pop_back();
- }
+ for (Constant *O : Operands)
+ RemoveDeadConstant(O);
}
// Strip the symbol table of its names.
//
-static void StripSymtab(ValueSymbolTable &ST) {
+static void StripSymtab(ValueSymbolTable &ST, bool PreserveDbgInfo) {
for (ValueSymbolTable::iterator VI = ST.begin(), VE = ST.end(); VI != VE; ) {
Value *V = VI->getValue();
++VI;
- if (!isa<GlobalValue>(V) || cast<GlobalValue>(V)->hasInternalLinkage()) {
- // Set name to "", removing from symbol table!
- V->setName("");
+ if (!isa<GlobalValue>(V) || cast<GlobalValue>(V)->hasLocalLinkage()) {
+ if (!PreserveDbgInfo || !V->getName().startswith("llvm.dbg"))
+ // Set name to "", removing from symbol table!
+ V->setName("");
}
}
}
-// Strip the symbol table of its names.
-static void StripTypeSymtab(TypeSymbolTable &ST) {
- for (TypeSymbolTable::iterator TI = ST.begin(), E = ST.end(); TI != E; )
- ST.remove(TI++);
+// Strip any named types of their names.
+static void StripTypeNames(Module &M, bool PreserveDbgInfo) {
+ TypeFinder StructTypes;
+ StructTypes.run(M, false);
+
+ for (unsigned i = 0, e = StructTypes.size(); i != e; ++i) {
+ StructType *STy = StructTypes[i];
+ if (STy->isLiteral() || STy->getName().empty()) continue;
+
+ if (PreserveDbgInfo && STy->getName().startswith("llvm.dbg"))
+ continue;
+
+ STy->setName("");
+ }
}
+/// Find values that are marked as llvm.used.
+static void findUsedValues(GlobalVariable *LLVMUsed,
+ SmallPtrSetImpl<const GlobalValue*> &UsedValues) {
+ if (!LLVMUsed) return;
+ UsedValues.insert(LLVMUsed);
+ ConstantArray *Inits = cast<ConstantArray>(LLVMUsed->getInitializer());
-bool StripSymbols::runOnModule(Module &M) {
- // If we're not just stripping debug info, strip all symbols from the
- // functions and the names from any internal globals.
- if (!OnlyDebugInfo) {
- for (Module::global_iterator I = M.global_begin(), E = M.global_end();
- I != E; ++I)
- if (I->hasInternalLinkage())
+ for (unsigned i = 0, e = Inits->getNumOperands(); i != e; ++i)
+ if (GlobalValue *GV =
+ dyn_cast<GlobalValue>(Inits->getOperand(i)->stripPointerCasts()))
+ UsedValues.insert(GV);
+}
+
+/// StripSymbolNames - Strip symbol names.
+static bool StripSymbolNames(Module &M, bool PreserveDbgInfo) {
+
+ SmallPtrSet<const GlobalValue*, 8> llvmUsedValues;
+ findUsedValues(M.getGlobalVariable("llvm.used"), llvmUsedValues);
+ findUsedValues(M.getGlobalVariable("llvm.compiler.used"), llvmUsedValues);
+
+ for (Module::global_iterator I = M.global_begin(), E = M.global_end();
+ I != E; ++I) {
+ if (I->hasLocalLinkage() && llvmUsedValues.count(I) == 0)
+ if (!PreserveDbgInfo || !I->getName().startswith("llvm.dbg"))
I->setName(""); // Internal symbols can't participate in linkage
+ }
- for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
- if (I->hasInternalLinkage())
+ for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
+ if (I->hasLocalLinkage() && llvmUsedValues.count(I) == 0)
+ if (!PreserveDbgInfo || !I->getName().startswith("llvm.dbg"))
I->setName(""); // Internal symbols can't participate in linkage
- StripSymtab(I->getValueSymbolTable());
- }
-
- // Remove all names from types.
- StripTypeSymtab(M.getTypeSymbolTable());
+ StripSymtab(I->getValueSymbolTable(), PreserveDbgInfo);
}
- // Strip debug info in the module if it exists. To do this, we remove
- // llvm.dbg.func.start, llvm.dbg.stoppoint, and llvm.dbg.region.end calls, and
- // any globals they point to if now dead.
- Function *FuncStart = M.getFunction("llvm.dbg.func.start");
- Function *StopPoint = M.getFunction("llvm.dbg.stoppoint");
- Function *RegionStart = M.getFunction("llvm.dbg.region.start");
- Function *RegionEnd = M.getFunction("llvm.dbg.region.end");
- Function *Declare = M.getFunction("llvm.dbg.declare");
- if (!FuncStart && !StopPoint && !RegionStart && !RegionEnd && !Declare)
- return true;
+ // Remove all names from types.
+ StripTypeNames(M, PreserveDbgInfo);
- std::vector<GlobalVariable*> DeadGlobals;
+ return true;
+}
+
+bool StripSymbols::runOnModule(Module &M) {
+ bool Changed = false;
+ Changed |= StripDebugInfo(M);
+ if (!OnlyDebugInfo)
+ Changed |= StripSymbolNames(M, false);
+ return Changed;
+}
+
+bool StripNonDebugSymbols::runOnModule(Module &M) {
+ return StripSymbolNames(M, true);
+}
+
+bool StripDebugDeclare::runOnModule(Module &M) {
+
+ Function *Declare = M.getFunction("llvm.dbg.declare");
+ std::vector<Constant*> DeadConstants;
- // Remove all of the calls to the debugger intrinsics, and remove them from
- // the module.
- if (FuncStart) {
- while (!FuncStart->use_empty()) {
- CallInst *CI = cast<CallInst>(FuncStart->use_back());
- Value *Arg = CI->getOperand(1);
- assert(CI->use_empty() && "llvm.dbg intrinsic should have void result");
- CI->eraseFromParent();
- if (Arg->use_empty())
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Arg))
- DeadGlobals.push_back(GV);
- }
- FuncStart->eraseFromParent();
- }
- if (StopPoint) {
- while (!StopPoint->use_empty()) {
- CallInst *CI = cast<CallInst>(StopPoint->use_back());
- Value *Arg = CI->getOperand(3);
- assert(CI->use_empty() && "llvm.dbg intrinsic should have void result");
- CI->eraseFromParent();
- if (Arg->use_empty())
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Arg))
- DeadGlobals.push_back(GV);
- }
- StopPoint->eraseFromParent();
- }
- if (RegionStart) {
- while (!RegionStart->use_empty()) {
- CallInst *CI = cast<CallInst>(RegionStart->use_back());
- Value *Arg = CI->getOperand(1);
- assert(CI->use_empty() && "llvm.dbg intrinsic should have void result");
- CI->eraseFromParent();
- if (Arg->use_empty())
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Arg))
- DeadGlobals.push_back(GV);
- }
- RegionStart->eraseFromParent();
- }
- if (RegionEnd) {
- while (!RegionEnd->use_empty()) {
- CallInst *CI = cast<CallInst>(RegionEnd->use_back());
- Value *Arg = CI->getOperand(1);
- assert(CI->use_empty() && "llvm.dbg intrinsic should have void result");
- CI->eraseFromParent();
- if (Arg->use_empty())
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Arg))
- DeadGlobals.push_back(GV);
- }
- RegionEnd->eraseFromParent();
- }
if (Declare) {
while (!Declare->use_empty()) {
- CallInst *CI = cast<CallInst>(Declare->use_back());
- Value *Arg = CI->getOperand(2);
+ CallInst *CI = cast<CallInst>(Declare->user_back());
+ Value *Arg1 = CI->getArgOperand(0);
+ Value *Arg2 = CI->getArgOperand(1);
assert(CI->use_empty() && "llvm.dbg intrinsic should have void result");
CI->eraseFromParent();
- if (Arg->use_empty())
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Arg))
- DeadGlobals.push_back(GV);
+ if (Arg1->use_empty()) {
+ if (Constant *C = dyn_cast<Constant>(Arg1))
+ DeadConstants.push_back(C);
+ else
+ RecursivelyDeleteTriviallyDeadInstructions(Arg1);
+ }
+ if (Arg2->use_empty())
+ if (Constant *C = dyn_cast<Constant>(Arg2))
+ DeadConstants.push_back(C);
}
Declare->eraseFromParent();
}
- // Finally, delete any internal globals that were only used by the debugger
- // intrinsics.
- while (!DeadGlobals.empty()) {
- GlobalVariable *GV = DeadGlobals.back();
- DeadGlobals.pop_back();
- if (GV->hasInternalLinkage())
- RemoveDeadConstant(GV);
+ while (!DeadConstants.empty()) {
+ Constant *C = DeadConstants.back();
+ DeadConstants.pop_back();
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) {
+ if (GV->hasLocalLinkage())
+ RemoveDeadConstant(GV);
+ } else
+ RemoveDeadConstant(C);
}
return true;
}
+
+/// Remove any debug info for global variables/functions in the given module for
+/// which said global variable/function no longer exists (i.e. is null).
+///
+/// Debugging information is encoded in llvm IR using metadata. This is designed
+/// such a way that debug info for symbols preserved even if symbols are
+/// optimized away by the optimizer. This special pass removes debug info for
+/// such symbols.
+bool StripDeadDebugInfo::runOnModule(Module &M) {
+ bool Changed = false;
+
+ LLVMContext &C = M.getContext();
+
+ // Find all debug info in F. This is actually overkill in terms of what we
+ // want to do, but we want to try and be as resilient as possible in the face
+ // of potential debug info changes by using the formal interfaces given to us
+ // as much as possible.
+ DebugInfoFinder F;
+ F.processModule(M);
+
+ // For each compile unit, find the live set of global variables/functions and
+ // replace the current list of potentially dead global variables/functions
+ // with the live list.
+ SmallVector<Metadata *, 64> LiveGlobalVariables;
+ SmallVector<Metadata *, 64> LiveSubprograms;
+ DenseSet<const MDNode *> VisitedSet;
+
+ for (MDCompileUnit *DIC : F.compile_units()) {
+ // Create our live subprogram list.
+ MDSubprogramArray SPs = DIC->getSubprograms();
+ bool SubprogramChange = false;
+ for (unsigned i = 0, e = SPs.size(); i != e; ++i) {
+ DISubprogram DISP = SPs[i];
+
+ // Make sure we visit each subprogram only once.
+ if (!VisitedSet.insert(DISP).second)
+ continue;
+
+ // If the function referenced by DISP is not null, the function is live.
+ if (DISP->getFunction())
+ LiveSubprograms.push_back(DISP);
+ else
+ SubprogramChange = true;
+ }
+
+ // Create our live global variable list.
+ MDGlobalVariableArray GVs = DIC->getGlobalVariables();
+ bool GlobalVariableChange = false;
+ for (unsigned i = 0, e = GVs.size(); i != e; ++i) {
+ DIGlobalVariable DIG = GVs[i];
+
+ // Make sure we only visit each global variable only once.
+ if (!VisitedSet.insert(DIG).second)
+ continue;
+
+ // If the global variable referenced by DIG is not null, the global
+ // variable is live.
+ if (DIG->getVariable())
+ LiveGlobalVariables.push_back(DIG);
+ else
+ GlobalVariableChange = true;
+ }
+
+ // If we found dead subprograms or global variables, replace the current
+ // subprogram list/global variable list with our new live subprogram/global
+ // variable list.
+ if (SubprogramChange) {
+ DIC->replaceSubprograms(MDTuple::get(C, LiveSubprograms));
+ Changed = true;
+ }
+
+ if (GlobalVariableChange) {
+ DIC->replaceGlobalVariables(MDTuple::get(C, LiveGlobalVariables));
+ Changed = true;
+ }
+
+ // Reset lists for the next iteration.
+ LiveSubprograms.clear();
+ LiveGlobalVariables.clear();
+ }
+
+ return Changed;
+}