X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FIPO%2FLowerBitSets.cpp;h=65d082b27a8460606c6cc7c3ad863b0d2d755a7d;hb=b2bc86f25144aa907b977e27fd93e316e15500d6;hp=fae75a8ac418458e1749a24b592dead2e249a770;hpb=19443c1bcb863ba186abfe0bda3a1603488d17f7;p=oota-llvm.git diff --git a/lib/Transforms/IPO/LowerBitSets.cpp b/lib/Transforms/IPO/LowerBitSets.cpp index fae75a8ac41..65d082b27a8 100644 --- a/lib/Transforms/IPO/LowerBitSets.cpp +++ b/lib/Transforms/IPO/LowerBitSets.cpp @@ -19,6 +19,8 @@ #include "llvm/ADT/Triple.h" #include "llvm/IR/Constant.h" #include "llvm/IR/Constants.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/GlobalObject.h" #include "llvm/IR/GlobalVariable.h" #include "llvm/IR/IRBuilder.h" #include "llvm/IR/Instructions.h" @@ -26,6 +28,8 @@ #include "llvm/IR/Module.h" #include "llvm/IR/Operator.h" #include "llvm/Pass.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/raw_ostream.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" using namespace llvm; @@ -59,9 +63,9 @@ bool BitSetInfo::containsGlobalOffset(uint64_t Offset) const { bool BitSetInfo::containsValue( const DataLayout &DL, - const DenseMap &GlobalLayout, Value *V, + const DenseMap &GlobalLayout, Value *V, uint64_t COffset) const { - if (auto GV = dyn_cast(V)) { + if (auto GV = dyn_cast(V)) { auto I = GlobalLayout.find(GV); if (I == GlobalLayout.end()) return false; @@ -90,6 +94,21 @@ bool BitSetInfo::containsValue( return false; } +void BitSetInfo::print(raw_ostream &OS) const { + OS << "offset " << ByteOffset << " size " << BitSize << " align " + << (1 << AlignLog2); + + if (isAllOnes()) { + OS << " all-ones\n"; + return; + } + + OS << " { "; + for (uint64_t B : Bits) + OS << B << ' '; + OS << "}\n"; +} + BitSetInfo BitSetBuilder::build() { if (Min > Max) Min = 0; @@ -193,34 +212,48 @@ struct LowerBitSets : public ModulePass { Module *M; bool LinkerSubsectionsViaSymbols; + Triple::ArchType Arch; + Triple::ObjectFormatType ObjectFormat; IntegerType *Int1Ty; IntegerType *Int8Ty; IntegerType *Int32Ty; Type *Int32PtrTy; IntegerType *Int64Ty; - Type *IntPtrTy; + IntegerType *IntPtrTy; // The llvm.bitsets named metadata. NamedMDNode *BitSetNM; - // Mapping from bitset mdstrings to the call sites that test them. - DenseMap> BitSetTestCallSites; + // Mapping from bitset identifiers to the call sites that test them. + DenseMap> BitSetTestCallSites; std::vector ByteArrayInfos; BitSetInfo - buildBitSet(MDString *BitSet, - const DenseMap &GlobalLayout); + buildBitSet(Metadata *BitSet, + const DenseMap &GlobalLayout); ByteArrayInfo *createByteArray(BitSetInfo &BSI); void allocateByteArrays(); Value *createBitSetTest(IRBuilder<> &B, BitSetInfo &BSI, ByteArrayInfo *&BAI, Value *BitOffset); + void lowerBitSetCalls(ArrayRef BitSets, + Constant *CombinedGlobalAddr, + const DenseMap &GlobalLayout); Value * lowerBitSetCall(CallInst *CI, BitSetInfo &BSI, ByteArrayInfo *&BAI, - GlobalVariable *CombinedGlobal, - const DenseMap &GlobalLayout); - void buildBitSetsFromGlobals(const std::vector &BitSets, - const std::vector &Globals); + Constant *CombinedGlobal, + const DenseMap &GlobalLayout); + void buildBitSetsFromGlobalVariables(ArrayRef BitSets, + ArrayRef Globals); + unsigned getJumpTableEntrySize(); + Type *getJumpTableEntryType(); + Constant *createJumpTableEntry(GlobalObject *Src, Function *Dest, + unsigned Distance); + void verifyBitSetMDNode(MDNode *Op); + void buildBitSetsFromFunctions(ArrayRef BitSets, + ArrayRef Functions); + void buildBitSetsFromDisjointSet(ArrayRef BitSets, + ArrayRef Globals); bool buildBitSets(); bool eraseBitSetMetadata(); @@ -228,7 +261,7 @@ struct LowerBitSets : public ModulePass { bool runOnModule(Module &M) override; }; -} // namespace +} // anonymous namespace INITIALIZE_PASS_BEGIN(LowerBitSets, "lowerbitsets", "Lower bitset metadata", false, false) @@ -244,6 +277,8 @@ bool LowerBitSets::doInitialization(Module &Mod) { Triple TargetTriple(M->getTargetTriple()); LinkerSubsectionsViaSymbols = TargetTriple.isMacOSX(); + Arch = TargetTriple.getArch(); + ObjectFormat = TargetTriple.getObjectFormat(); Int1Ty = Type::getInt1Ty(M->getContext()); Int8Ty = Type::getInt8Ty(M->getContext()); @@ -262,8 +297,8 @@ bool LowerBitSets::doInitialization(Module &Mod) { /// Build a bit set for BitSet using the object layouts in /// GlobalLayout. BitSetInfo LowerBitSets::buildBitSet( - MDString *BitSet, - const DenseMap &GlobalLayout) { + Metadata *BitSet, + const DenseMap &GlobalLayout) { BitSetBuilder BSB; // Compute the byte offset of each element of this bitset. @@ -271,8 +306,13 @@ BitSetInfo LowerBitSets::buildBitSet( for (MDNode *Op : BitSetNM->operands()) { if (Op->getOperand(0) != BitSet || !Op->getOperand(1)) continue; - auto OpGlobal = cast( - cast(Op->getOperand(1))->getValue()); + Constant *OpConst = + cast(Op->getOperand(1))->getValue(); + if (auto GA = dyn_cast(OpConst)) + OpConst = GA->getAliasee(); + auto OpGlobal = dyn_cast(OpConst); + if (!OpGlobal) + continue; uint64_t Offset = cast(cast(Op->getOperand(2)) ->getValue())->getZExtValue(); @@ -396,16 +436,17 @@ Value *LowerBitSets::createBitSetTest(IRBuilder<> &B, BitSetInfo &BSI, } Constant *ByteArray = BAI->ByteArray; + Type *Ty = BAI->ByteArray->getValueType(); if (!LinkerSubsectionsViaSymbols && AvoidReuse) { // Each use of the byte array uses a different alias. This makes the // backend less likely to reuse previously computed byte array addresses, // improving the security of the CFI mechanism based on this pass. - ByteArray = GlobalAlias::create( - BAI->ByteArray->getType()->getElementType(), 0, - GlobalValue::PrivateLinkage, "bits_use", ByteArray, M); + ByteArray = GlobalAlias::create(BAI->ByteArray->getValueType(), 0, + GlobalValue::PrivateLinkage, "bits_use", + ByteArray, M); } - Value *ByteAddr = B.CreateGEP(ByteArray, BitOffset); + Value *ByteAddr = B.CreateGEP(Ty, ByteArray, BitOffset); Value *Byte = B.CreateLoad(ByteAddr); Value *ByteAndMask = B.CreateAnd(Byte, BAI->Mask); @@ -417,17 +458,16 @@ Value *LowerBitSets::createBitSetTest(IRBuilder<> &B, BitSetInfo &BSI, /// replace the call with. Value *LowerBitSets::lowerBitSetCall( CallInst *CI, BitSetInfo &BSI, ByteArrayInfo *&BAI, - GlobalVariable *CombinedGlobal, - const DenseMap &GlobalLayout) { + Constant *CombinedGlobalIntAddr, + const DenseMap &GlobalLayout) { Value *Ptr = CI->getArgOperand(0); const DataLayout &DL = M->getDataLayout(); if (BSI.containsValue(DL, GlobalLayout, Ptr)) - return ConstantInt::getTrue(CombinedGlobal->getParent()->getContext()); + return ConstantInt::getTrue(M->getContext()); - Constant *GlobalAsInt = ConstantExpr::getPtrToInt(CombinedGlobal, IntPtrTy); Constant *OffsetedGlobalAsInt = ConstantExpr::getAdd( - GlobalAsInt, ConstantInt::get(IntPtrTy, BSI.ByteOffset)); + CombinedGlobalIntAddr, ConstantInt::get(IntPtrTy, BSI.ByteOffset)); BasicBlock *InitialBB = CI->getParent(); @@ -486,18 +526,19 @@ Value *LowerBitSets::lowerBitSetCall( /// Given a disjoint set of bitsets and globals, layout the globals, build the /// bit sets and lower the llvm.bitset.test calls. -void LowerBitSets::buildBitSetsFromGlobals( - const std::vector &BitSets, - const std::vector &Globals) { +void LowerBitSets::buildBitSetsFromGlobalVariables( + ArrayRef BitSets, ArrayRef Globals) { // Build a new global with the combined contents of the referenced globals. + // This global is a struct whose even-indexed elements contain the original + // contents of the referenced globals and whose odd-indexed elements contain + // any padding required to align the next element to the next power of 2. std::vector GlobalInits; const DataLayout &DL = M->getDataLayout(); for (GlobalVariable *G : Globals) { GlobalInits.push_back(G->getInitializer()); - uint64_t InitSize = DL.getTypeAllocSize(G->getInitializer()->getType()); + uint64_t InitSize = DL.getTypeAllocSize(G->getValueType()); - // Compute the amount of padding required to align the next element to the - // next power of 2. + // Compute the amount of padding required. uint64_t Padding = NextPowerOf2(InitSize - 1) - InitSize; // Cap at 128 was found experimentally to have a good data/instruction @@ -511,34 +552,20 @@ void LowerBitSets::buildBitSetsFromGlobals( if (!GlobalInits.empty()) GlobalInits.pop_back(); Constant *NewInit = ConstantStruct::getAnon(M->getContext(), GlobalInits); - auto CombinedGlobal = + auto *CombinedGlobal = new GlobalVariable(*M, NewInit->getType(), /*isConstant=*/true, GlobalValue::PrivateLinkage, NewInit); - const StructLayout *CombinedGlobalLayout = - DL.getStructLayout(cast(NewInit->getType())); + StructType *NewTy = cast(NewInit->getType()); + const StructLayout *CombinedGlobalLayout = DL.getStructLayout(NewTy); // Compute the offsets of the original globals within the new global. - DenseMap GlobalLayout; + DenseMap GlobalLayout; for (unsigned I = 0; I != Globals.size(); ++I) // Multiply by 2 to account for padding elements. GlobalLayout[Globals[I]] = CombinedGlobalLayout->getElementOffset(I * 2); - // For each bitset in this disjoint set... - for (MDString *BS : BitSets) { - // Build the bitset. - BitSetInfo BSI = buildBitSet(BS, GlobalLayout); - - ByteArrayInfo *BAI = 0; - - // Lower each call to llvm.bitset.test for this bitset. - for (CallInst *CI : BitSetTestCallSites[BS]) { - ++NumBitSetCallsLowered; - Value *Lowered = lowerBitSetCall(CI, BSI, BAI, CombinedGlobal, GlobalLayout); - CI->replaceAllUsesWith(Lowered); - CI->eraseFromParent(); - } - } + lowerBitSetCalls(BitSets, CombinedGlobal, GlobalLayout); // Build aliases pointing to offsets into the combined global for each // global from which we built the combined global, and replace references @@ -552,10 +579,11 @@ void LowerBitSets::buildBitSetsFromGlobals( if (LinkerSubsectionsViaSymbols) { Globals[I]->replaceAllUsesWith(CombinedGlobalElemPtr); } else { - GlobalAlias *GAlias = GlobalAlias::create( - Globals[I]->getType()->getElementType(), - Globals[I]->getType()->getAddressSpace(), Globals[I]->getLinkage(), - "", CombinedGlobalElemPtr, M); + assert(Globals[I]->getType()->getAddressSpace() == 0); + GlobalAlias *GAlias = GlobalAlias::create(NewTy->getElementType(I * 2), 0, + Globals[I]->getLinkage(), "", + CombinedGlobalElemPtr, M); + GAlias->setVisibility(Globals[I]->getVisibility()); GAlias->takeName(Globals[I]); Globals[I]->replaceAllUsesWith(GAlias); } @@ -563,6 +591,331 @@ void LowerBitSets::buildBitSetsFromGlobals( } } +void LowerBitSets::lowerBitSetCalls( + ArrayRef BitSets, Constant *CombinedGlobalAddr, + const DenseMap &GlobalLayout) { + Constant *CombinedGlobalIntAddr = + ConstantExpr::getPtrToInt(CombinedGlobalAddr, IntPtrTy); + + // For each bitset in this disjoint set... + for (Metadata *BS : BitSets) { + // Build the bitset. + BitSetInfo BSI = buildBitSet(BS, GlobalLayout); + DEBUG({ + if (auto BSS = dyn_cast(BS)) + dbgs() << BSS->getString() << ": "; + else + dbgs() << ": "; + BSI.print(dbgs()); + }); + + ByteArrayInfo *BAI = nullptr; + + // Lower each call to llvm.bitset.test for this bitset. + for (CallInst *CI : BitSetTestCallSites[BS]) { + ++NumBitSetCallsLowered; + Value *Lowered = + lowerBitSetCall(CI, BSI, BAI, CombinedGlobalIntAddr, GlobalLayout); + CI->replaceAllUsesWith(Lowered); + CI->eraseFromParent(); + } + } +} + +void LowerBitSets::verifyBitSetMDNode(MDNode *Op) { + if (Op->getNumOperands() != 3) + report_fatal_error( + "All operands of llvm.bitsets metadata must have 3 elements"); + if (!Op->getOperand(1)) + return; + + auto OpConstMD = dyn_cast(Op->getOperand(1)); + if (!OpConstMD) + report_fatal_error("Bit set element must be a constant"); + auto OpGlobal = dyn_cast(OpConstMD->getValue()); + if (!OpGlobal) + return; + + if (OpGlobal->isThreadLocal()) + report_fatal_error("Bit set element may not be thread-local"); + if (OpGlobal->hasSection()) + report_fatal_error("Bit set element may not have an explicit section"); + + if (isa(OpGlobal) && OpGlobal->isDeclarationForLinker()) + report_fatal_error("Bit set global var element must be a definition"); + + auto OffsetConstMD = dyn_cast(Op->getOperand(2)); + if (!OffsetConstMD) + report_fatal_error("Bit set element offset must be a constant"); + auto OffsetInt = dyn_cast(OffsetConstMD->getValue()); + if (!OffsetInt) + report_fatal_error("Bit set element offset must be an integer constant"); +} + +static const unsigned kX86JumpTableEntrySize = 8; + +unsigned LowerBitSets::getJumpTableEntrySize() { + if (Arch != Triple::x86 && Arch != Triple::x86_64) + report_fatal_error("Unsupported architecture for jump tables"); + + return kX86JumpTableEntrySize; +} + +// Create a constant representing a jump table entry for the target. This +// consists of an instruction sequence containing a relative branch to Dest. The +// constant will be laid out at address Src+(Len*Distance) where Len is the +// target-specific jump table entry size. +Constant *LowerBitSets::createJumpTableEntry(GlobalObject *Src, Function *Dest, + unsigned Distance) { + if (Arch != Triple::x86 && Arch != Triple::x86_64) + report_fatal_error("Unsupported architecture for jump tables"); + + const unsigned kJmpPCRel32Code = 0xe9; + const unsigned kInt3Code = 0xcc; + + ConstantInt *Jmp = ConstantInt::get(Int8Ty, kJmpPCRel32Code); + + // Build a constant representing the displacement between the constant's + // address and Dest. This will resolve to a PC32 relocation referring to Dest. + Constant *DestInt = ConstantExpr::getPtrToInt(Dest, IntPtrTy); + Constant *SrcInt = ConstantExpr::getPtrToInt(Src, IntPtrTy); + Constant *Disp = ConstantExpr::getSub(DestInt, SrcInt); + ConstantInt *DispOffset = + ConstantInt::get(IntPtrTy, Distance * kX86JumpTableEntrySize + 5); + Constant *OffsetedDisp = ConstantExpr::getSub(Disp, DispOffset); + OffsetedDisp = ConstantExpr::getTrunc(OffsetedDisp, Int32Ty); + + ConstantInt *Int3 = ConstantInt::get(Int8Ty, kInt3Code); + + Constant *Fields[] = { + Jmp, OffsetedDisp, Int3, Int3, Int3, + }; + return ConstantStruct::getAnon(Fields, /*Packed=*/true); +} + +Type *LowerBitSets::getJumpTableEntryType() { + if (Arch != Triple::x86 && Arch != Triple::x86_64) + report_fatal_error("Unsupported architecture for jump tables"); + + return StructType::get(M->getContext(), + {Int8Ty, Int32Ty, Int8Ty, Int8Ty, Int8Ty}, + /*Packed=*/true); +} + +/// Given a disjoint set of bitsets and functions, build a jump table for the +/// functions, build the bit sets and lower the llvm.bitset.test calls. +void LowerBitSets::buildBitSetsFromFunctions(ArrayRef BitSets, + ArrayRef Functions) { + // Unlike the global bitset builder, the function bitset builder cannot + // re-arrange functions in a particular order and base its calculations on the + // layout of the functions' entry points, as we have no idea how large a + // particular function will end up being (the size could even depend on what + // this pass does!) Instead, we build a jump table, which is a block of code + // consisting of one branch instruction for each of the functions in the bit + // set that branches to the target function, and redirect any taken function + // addresses to the corresponding jump table entry. In the object file's + // symbol table, the symbols for the target functions also refer to the jump + // table entries, so that addresses taken outside the module will pass any + // verification done inside the module. + // + // In more concrete terms, suppose we have three functions f, g, h which are + // members of a single bitset, and a function foo that returns their + // addresses: + // + // f: + // mov 0, %eax + // ret + // + // g: + // mov 1, %eax + // ret + // + // h: + // mov 2, %eax + // ret + // + // foo: + // mov f, %eax + // mov g, %edx + // mov h, %ecx + // ret + // + // To create a jump table for these functions, we instruct the LLVM code + // generator to output a jump table in the .text section. This is done by + // representing the instructions in the jump table as an LLVM constant and + // placing them in a global variable in the .text section. The end result will + // (conceptually) look like this: + // + // f: + // jmp .Ltmp0 ; 5 bytes + // int3 ; 1 byte + // int3 ; 1 byte + // int3 ; 1 byte + // + // g: + // jmp .Ltmp1 ; 5 bytes + // int3 ; 1 byte + // int3 ; 1 byte + // int3 ; 1 byte + // + // h: + // jmp .Ltmp2 ; 5 bytes + // int3 ; 1 byte + // int3 ; 1 byte + // int3 ; 1 byte + // + // .Ltmp0: + // mov 0, %eax + // ret + // + // .Ltmp1: + // mov 1, %eax + // ret + // + // .Ltmp2: + // mov 2, %eax + // ret + // + // foo: + // mov f, %eax + // mov g, %edx + // mov h, %ecx + // ret + // + // Because the addresses of f, g, h are evenly spaced at a power of 2, in the + // normal case the check can be carried out using the same kind of simple + // arithmetic that we normally use for globals. + + assert(!Functions.empty()); + + // Build a simple layout based on the regular layout of jump tables. + DenseMap GlobalLayout; + unsigned EntrySize = getJumpTableEntrySize(); + for (unsigned I = 0; I != Functions.size(); ++I) + GlobalLayout[Functions[I]] = I * EntrySize; + + // Create a constant to hold the jump table. + ArrayType *JumpTableType = + ArrayType::get(getJumpTableEntryType(), Functions.size()); + auto JumpTable = new GlobalVariable(*M, JumpTableType, + /*isConstant=*/true, + GlobalValue::PrivateLinkage, nullptr); + JumpTable->setSection(ObjectFormat == Triple::MachO + ? "__TEXT,__text,regular,pure_instructions" + : ".text"); + lowerBitSetCalls(BitSets, JumpTable, GlobalLayout); + + // Build aliases pointing to offsets into the jump table, and replace + // references to the original functions with references to the aliases. + for (unsigned I = 0; I != Functions.size(); ++I) { + Constant *CombinedGlobalElemPtr = ConstantExpr::getBitCast( + ConstantExpr::getGetElementPtr( + JumpTableType, JumpTable, + ArrayRef{ConstantInt::get(IntPtrTy, 0), + ConstantInt::get(IntPtrTy, I)}), + Functions[I]->getType()); + if (LinkerSubsectionsViaSymbols || Functions[I]->isDeclarationForLinker()) { + Functions[I]->replaceAllUsesWith(CombinedGlobalElemPtr); + } else { + assert(Functions[I]->getType()->getAddressSpace() == 0); + GlobalAlias *GAlias = GlobalAlias::create(Functions[I]->getValueType(), 0, + Functions[I]->getLinkage(), "", + CombinedGlobalElemPtr, M); + GAlias->setVisibility(Functions[I]->getVisibility()); + GAlias->takeName(Functions[I]); + Functions[I]->replaceAllUsesWith(GAlias); + } + if (!Functions[I]->isDeclarationForLinker()) + Functions[I]->setLinkage(GlobalValue::PrivateLinkage); + } + + // Build and set the jump table's initializer. + std::vector JumpTableEntries; + for (unsigned I = 0; I != Functions.size(); ++I) + JumpTableEntries.push_back( + createJumpTableEntry(JumpTable, Functions[I], I)); + JumpTable->setInitializer( + ConstantArray::get(JumpTableType, JumpTableEntries)); +} + +void LowerBitSets::buildBitSetsFromDisjointSet( + ArrayRef BitSets, ArrayRef Globals) { + llvm::DenseMap BitSetIndices; + llvm::DenseMap GlobalIndices; + for (unsigned I = 0; I != BitSets.size(); ++I) + BitSetIndices[BitSets[I]] = I; + for (unsigned I = 0; I != Globals.size(); ++I) + GlobalIndices[Globals[I]] = I; + + // For each bitset, build a set of indices that refer to globals referenced by + // the bitset. + std::vector> BitSetMembers(BitSets.size()); + if (BitSetNM) { + for (MDNode *Op : BitSetNM->operands()) { + // Op = { bitset name, global, offset } + if (!Op->getOperand(1)) + continue; + auto I = BitSetIndices.find(Op->getOperand(0)); + if (I == BitSetIndices.end()) + continue; + + auto OpGlobal = dyn_cast( + cast(Op->getOperand(1))->getValue()); + if (!OpGlobal) + continue; + BitSetMembers[I->second].insert(GlobalIndices[OpGlobal]); + } + } + + // Order the sets of indices by size. The GlobalLayoutBuilder works best + // when given small index sets first. + std::stable_sort( + BitSetMembers.begin(), BitSetMembers.end(), + [](const std::set &O1, const std::set &O2) { + return O1.size() < O2.size(); + }); + + // Create a GlobalLayoutBuilder and provide it with index sets as layout + // fragments. The GlobalLayoutBuilder tries to lay out members of fragments as + // close together as possible. + GlobalLayoutBuilder GLB(Globals.size()); + for (auto &&MemSet : BitSetMembers) + GLB.addFragment(MemSet); + + // Build the bitsets from this disjoint set. + if (Globals.empty() || isa(Globals[0])) { + // Build a vector of global variables with the computed layout. + std::vector OrderedGVs(Globals.size()); + auto OGI = OrderedGVs.begin(); + for (auto &&F : GLB.Fragments) { + for (auto &&Offset : F) { + auto GV = dyn_cast(Globals[Offset]); + if (!GV) + report_fatal_error( + "Bit set may not contain both global variables and functions"); + *OGI++ = GV; + } + } + + buildBitSetsFromGlobalVariables(BitSets, OrderedGVs); + } else { + // Build a vector of functions with the computed layout. + std::vector OrderedFns(Globals.size()); + auto OFI = OrderedFns.begin(); + for (auto &&F : GLB.Fragments) { + for (auto &&Offset : F) { + auto Fn = dyn_cast(Globals[Offset]); + if (!Fn) + report_fatal_error( + "Bit set may not contain both global variables and functions"); + *OFI++ = Fn; + } + } + + buildBitSetsFromFunctions(BitSets, OrderedFns); + } +} + /// Lower all bit sets in this module. bool LowerBitSets::buildBitSets() { Function *BitSetTestFunc = @@ -573,24 +926,36 @@ bool LowerBitSets::buildBitSets() { // Equivalence class set containing bitsets and the globals they reference. // This is used to partition the set of bitsets in the module into disjoint // sets. - typedef EquivalenceClasses> + typedef EquivalenceClasses> GlobalClassesTy; GlobalClassesTy GlobalClasses; + // Verify the bitset metadata and build a mapping from bitset identifiers to + // their last observed index in BitSetNM. This will used later to + // deterministically order the list of bitset identifiers. + llvm::DenseMap BitSetIdIndices; + if (BitSetNM) { + for (unsigned I = 0, E = BitSetNM->getNumOperands(); I != E; ++I) { + MDNode *Op = BitSetNM->getOperand(I); + verifyBitSetMDNode(Op); + BitSetIdIndices[Op->getOperand(0)] = I; + } + } + for (const Use &U : BitSetTestFunc->uses()) { auto CI = cast(U.getUser()); auto BitSetMDVal = dyn_cast(CI->getArgOperand(1)); - if (!BitSetMDVal || !isa(BitSetMDVal->getMetadata())) + if (!BitSetMDVal) report_fatal_error( - "Second argument of llvm.bitset.test must be metadata string"); - auto BitSet = cast(BitSetMDVal->getMetadata()); + "Second argument of llvm.bitset.test must be metadata"); + auto BitSet = BitSetMDVal->getMetadata(); // Add the call site to the list of call sites for this bit set. We also use // BitSetTestCallSites to keep track of whether we have seen this bit set // before. If we have, we don't need to re-add the referenced globals to the // equivalence class. - std::pair>::iterator, + std::pair>::iterator, bool> Ins = BitSetTestCallSites.insert( std::make_pair(BitSet, std::vector())); @@ -605,30 +970,15 @@ bool LowerBitSets::buildBitSets() { if (!BitSetNM) continue; - // Verify the bitset metadata and add the referenced globals to the bitset's - // equivalence class. + // Add the referenced globals to the bitset's equivalence class. for (MDNode *Op : BitSetNM->operands()) { - if (Op->getNumOperands() != 3) - report_fatal_error( - "All operands of llvm.bitsets metadata must have 3 elements"); - if (Op->getOperand(0) != BitSet || !Op->getOperand(1)) continue; - auto OpConstMD = dyn_cast(Op->getOperand(1)); - if (!OpConstMD) - report_fatal_error("Bit set element must be a constant"); - auto OpGlobal = dyn_cast(OpConstMD->getValue()); + auto OpGlobal = dyn_cast( + cast(Op->getOperand(1))->getValue()); if (!OpGlobal) - report_fatal_error("Bit set element must refer to global"); - - auto OffsetConstMD = dyn_cast(Op->getOperand(2)); - if (!OffsetConstMD) - report_fatal_error("Bit set element offset must be a constant"); - auto OffsetInt = dyn_cast(OffsetConstMD->getValue()); - if (!OffsetInt) - report_fatal_error( - "Bit set element offset must be an integer constant"); + continue; CurSet = GlobalClasses.unionSets( CurSet, GlobalClasses.findLeader(GlobalClasses.insert(OpGlobal))); @@ -638,77 +988,51 @@ bool LowerBitSets::buildBitSets() { if (GlobalClasses.empty()) return false; - // For each disjoint set we found... + // Build a list of disjoint sets ordered by their maximum BitSetNM index + // for determinism. + std::vector> Sets; for (GlobalClassesTy::iterator I = GlobalClasses.begin(), E = GlobalClasses.end(); I != E; ++I) { if (!I->isLeader()) continue; - ++NumBitSetDisjointSets; - // Build the list of bitsets and referenced globals in this disjoint set. - std::vector BitSets; - std::vector Globals; - llvm::DenseMap BitSetIndices; - llvm::DenseMap GlobalIndices; + unsigned MaxIndex = 0; for (GlobalClassesTy::member_iterator MI = GlobalClasses.member_begin(I); MI != GlobalClasses.member_end(); ++MI) { - if ((*MI).is()) { - BitSetIndices[MI->get()] = BitSets.size(); - BitSets.push_back(MI->get()); - } else { - GlobalIndices[MI->get()] = Globals.size(); - Globals.push_back(MI->get()); - } + if ((*MI).is()) + MaxIndex = std::max(MaxIndex, BitSetIdIndices[MI->get()]); } + Sets.emplace_back(I, MaxIndex); + } + std::sort(Sets.begin(), Sets.end(), + [](const std::pair &S1, + const std::pair &S2) { + return S1.second < S2.second; + }); - // For each bitset, build a set of indices that refer to globals referenced - // by the bitset. - std::vector> BitSetMembers(BitSets.size()); - if (BitSetNM) { - for (MDNode *Op : BitSetNM->operands()) { - // Op = { bitset name, global, offset } - if (!Op->getOperand(1)) - continue; - auto I = BitSetIndices.find(cast(Op->getOperand(0))); - if (I == BitSetIndices.end()) - continue; - - auto OpGlobal = cast( - cast(Op->getOperand(1))->getValue()); - BitSetMembers[I->second].insert(GlobalIndices[OpGlobal]); - } + // For each disjoint set we found... + for (const auto &S : Sets) { + // Build the list of bitsets in this disjoint set. + std::vector BitSets; + std::vector Globals; + for (GlobalClassesTy::member_iterator MI = + GlobalClasses.member_begin(S.first); + MI != GlobalClasses.member_end(); ++MI) { + if ((*MI).is()) + BitSets.push_back(MI->get()); + else + Globals.push_back(MI->get()); } - // Order the sets of indices by size. The GlobalLayoutBuilder works best - // when given small index sets first. - std::stable_sort( - BitSetMembers.begin(), BitSetMembers.end(), - [](const std::set &O1, const std::set &O2) { - return O1.size() < O2.size(); - }); - - // Create a GlobalLayoutBuilder and provide it with index sets as layout - // fragments. The GlobalLayoutBuilder tries to lay out members of fragments - // as close together as possible. - GlobalLayoutBuilder GLB(Globals.size()); - for (auto &&MemSet : BitSetMembers) - GLB.addFragment(MemSet); - - // Build a vector of globals with the computed layout. - std::vector OrderedGlobals(Globals.size()); - auto OGI = OrderedGlobals.begin(); - for (auto &&F : GLB.Fragments) - for (auto &&Offset : F) - *OGI++ = Globals[Offset]; - - // Order bitsets by name for determinism. - std::sort(BitSets.begin(), BitSets.end(), [](MDString *S1, MDString *S2) { - return S1->getString() < S2->getString(); + // Order bitsets by BitSetNM index for determinism. This ordering is stable + // as there is a one-to-one mapping between metadata and indices. + std::sort(BitSets.begin(), BitSets.end(), [&](Metadata *M1, Metadata *M2) { + return BitSetIdIndices[M1] < BitSetIdIndices[M2]; }); - // Build the bitsets from this disjoint set. - buildBitSetsFromGlobals(BitSets, OrderedGlobals); + // Lower the bitsets in this disjoint set. + buildBitSetsFromDisjointSet(BitSets, Globals); } allocateByteArrays();