11dc6a8d1SPeter Collingbourne //===- SplitModule.cpp - Split a module into partitions -------------------===// 21dc6a8d1SPeter Collingbourne // 31dc6a8d1SPeter Collingbourne // The LLVM Compiler Infrastructure 41dc6a8d1SPeter Collingbourne // 51dc6a8d1SPeter Collingbourne // This file is distributed under the University of Illinois Open Source 61dc6a8d1SPeter Collingbourne // License. See LICENSE.TXT for details. 71dc6a8d1SPeter Collingbourne // 81dc6a8d1SPeter Collingbourne //===----------------------------------------------------------------------===// 91dc6a8d1SPeter Collingbourne // 101dc6a8d1SPeter Collingbourne // This file defines the function llvm::SplitModule, which splits a module 111dc6a8d1SPeter Collingbourne // into multiple linkable partitions. It can be used to implement parallel code 121dc6a8d1SPeter Collingbourne // generation for link-time optimization. 131dc6a8d1SPeter Collingbourne // 141dc6a8d1SPeter Collingbourne //===----------------------------------------------------------------------===// 151dc6a8d1SPeter Collingbourne 16*d19d4d30SSergei Larin #define DEBUG_TYPE "split-module" 17*d19d4d30SSergei Larin 181dc6a8d1SPeter Collingbourne #include "llvm/Transforms/Utils/SplitModule.h" 19*d19d4d30SSergei Larin #include "llvm/ADT/EquivalenceClasses.h" 201dc6a8d1SPeter Collingbourne #include "llvm/ADT/Hashing.h" 21*d19d4d30SSergei Larin #include "llvm/ADT/MapVector.h" 22*d19d4d30SSergei Larin #include "llvm/ADT/SetVector.h" 231dc6a8d1SPeter Collingbourne #include "llvm/IR/Function.h" 241dc6a8d1SPeter Collingbourne #include "llvm/IR/GlobalAlias.h" 251dc6a8d1SPeter Collingbourne #include "llvm/IR/GlobalObject.h" 261dc6a8d1SPeter Collingbourne #include "llvm/IR/GlobalValue.h" 271dc6a8d1SPeter Collingbourne #include "llvm/IR/Module.h" 28*d19d4d30SSergei Larin #include "llvm/Support/CommandLine.h" 29*d19d4d30SSergei Larin #include "llvm/Support/Debug.h" 301dc6a8d1SPeter Collingbourne #include "llvm/Support/MD5.h" 311dc6a8d1SPeter Collingbourne #include "llvm/Support/raw_ostream.h" 321dc6a8d1SPeter Collingbourne #include "llvm/Transforms/Utils/Cloning.h" 33*d19d4d30SSergei Larin #include <queue> 341dc6a8d1SPeter Collingbourne 351dc6a8d1SPeter Collingbourne using namespace llvm; 361dc6a8d1SPeter Collingbourne 37*d19d4d30SSergei Larin namespace { 38*d19d4d30SSergei Larin typedef EquivalenceClasses<const GlobalValue *> ClusterMapType; 39*d19d4d30SSergei Larin typedef DenseMap<const Comdat *, const GlobalValue *> ComdatMembersType; 40*d19d4d30SSergei Larin typedef DenseMap<const GlobalValue *, unsigned> ClusterIDMapType; 41*d19d4d30SSergei Larin } 42*d19d4d30SSergei Larin 43*d19d4d30SSergei Larin static void addNonConstUser(ClusterMapType &GVtoClusterMap, 44*d19d4d30SSergei Larin const GlobalValue *GV, const User *U) { 45*d19d4d30SSergei Larin assert((!isa<Constant>(U) || isa<GlobalValue>(U)) && "Bad user"); 46*d19d4d30SSergei Larin 47*d19d4d30SSergei Larin if (const Instruction *I = dyn_cast<Instruction>(U)) { 48*d19d4d30SSergei Larin const GlobalValue *F = I->getParent()->getParent(); 49*d19d4d30SSergei Larin GVtoClusterMap.unionSets(GV, F); 50*d19d4d30SSergei Larin } else if (isa<GlobalAlias>(U) || isa<Function>(U) || 51*d19d4d30SSergei Larin isa<GlobalVariable>(U)) { 52*d19d4d30SSergei Larin GVtoClusterMap.unionSets(GV, cast<GlobalValue>(U)); 53*d19d4d30SSergei Larin } else { 54*d19d4d30SSergei Larin llvm_unreachable("Underimplemented use case"); 55*d19d4d30SSergei Larin } 56*d19d4d30SSergei Larin } 57*d19d4d30SSergei Larin 58*d19d4d30SSergei Larin // Find partitions for module in the way that no locals need to be 59*d19d4d30SSergei Larin // globalized. 60*d19d4d30SSergei Larin // Try to balance pack those partitions into N files since this roughly equals 61*d19d4d30SSergei Larin // thread balancing for the backend codegen step. 62*d19d4d30SSergei Larin static void findPartitions(Module *M, ClusterIDMapType &ClusterIDMap, 63*d19d4d30SSergei Larin unsigned N) { 64*d19d4d30SSergei Larin // At this point module should have the proper mix of globals and locals. 65*d19d4d30SSergei Larin // As we attempt to partition this module, we must not change any 66*d19d4d30SSergei Larin // locals to globals. 67*d19d4d30SSergei Larin 68*d19d4d30SSergei Larin DEBUG(dbgs() << "Partition module with (" << M->size() << ")functions\n"); 69*d19d4d30SSergei Larin ClusterMapType GVtoClusterMap; 70*d19d4d30SSergei Larin ComdatMembersType ComdatMembers; 71*d19d4d30SSergei Larin 72*d19d4d30SSergei Larin auto recordGVSet = [&GVtoClusterMap, &ComdatMembers](GlobalValue &GV) { 73*d19d4d30SSergei Larin if (GV.isDeclaration()) 74*d19d4d30SSergei Larin return; 75*d19d4d30SSergei Larin 76*d19d4d30SSergei Larin if (!GV.hasName()) 77*d19d4d30SSergei Larin GV.setName("__llvmsplit_unnamed"); 78*d19d4d30SSergei Larin 79*d19d4d30SSergei Larin // Comdat groups must not be partitioned. For comdat groups that contain 80*d19d4d30SSergei Larin // locals, record all their members here so we can keep them together. 81*d19d4d30SSergei Larin // Comdat groups that only contain external globals are already handled by 82*d19d4d30SSergei Larin // the MD5-based partitioning. 83*d19d4d30SSergei Larin if (const Comdat *C = GV.getComdat()) { 84*d19d4d30SSergei Larin auto &Member = ComdatMembers[C]; 85*d19d4d30SSergei Larin if (Member) 86*d19d4d30SSergei Larin GVtoClusterMap.unionSets(Member, &GV); 87*d19d4d30SSergei Larin else 88*d19d4d30SSergei Larin Member = &GV; 89*d19d4d30SSergei Larin } 90*d19d4d30SSergei Larin 91*d19d4d30SSergei Larin // Further only iterate over local GVs. 92*d19d4d30SSergei Larin if (!GV.hasLocalLinkage()) 93*d19d4d30SSergei Larin return; 94*d19d4d30SSergei Larin 95*d19d4d30SSergei Larin for (auto *U : GV.users()) { 96*d19d4d30SSergei Larin SmallVector<const User *, 4> Worklist; 97*d19d4d30SSergei Larin Worklist.push_back(U); 98*d19d4d30SSergei Larin while (!Worklist.empty()) { 99*d19d4d30SSergei Larin const User *UU = Worklist.pop_back_val(); 100*d19d4d30SSergei Larin // For each constant that is not a GV (a pure const) recurse. 101*d19d4d30SSergei Larin if (isa<Constant>(UU) && !isa<GlobalValue>(UU)) { 102*d19d4d30SSergei Larin Worklist.append(UU->user_begin(), UU->user_end()); 103*d19d4d30SSergei Larin continue; 104*d19d4d30SSergei Larin } 105*d19d4d30SSergei Larin addNonConstUser(GVtoClusterMap, &GV, UU); 106*d19d4d30SSergei Larin } 107*d19d4d30SSergei Larin } 108*d19d4d30SSergei Larin }; 109*d19d4d30SSergei Larin 110*d19d4d30SSergei Larin std::for_each(M->begin(), M->end(), recordGVSet); 111*d19d4d30SSergei Larin std::for_each(M->global_begin(), M->global_end(), recordGVSet); 112*d19d4d30SSergei Larin std::for_each(M->alias_begin(), M->alias_end(), recordGVSet); 113*d19d4d30SSergei Larin 114*d19d4d30SSergei Larin // Assigned all GVs to merged clusters while balancing number of objects in 115*d19d4d30SSergei Larin // each. 116*d19d4d30SSergei Larin auto CompareClusters = [](const std::pair<unsigned, unsigned> &a, 117*d19d4d30SSergei Larin const std::pair<unsigned, unsigned> &b) { 118*d19d4d30SSergei Larin if (a.second || b.second) 119*d19d4d30SSergei Larin return a.second > b.second; 120*d19d4d30SSergei Larin else 121*d19d4d30SSergei Larin return a.first > b.first; 122*d19d4d30SSergei Larin }; 123*d19d4d30SSergei Larin 124*d19d4d30SSergei Larin std::priority_queue<std::pair<unsigned, unsigned>, 125*d19d4d30SSergei Larin std::vector<std::pair<unsigned, unsigned>>, 126*d19d4d30SSergei Larin decltype(CompareClusters)> 127*d19d4d30SSergei Larin BalancinQueue(CompareClusters); 128*d19d4d30SSergei Larin // Pre-populate priority queue with N slot blanks. 129*d19d4d30SSergei Larin for (unsigned i = 0; i < N; ++i) 130*d19d4d30SSergei Larin BalancinQueue.push(std::make_pair(i, 0)); 131*d19d4d30SSergei Larin 132*d19d4d30SSergei Larin typedef std::pair<unsigned, ClusterMapType::iterator> SortType; 133*d19d4d30SSergei Larin SmallVector<SortType, 64> Sets; 134*d19d4d30SSergei Larin SmallPtrSet<const GlobalValue *, 64> Visited; 135*d19d4d30SSergei Larin 136*d19d4d30SSergei Larin // To guarantee determinism, we have to sort SCC according to size. 137*d19d4d30SSergei Larin // When size is the same, use leader's name. 138*d19d4d30SSergei Larin for (ClusterMapType::iterator I = GVtoClusterMap.begin(), 139*d19d4d30SSergei Larin E = GVtoClusterMap.end(); I != E; ++I) 140*d19d4d30SSergei Larin if (I->isLeader()) 141*d19d4d30SSergei Larin Sets.push_back( 142*d19d4d30SSergei Larin std::make_pair(std::distance(GVtoClusterMap.member_begin(I), 143*d19d4d30SSergei Larin GVtoClusterMap.member_end()), I)); 144*d19d4d30SSergei Larin 145*d19d4d30SSergei Larin std::sort(Sets.begin(), Sets.end(), [](const SortType &a, const SortType &b) { 146*d19d4d30SSergei Larin if (a.first == b.first) 147*d19d4d30SSergei Larin return a.second->getData()->getName() > b.second->getData()->getName(); 148*d19d4d30SSergei Larin else 149*d19d4d30SSergei Larin return a.first > b.first; 150*d19d4d30SSergei Larin }); 151*d19d4d30SSergei Larin 152*d19d4d30SSergei Larin for (auto &I : Sets) { 153*d19d4d30SSergei Larin unsigned CurrentClusterID = BalancinQueue.top().first; 154*d19d4d30SSergei Larin unsigned CurrentClusterSize = BalancinQueue.top().second; 155*d19d4d30SSergei Larin BalancinQueue.pop(); 156*d19d4d30SSergei Larin 157*d19d4d30SSergei Larin DEBUG(dbgs() << "Root[" << CurrentClusterID << "] cluster_size(" << I.first 158*d19d4d30SSergei Larin << ") ----> " << I.second->getData()->getName() << "\n"); 159*d19d4d30SSergei Larin 160*d19d4d30SSergei Larin for (ClusterMapType::member_iterator MI = 161*d19d4d30SSergei Larin GVtoClusterMap.findLeader(I.second); 162*d19d4d30SSergei Larin MI != GVtoClusterMap.member_end(); ++MI) { 163*d19d4d30SSergei Larin if (!Visited.insert(*MI).second) 164*d19d4d30SSergei Larin continue; 165*d19d4d30SSergei Larin DEBUG(dbgs() << "----> " << (*MI)->getName() 166*d19d4d30SSergei Larin << ((*MI)->hasLocalLinkage() ? " l " : " e ") << "\n"); 167*d19d4d30SSergei Larin Visited.insert(*MI); 168*d19d4d30SSergei Larin ClusterIDMap[*MI] = CurrentClusterID; 169*d19d4d30SSergei Larin CurrentClusterSize++; 170*d19d4d30SSergei Larin } 171*d19d4d30SSergei Larin // Add this set size to the number of entries in this cluster. 172*d19d4d30SSergei Larin BalancinQueue.push(std::make_pair(CurrentClusterID, CurrentClusterSize)); 173*d19d4d30SSergei Larin } 174*d19d4d30SSergei Larin } 175*d19d4d30SSergei Larin 1761dc6a8d1SPeter Collingbourne static void externalize(GlobalValue *GV) { 1771dc6a8d1SPeter Collingbourne if (GV->hasLocalLinkage()) { 1781dc6a8d1SPeter Collingbourne GV->setLinkage(GlobalValue::ExternalLinkage); 1791dc6a8d1SPeter Collingbourne GV->setVisibility(GlobalValue::HiddenVisibility); 1801dc6a8d1SPeter Collingbourne } 1811dc6a8d1SPeter Collingbourne 1821dc6a8d1SPeter Collingbourne // Unnamed entities must be named consistently between modules. setName will 1831dc6a8d1SPeter Collingbourne // give a distinct name to each such entity. 1841dc6a8d1SPeter Collingbourne if (!GV->hasName()) 1851dc6a8d1SPeter Collingbourne GV->setName("__llvmsplit_unnamed"); 1861dc6a8d1SPeter Collingbourne } 1871dc6a8d1SPeter Collingbourne 1881dc6a8d1SPeter Collingbourne // Returns whether GV should be in partition (0-based) I of N. 1891dc6a8d1SPeter Collingbourne static bool isInPartition(const GlobalValue *GV, unsigned I, unsigned N) { 1901dc6a8d1SPeter Collingbourne if (auto GA = dyn_cast<GlobalAlias>(GV)) 1911dc6a8d1SPeter Collingbourne if (const GlobalObject *Base = GA->getBaseObject()) 1921dc6a8d1SPeter Collingbourne GV = Base; 1931dc6a8d1SPeter Collingbourne 1941dc6a8d1SPeter Collingbourne StringRef Name; 1951dc6a8d1SPeter Collingbourne if (const Comdat *C = GV->getComdat()) 1961dc6a8d1SPeter Collingbourne Name = C->getName(); 1971dc6a8d1SPeter Collingbourne else 1981dc6a8d1SPeter Collingbourne Name = GV->getName(); 1991dc6a8d1SPeter Collingbourne 2001dc6a8d1SPeter Collingbourne // Partition by MD5 hash. We only need a few bits for evenness as the number 2011dc6a8d1SPeter Collingbourne // of partitions will generally be in the 1-2 figure range; the low 16 bits 2021dc6a8d1SPeter Collingbourne // are enough. 2031dc6a8d1SPeter Collingbourne MD5 H; 2041dc6a8d1SPeter Collingbourne MD5::MD5Result R; 2051dc6a8d1SPeter Collingbourne H.update(Name); 2061dc6a8d1SPeter Collingbourne H.final(R); 2071dc6a8d1SPeter Collingbourne return (R[0] | (R[1] << 8)) % N == I; 2081dc6a8d1SPeter Collingbourne } 2091dc6a8d1SPeter Collingbourne 2101dc6a8d1SPeter Collingbourne void llvm::SplitModule( 2111dc6a8d1SPeter Collingbourne std::unique_ptr<Module> M, unsigned N, 212*d19d4d30SSergei Larin std::function<void(std::unique_ptr<Module> MPart)> ModuleCallback, 213*d19d4d30SSergei Larin bool PreserveLocals) { 214*d19d4d30SSergei Larin if (!PreserveLocals) { 2151dc6a8d1SPeter Collingbourne for (Function &F : *M) 2161dc6a8d1SPeter Collingbourne externalize(&F); 2171dc6a8d1SPeter Collingbourne for (GlobalVariable &GV : M->globals()) 2181dc6a8d1SPeter Collingbourne externalize(&GV); 2191dc6a8d1SPeter Collingbourne for (GlobalAlias &GA : M->aliases()) 2201dc6a8d1SPeter Collingbourne externalize(&GA); 2211dc6a8d1SPeter Collingbourne 2221dc6a8d1SPeter Collingbourne // FIXME: We should be able to reuse M as the last partition instead of 2231dc6a8d1SPeter Collingbourne // cloning it. 2241dc6a8d1SPeter Collingbourne for (unsigned I = 0; I != N; ++I) { 2251dc6a8d1SPeter Collingbourne ValueToValueMapTy VMap; 2261dc6a8d1SPeter Collingbourne std::unique_ptr<Module> MPart( 2271dc6a8d1SPeter Collingbourne CloneModule(M.get(), VMap, [=](const GlobalValue *GV) { 2281dc6a8d1SPeter Collingbourne return isInPartition(GV, I, N); 2291dc6a8d1SPeter Collingbourne })); 2301dc6a8d1SPeter Collingbourne if (I != 0) 2311dc6a8d1SPeter Collingbourne MPart->setModuleInlineAsm(""); 2321dc6a8d1SPeter Collingbourne ModuleCallback(std::move(MPart)); 2331dc6a8d1SPeter Collingbourne } 234*d19d4d30SSergei Larin } else { 235*d19d4d30SSergei Larin // This performs splitting without a need for externalization, which might not 236*d19d4d30SSergei Larin // always be possible. 237*d19d4d30SSergei Larin ClusterIDMapType ClusterIDMap; 238*d19d4d30SSergei Larin findPartitions(M.get(), ClusterIDMap, N); 239*d19d4d30SSergei Larin 240*d19d4d30SSergei Larin for (unsigned I = 0; I < N; ++I) { 241*d19d4d30SSergei Larin ValueToValueMapTy VMap; 242*d19d4d30SSergei Larin std::unique_ptr<Module> MPart( 243*d19d4d30SSergei Larin CloneModule(M.get(), VMap, [&](const GlobalValue *GV) { 244*d19d4d30SSergei Larin if (ClusterIDMap.count(GV)) 245*d19d4d30SSergei Larin return (ClusterIDMap[GV] == I); 246*d19d4d30SSergei Larin else 247*d19d4d30SSergei Larin return isInPartition(GV, I, N); 248*d19d4d30SSergei Larin })); 249*d19d4d30SSergei Larin if (I != 0) 250*d19d4d30SSergei Larin MPart->setModuleInlineAsm(""); 251*d19d4d30SSergei Larin ModuleCallback(std::move(MPart)); 252*d19d4d30SSergei Larin } 253*d19d4d30SSergei Larin } 2541dc6a8d1SPeter Collingbourne } 255