1 //===- ModuleSummaryAnalysis.cpp - Module summary index builder -----------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This pass builds a ModuleSummaryIndex object for the module, to be written
10 // to bitcode or LLVM assembly.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/Analysis/ModuleSummaryAnalysis.h"
15 #include "llvm/ADT/ArrayRef.h"
16 #include "llvm/ADT/DenseSet.h"
17 #include "llvm/ADT/MapVector.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/ADT/SetVector.h"
20 #include "llvm/ADT/SmallPtrSet.h"
21 #include "llvm/ADT/SmallVector.h"
22 #include "llvm/ADT/StringRef.h"
23 #include "llvm/Analysis/BlockFrequencyInfo.h"
24 #include "llvm/Analysis/BranchProbabilityInfo.h"
25 #include "llvm/Analysis/IndirectCallPromotionAnalysis.h"
26 #include "llvm/Analysis/LoopInfo.h"
27 #include "llvm/Analysis/MemoryProfileInfo.h"
28 #include "llvm/Analysis/ProfileSummaryInfo.h"
29 #include "llvm/Analysis/StackSafetyAnalysis.h"
30 #include "llvm/Analysis/TypeMetadataUtils.h"
31 #include "llvm/IR/Attributes.h"
32 #include "llvm/IR/BasicBlock.h"
33 #include "llvm/IR/Constant.h"
34 #include "llvm/IR/Constants.h"
35 #include "llvm/IR/Dominators.h"
36 #include "llvm/IR/Function.h"
37 #include "llvm/IR/GlobalAlias.h"
38 #include "llvm/IR/GlobalValue.h"
39 #include "llvm/IR/GlobalVariable.h"
40 #include "llvm/IR/Instructions.h"
41 #include "llvm/IR/IntrinsicInst.h"
42 #include "llvm/IR/Metadata.h"
43 #include "llvm/IR/Module.h"
44 #include "llvm/IR/ModuleSummaryIndex.h"
45 #include "llvm/IR/Use.h"
46 #include "llvm/IR/User.h"
47 #include "llvm/InitializePasses.h"
48 #include "llvm/Object/ModuleSymbolTable.h"
49 #include "llvm/Object/SymbolicFile.h"
50 #include "llvm/Pass.h"
51 #include "llvm/Support/Casting.h"
52 #include "llvm/Support/CommandLine.h"
53 #include "llvm/Support/FileSystem.h"
54 #include <algorithm>
55 #include <cassert>
56 #include <cstdint>
57 #include <vector>
58 
59 using namespace llvm;
60 using namespace llvm::memprof;
61 
62 #define DEBUG_TYPE "module-summary-analysis"
63 
64 // Option to force edges cold which will block importing when the
65 // -import-cold-multiplier is set to 0. Useful for debugging.
66 namespace llvm {
67 FunctionSummary::ForceSummaryHotnessType ForceSummaryEdgesCold =
68     FunctionSummary::FSHT_None;
69 } // namespace llvm
70 
71 static cl::opt<FunctionSummary::ForceSummaryHotnessType, true> FSEC(
72     "force-summary-edges-cold", cl::Hidden, cl::location(ForceSummaryEdgesCold),
73     cl::desc("Force all edges in the function summary to cold"),
74     cl::values(clEnumValN(FunctionSummary::FSHT_None, "none", "None."),
75                clEnumValN(FunctionSummary::FSHT_AllNonCritical,
76                           "all-non-critical", "All non-critical edges."),
77                clEnumValN(FunctionSummary::FSHT_All, "all", "All edges.")));
78 
79 static cl::opt<std::string> ModuleSummaryDotFile(
80     "module-summary-dot-file", cl::Hidden, cl::value_desc("filename"),
81     cl::desc("File to emit dot graph of new summary into"));
82 
83 extern cl::opt<bool> ScalePartialSampleProfileWorkingSetSize;
84 
85 // Walk through the operands of a given User via worklist iteration and populate
86 // the set of GlobalValue references encountered. Invoked either on an
87 // Instruction or a GlobalVariable (which walks its initializer).
88 // Return true if any of the operands contains blockaddress. This is important
89 // to know when computing summary for global var, because if global variable
90 // references basic block address we can't import it separately from function
91 // containing that basic block. For simplicity we currently don't import such
92 // global vars at all. When importing function we aren't interested if any
93 // instruction in it takes an address of any basic block, because instruction
94 // can only take an address of basic block located in the same function.
findRefEdges(ModuleSummaryIndex & Index,const User * CurUser,SetVector<ValueInfo,std::vector<ValueInfo>> & RefEdges,SmallPtrSet<const User *,8> & Visited)95 static bool findRefEdges(ModuleSummaryIndex &Index, const User *CurUser,
96                          SetVector<ValueInfo, std::vector<ValueInfo>> &RefEdges,
97                          SmallPtrSet<const User *, 8> &Visited) {
98   bool HasBlockAddress = false;
99   SmallVector<const User *, 32> Worklist;
100   if (Visited.insert(CurUser).second)
101     Worklist.push_back(CurUser);
102 
103   while (!Worklist.empty()) {
104     const User *U = Worklist.pop_back_val();
105     const auto *CB = dyn_cast<CallBase>(U);
106 
107     for (const auto &OI : U->operands()) {
108       const User *Operand = dyn_cast<User>(OI);
109       if (!Operand)
110         continue;
111       if (isa<BlockAddress>(Operand)) {
112         HasBlockAddress = true;
113         continue;
114       }
115       if (auto *GV = dyn_cast<GlobalValue>(Operand)) {
116         // We have a reference to a global value. This should be added to
117         // the reference set unless it is a callee. Callees are handled
118         // specially by WriteFunction and are added to a separate list.
119         if (!(CB && CB->isCallee(&OI)))
120           RefEdges.insert(Index.getOrInsertValueInfo(GV));
121         continue;
122       }
123       if (Visited.insert(Operand).second)
124         Worklist.push_back(Operand);
125     }
126   }
127   return HasBlockAddress;
128 }
129 
getHotness(uint64_t ProfileCount,ProfileSummaryInfo * PSI)130 static CalleeInfo::HotnessType getHotness(uint64_t ProfileCount,
131                                           ProfileSummaryInfo *PSI) {
132   if (!PSI)
133     return CalleeInfo::HotnessType::Unknown;
134   if (PSI->isHotCount(ProfileCount))
135     return CalleeInfo::HotnessType::Hot;
136   if (PSI->isColdCount(ProfileCount))
137     return CalleeInfo::HotnessType::Cold;
138   return CalleeInfo::HotnessType::None;
139 }
140 
isNonRenamableLocal(const GlobalValue & GV)141 static bool isNonRenamableLocal(const GlobalValue &GV) {
142   return GV.hasSection() && GV.hasLocalLinkage();
143 }
144 
145 /// Determine whether this call has all constant integer arguments (excluding
146 /// "this") and summarize it to VCalls or ConstVCalls as appropriate.
addVCallToSet(DevirtCallSite Call,GlobalValue::GUID Guid,SetVector<FunctionSummary::VFuncId,std::vector<FunctionSummary::VFuncId>> & VCalls,SetVector<FunctionSummary::ConstVCall,std::vector<FunctionSummary::ConstVCall>> & ConstVCalls)147 static void addVCallToSet(
148     DevirtCallSite Call, GlobalValue::GUID Guid,
149     SetVector<FunctionSummary::VFuncId, std::vector<FunctionSummary::VFuncId>>
150         &VCalls,
151     SetVector<FunctionSummary::ConstVCall,
152               std::vector<FunctionSummary::ConstVCall>> &ConstVCalls) {
153   std::vector<uint64_t> Args;
154   // Start from the second argument to skip the "this" pointer.
155   for (auto &Arg : drop_begin(Call.CB.args())) {
156     auto *CI = dyn_cast<ConstantInt>(Arg);
157     if (!CI || CI->getBitWidth() > 64) {
158       VCalls.insert({Guid, Call.Offset});
159       return;
160     }
161     Args.push_back(CI->getZExtValue());
162   }
163   ConstVCalls.insert({{Guid, Call.Offset}, std::move(Args)});
164 }
165 
166 /// If this intrinsic call requires that we add information to the function
167 /// summary, do so via the non-constant reference arguments.
addIntrinsicToSummary(const CallInst * CI,SetVector<GlobalValue::GUID,std::vector<GlobalValue::GUID>> & TypeTests,SetVector<FunctionSummary::VFuncId,std::vector<FunctionSummary::VFuncId>> & TypeTestAssumeVCalls,SetVector<FunctionSummary::VFuncId,std::vector<FunctionSummary::VFuncId>> & TypeCheckedLoadVCalls,SetVector<FunctionSummary::ConstVCall,std::vector<FunctionSummary::ConstVCall>> & TypeTestAssumeConstVCalls,SetVector<FunctionSummary::ConstVCall,std::vector<FunctionSummary::ConstVCall>> & TypeCheckedLoadConstVCalls,DominatorTree & DT)168 static void addIntrinsicToSummary(
169     const CallInst *CI,
170     SetVector<GlobalValue::GUID, std::vector<GlobalValue::GUID>> &TypeTests,
171     SetVector<FunctionSummary::VFuncId, std::vector<FunctionSummary::VFuncId>>
172         &TypeTestAssumeVCalls,
173     SetVector<FunctionSummary::VFuncId, std::vector<FunctionSummary::VFuncId>>
174         &TypeCheckedLoadVCalls,
175     SetVector<FunctionSummary::ConstVCall,
176               std::vector<FunctionSummary::ConstVCall>>
177         &TypeTestAssumeConstVCalls,
178     SetVector<FunctionSummary::ConstVCall,
179               std::vector<FunctionSummary::ConstVCall>>
180         &TypeCheckedLoadConstVCalls,
181     DominatorTree &DT) {
182   switch (CI->getCalledFunction()->getIntrinsicID()) {
183   case Intrinsic::type_test:
184   case Intrinsic::public_type_test: {
185     auto *TypeMDVal = cast<MetadataAsValue>(CI->getArgOperand(1));
186     auto *TypeId = dyn_cast<MDString>(TypeMDVal->getMetadata());
187     if (!TypeId)
188       break;
189     GlobalValue::GUID Guid = GlobalValue::getGUID(TypeId->getString());
190 
191     // Produce a summary from type.test intrinsics. We only summarize type.test
192     // intrinsics that are used other than by an llvm.assume intrinsic.
193     // Intrinsics that are assumed are relevant only to the devirtualization
194     // pass, not the type test lowering pass.
195     bool HasNonAssumeUses = llvm::any_of(CI->uses(), [](const Use &CIU) {
196       return !isa<AssumeInst>(CIU.getUser());
197     });
198     if (HasNonAssumeUses)
199       TypeTests.insert(Guid);
200 
201     SmallVector<DevirtCallSite, 4> DevirtCalls;
202     SmallVector<CallInst *, 4> Assumes;
203     findDevirtualizableCallsForTypeTest(DevirtCalls, Assumes, CI, DT);
204     for (auto &Call : DevirtCalls)
205       addVCallToSet(Call, Guid, TypeTestAssumeVCalls,
206                     TypeTestAssumeConstVCalls);
207 
208     break;
209   }
210 
211   case Intrinsic::type_checked_load_relative:
212   case Intrinsic::type_checked_load: {
213     auto *TypeMDVal = cast<MetadataAsValue>(CI->getArgOperand(2));
214     auto *TypeId = dyn_cast<MDString>(TypeMDVal->getMetadata());
215     if (!TypeId)
216       break;
217     GlobalValue::GUID Guid = GlobalValue::getGUID(TypeId->getString());
218 
219     SmallVector<DevirtCallSite, 4> DevirtCalls;
220     SmallVector<Instruction *, 4> LoadedPtrs;
221     SmallVector<Instruction *, 4> Preds;
222     bool HasNonCallUses = false;
223     findDevirtualizableCallsForTypeCheckedLoad(DevirtCalls, LoadedPtrs, Preds,
224                                                HasNonCallUses, CI, DT);
225     // Any non-call uses of the result of llvm.type.checked.load will
226     // prevent us from optimizing away the llvm.type.test.
227     if (HasNonCallUses)
228       TypeTests.insert(Guid);
229     for (auto &Call : DevirtCalls)
230       addVCallToSet(Call, Guid, TypeCheckedLoadVCalls,
231                     TypeCheckedLoadConstVCalls);
232 
233     break;
234   }
235   default:
236     break;
237   }
238 }
239 
isNonVolatileLoad(const Instruction * I)240 static bool isNonVolatileLoad(const Instruction *I) {
241   if (const auto *LI = dyn_cast<LoadInst>(I))
242     return !LI->isVolatile();
243 
244   return false;
245 }
246 
isNonVolatileStore(const Instruction * I)247 static bool isNonVolatileStore(const Instruction *I) {
248   if (const auto *SI = dyn_cast<StoreInst>(I))
249     return !SI->isVolatile();
250 
251   return false;
252 }
253 
254 // Returns true if the function definition must be unreachable.
255 //
256 // Note if this helper function returns true, `F` is guaranteed
257 // to be unreachable; if it returns false, `F` might still
258 // be unreachable but not covered by this helper function.
mustBeUnreachableFunction(const Function & F)259 static bool mustBeUnreachableFunction(const Function &F) {
260   // A function must be unreachable if its entry block ends with an
261   // 'unreachable'.
262   assert(!F.isDeclaration());
263   return isa<UnreachableInst>(F.getEntryBlock().getTerminator());
264 }
265 
computeFunctionSummary(ModuleSummaryIndex & Index,const Module & M,const Function & F,BlockFrequencyInfo * BFI,ProfileSummaryInfo * PSI,DominatorTree & DT,bool HasLocalsInUsedOrAsm,DenseSet<GlobalValue::GUID> & CantBePromoted,bool IsThinLTO,std::function<const StackSafetyInfo * (const Function & F)> GetSSICallback)266 static void computeFunctionSummary(
267     ModuleSummaryIndex &Index, const Module &M, const Function &F,
268     BlockFrequencyInfo *BFI, ProfileSummaryInfo *PSI, DominatorTree &DT,
269     bool HasLocalsInUsedOrAsm, DenseSet<GlobalValue::GUID> &CantBePromoted,
270     bool IsThinLTO,
271     std::function<const StackSafetyInfo *(const Function &F)> GetSSICallback) {
272   // Summary not currently supported for anonymous functions, they should
273   // have been named.
274   assert(F.hasName());
275 
276   unsigned NumInsts = 0;
277   // Map from callee ValueId to profile count. Used to accumulate profile
278   // counts for all static calls to a given callee.
279   MapVector<ValueInfo, CalleeInfo, DenseMap<ValueInfo, unsigned>,
280             std::vector<std::pair<ValueInfo, CalleeInfo>>>
281       CallGraphEdges;
282   SetVector<ValueInfo, std::vector<ValueInfo>> RefEdges, LoadRefEdges,
283       StoreRefEdges;
284   SetVector<GlobalValue::GUID, std::vector<GlobalValue::GUID>> TypeTests;
285   SetVector<FunctionSummary::VFuncId, std::vector<FunctionSummary::VFuncId>>
286       TypeTestAssumeVCalls, TypeCheckedLoadVCalls;
287   SetVector<FunctionSummary::ConstVCall,
288             std::vector<FunctionSummary::ConstVCall>>
289       TypeTestAssumeConstVCalls, TypeCheckedLoadConstVCalls;
290   ICallPromotionAnalysis ICallAnalysis;
291   SmallPtrSet<const User *, 8> Visited;
292 
293   // Add personality function, prefix data and prologue data to function's ref
294   // list.
295   findRefEdges(Index, &F, RefEdges, Visited);
296   std::vector<const Instruction *> NonVolatileLoads;
297   std::vector<const Instruction *> NonVolatileStores;
298 
299   std::vector<CallsiteInfo> Callsites;
300   std::vector<AllocInfo> Allocs;
301 
302 #ifndef NDEBUG
303   DenseSet<const CallBase *> CallsThatMayHaveMemprofSummary;
304 #endif
305 
306   bool HasInlineAsmMaybeReferencingInternal = false;
307   bool HasIndirBranchToBlockAddress = false;
308   bool HasIFuncCall = false;
309   bool HasUnknownCall = false;
310   bool MayThrow = false;
311   for (const BasicBlock &BB : F) {
312     // We don't allow inlining of function with indirect branch to blockaddress.
313     // If the blockaddress escapes the function, e.g., via a global variable,
314     // inlining may lead to an invalid cross-function reference. So we shouldn't
315     // import such function either.
316     if (BB.hasAddressTaken()) {
317       for (User *U : BlockAddress::get(const_cast<BasicBlock *>(&BB))->users())
318         if (!isa<CallBrInst>(*U)) {
319           HasIndirBranchToBlockAddress = true;
320           break;
321         }
322     }
323 
324     for (const Instruction &I : BB) {
325       if (I.isDebugOrPseudoInst())
326         continue;
327       ++NumInsts;
328 
329       // Regular LTO module doesn't participate in ThinLTO import,
330       // so no reference from it can be read/writeonly, since this
331       // would require importing variable as local copy
332       if (IsThinLTO) {
333         if (isNonVolatileLoad(&I)) {
334           // Postpone processing of non-volatile load instructions
335           // See comments below
336           Visited.insert(&I);
337           NonVolatileLoads.push_back(&I);
338           continue;
339         } else if (isNonVolatileStore(&I)) {
340           Visited.insert(&I);
341           NonVolatileStores.push_back(&I);
342           // All references from second operand of store (destination address)
343           // can be considered write-only if they're not referenced by any
344           // non-store instruction. References from first operand of store
345           // (stored value) can't be treated either as read- or as write-only
346           // so we add them to RefEdges as we do with all other instructions
347           // except non-volatile load.
348           Value *Stored = I.getOperand(0);
349           if (auto *GV = dyn_cast<GlobalValue>(Stored))
350             // findRefEdges will try to examine GV operands, so instead
351             // of calling it we should add GV to RefEdges directly.
352             RefEdges.insert(Index.getOrInsertValueInfo(GV));
353           else if (auto *U = dyn_cast<User>(Stored))
354             findRefEdges(Index, U, RefEdges, Visited);
355           continue;
356         }
357       }
358       findRefEdges(Index, &I, RefEdges, Visited);
359       const auto *CB = dyn_cast<CallBase>(&I);
360       if (!CB) {
361         if (I.mayThrow())
362           MayThrow = true;
363         continue;
364       }
365 
366       const auto *CI = dyn_cast<CallInst>(&I);
367       // Since we don't know exactly which local values are referenced in inline
368       // assembly, conservatively mark the function as possibly referencing
369       // a local value from inline assembly to ensure we don't export a
370       // reference (which would require renaming and promotion of the
371       // referenced value).
372       if (HasLocalsInUsedOrAsm && CI && CI->isInlineAsm())
373         HasInlineAsmMaybeReferencingInternal = true;
374 
375       auto *CalledValue = CB->getCalledOperand();
376       auto *CalledFunction = CB->getCalledFunction();
377       if (CalledValue && !CalledFunction) {
378         CalledValue = CalledValue->stripPointerCasts();
379         // Stripping pointer casts can reveal a called function.
380         CalledFunction = dyn_cast<Function>(CalledValue);
381       }
382       // Check if this is an alias to a function. If so, get the
383       // called aliasee for the checks below.
384       if (auto *GA = dyn_cast<GlobalAlias>(CalledValue)) {
385         assert(!CalledFunction && "Expected null called function in callsite for alias");
386         CalledFunction = dyn_cast<Function>(GA->getAliaseeObject());
387       }
388       // Check if this is a direct call to a known function or a known
389       // intrinsic, or an indirect call with profile data.
390       if (CalledFunction) {
391         if (CI && CalledFunction->isIntrinsic()) {
392           addIntrinsicToSummary(
393               CI, TypeTests, TypeTestAssumeVCalls, TypeCheckedLoadVCalls,
394               TypeTestAssumeConstVCalls, TypeCheckedLoadConstVCalls, DT);
395           continue;
396         }
397         // We should have named any anonymous globals
398         assert(CalledFunction->hasName());
399         auto ScaledCount = PSI->getProfileCount(*CB, BFI);
400         auto Hotness = ScaledCount ? getHotness(*ScaledCount, PSI)
401                                    : CalleeInfo::HotnessType::Unknown;
402         if (ForceSummaryEdgesCold != FunctionSummary::FSHT_None)
403           Hotness = CalleeInfo::HotnessType::Cold;
404 
405         // Use the original CalledValue, in case it was an alias. We want
406         // to record the call edge to the alias in that case. Eventually
407         // an alias summary will be created to associate the alias and
408         // aliasee.
409         auto &ValueInfo = CallGraphEdges[Index.getOrInsertValueInfo(
410             cast<GlobalValue>(CalledValue))];
411         ValueInfo.updateHotness(Hotness);
412         if (CB->isTailCall())
413           ValueInfo.setHasTailCall(true);
414         // Add the relative block frequency to CalleeInfo if there is no profile
415         // information.
416         if (BFI != nullptr && Hotness == CalleeInfo::HotnessType::Unknown) {
417           uint64_t BBFreq = BFI->getBlockFreq(&BB).getFrequency();
418           uint64_t EntryFreq = BFI->getEntryFreq().getFrequency();
419           ValueInfo.updateRelBlockFreq(BBFreq, EntryFreq);
420         }
421       } else {
422         HasUnknownCall = true;
423         // If F is imported, a local linkage ifunc (e.g. target_clones on a
424         // static function) called by F will be cloned. Since summaries don't
425         // track ifunc, we do not know implementation functions referenced by
426         // the ifunc resolver need to be promoted in the exporter, and we will
427         // get linker errors due to cloned declarations for implementation
428         // functions. As a simple fix, just mark F as not eligible for import.
429         // Non-local ifunc is not cloned and does not have the issue.
430         if (auto *GI = dyn_cast_if_present<GlobalIFunc>(CalledValue))
431           if (GI->hasLocalLinkage())
432             HasIFuncCall = true;
433         // Skip inline assembly calls.
434         if (CI && CI->isInlineAsm())
435           continue;
436         // Skip direct calls.
437         if (!CalledValue || isa<Constant>(CalledValue))
438           continue;
439 
440         // Check if the instruction has a callees metadata. If so, add callees
441         // to CallGraphEdges to reflect the references from the metadata, and
442         // to enable importing for subsequent indirect call promotion and
443         // inlining.
444         if (auto *MD = I.getMetadata(LLVMContext::MD_callees)) {
445           for (const auto &Op : MD->operands()) {
446             Function *Callee = mdconst::extract_or_null<Function>(Op);
447             if (Callee)
448               CallGraphEdges[Index.getOrInsertValueInfo(Callee)];
449           }
450         }
451 
452         uint32_t NumVals, NumCandidates;
453         uint64_t TotalCount;
454         auto CandidateProfileData =
455             ICallAnalysis.getPromotionCandidatesForInstruction(
456                 &I, NumVals, TotalCount, NumCandidates);
457         for (const auto &Candidate : CandidateProfileData)
458           CallGraphEdges[Index.getOrInsertValueInfo(Candidate.Value)]
459               .updateHotness(getHotness(Candidate.Count, PSI));
460       }
461 
462       // Summarize memprof related metadata. This is only needed for ThinLTO.
463       if (!IsThinLTO)
464         continue;
465 
466       // TODO: Skip indirect calls for now. Need to handle these better, likely
467       // by creating multiple Callsites, one per target, then speculatively
468       // devirtualize while applying clone info in the ThinLTO backends. This
469       // will also be important because we will have a different set of clone
470       // versions per target. This handling needs to match that in the ThinLTO
471       // backend so we handle things consistently for matching of callsite
472       // summaries to instructions.
473       if (!CalledFunction)
474         continue;
475 
476       // Ensure we keep this analysis in sync with the handling in the ThinLTO
477       // backend (see MemProfContextDisambiguation::applyImport). Save this call
478       // so that we can skip it in checking the reverse case later.
479       assert(mayHaveMemprofSummary(CB));
480 #ifndef NDEBUG
481       CallsThatMayHaveMemprofSummary.insert(CB);
482 #endif
483 
484       // Compute the list of stack ids first (so we can trim them from the stack
485       // ids on any MIBs).
486       CallStack<MDNode, MDNode::op_iterator> InstCallsite(
487           I.getMetadata(LLVMContext::MD_callsite));
488       auto *MemProfMD = I.getMetadata(LLVMContext::MD_memprof);
489       if (MemProfMD) {
490         std::vector<MIBInfo> MIBs;
491         for (auto &MDOp : MemProfMD->operands()) {
492           auto *MIBMD = cast<const MDNode>(MDOp);
493           MDNode *StackNode = getMIBStackNode(MIBMD);
494           assert(StackNode);
495           SmallVector<unsigned> StackIdIndices;
496           CallStack<MDNode, MDNode::op_iterator> StackContext(StackNode);
497           // Collapse out any on the allocation call (inlining).
498           for (auto ContextIter =
499                    StackContext.beginAfterSharedPrefix(InstCallsite);
500                ContextIter != StackContext.end(); ++ContextIter) {
501             unsigned StackIdIdx = Index.addOrGetStackIdIndex(*ContextIter);
502             // If this is a direct recursion, simply skip the duplicate
503             // entries. If this is mutual recursion, handling is left to
504             // the LTO link analysis client.
505             if (StackIdIndices.empty() || StackIdIndices.back() != StackIdIdx)
506               StackIdIndices.push_back(StackIdIdx);
507           }
508           MIBs.push_back(
509               MIBInfo(getMIBAllocType(MIBMD), std::move(StackIdIndices)));
510         }
511         Allocs.push_back(AllocInfo(std::move(MIBs)));
512       } else if (!InstCallsite.empty()) {
513         SmallVector<unsigned> StackIdIndices;
514         for (auto StackId : InstCallsite)
515           StackIdIndices.push_back(Index.addOrGetStackIdIndex(StackId));
516         // Use the original CalledValue, in case it was an alias. We want
517         // to record the call edge to the alias in that case. Eventually
518         // an alias summary will be created to associate the alias and
519         // aliasee.
520         auto CalleeValueInfo =
521             Index.getOrInsertValueInfo(cast<GlobalValue>(CalledValue));
522         Callsites.push_back({CalleeValueInfo, StackIdIndices});
523       }
524     }
525   }
526 
527   if (PSI->hasPartialSampleProfile() && ScalePartialSampleProfileWorkingSetSize)
528     Index.addBlockCount(F.size());
529 
530   std::vector<ValueInfo> Refs;
531   if (IsThinLTO) {
532     auto AddRefEdges = [&](const std::vector<const Instruction *> &Instrs,
533                            SetVector<ValueInfo, std::vector<ValueInfo>> &Edges,
534                            SmallPtrSet<const User *, 8> &Cache) {
535       for (const auto *I : Instrs) {
536         Cache.erase(I);
537         findRefEdges(Index, I, Edges, Cache);
538       }
539     };
540 
541     // By now we processed all instructions in a function, except
542     // non-volatile loads and non-volatile value stores. Let's find
543     // ref edges for both of instruction sets
544     AddRefEdges(NonVolatileLoads, LoadRefEdges, Visited);
545     // We can add some values to the Visited set when processing load
546     // instructions which are also used by stores in NonVolatileStores.
547     // For example this can happen if we have following code:
548     //
549     // store %Derived* @foo, %Derived** bitcast (%Base** @bar to %Derived**)
550     // %42 = load %Derived*, %Derived** bitcast (%Base** @bar to %Derived**)
551     //
552     // After processing loads we'll add bitcast to the Visited set, and if
553     // we use the same set while processing stores, we'll never see store
554     // to @bar and @bar will be mistakenly treated as readonly.
555     SmallPtrSet<const llvm::User *, 8> StoreCache;
556     AddRefEdges(NonVolatileStores, StoreRefEdges, StoreCache);
557 
558     // If both load and store instruction reference the same variable
559     // we won't be able to optimize it. Add all such reference edges
560     // to RefEdges set.
561     for (const auto &VI : StoreRefEdges)
562       if (LoadRefEdges.remove(VI))
563         RefEdges.insert(VI);
564 
565     unsigned RefCnt = RefEdges.size();
566     // All new reference edges inserted in two loops below are either
567     // read or write only. They will be grouped in the end of RefEdges
568     // vector, so we can use a single integer value to identify them.
569     for (const auto &VI : LoadRefEdges)
570       RefEdges.insert(VI);
571 
572     unsigned FirstWORef = RefEdges.size();
573     for (const auto &VI : StoreRefEdges)
574       RefEdges.insert(VI);
575 
576     Refs = RefEdges.takeVector();
577     for (; RefCnt < FirstWORef; ++RefCnt)
578       Refs[RefCnt].setReadOnly();
579 
580     for (; RefCnt < Refs.size(); ++RefCnt)
581       Refs[RefCnt].setWriteOnly();
582   } else {
583     Refs = RefEdges.takeVector();
584   }
585   // Explicit add hot edges to enforce importing for designated GUIDs for
586   // sample PGO, to enable the same inlines as the profiled optimized binary.
587   for (auto &I : F.getImportGUIDs())
588     CallGraphEdges[Index.getOrInsertValueInfo(I)].updateHotness(
589         ForceSummaryEdgesCold == FunctionSummary::FSHT_All
590             ? CalleeInfo::HotnessType::Cold
591             : CalleeInfo::HotnessType::Critical);
592 
593 #ifndef NDEBUG
594   // Make sure that all calls we decided could not have memprof summaries get a
595   // false value for mayHaveMemprofSummary, to ensure that this handling remains
596   // in sync with the ThinLTO backend handling.
597   if (IsThinLTO) {
598     for (const BasicBlock &BB : F) {
599       for (const Instruction &I : BB) {
600         const auto *CB = dyn_cast<CallBase>(&I);
601         if (!CB)
602           continue;
603         // We already checked these above.
604         if (CallsThatMayHaveMemprofSummary.count(CB))
605           continue;
606         assert(!mayHaveMemprofSummary(CB));
607       }
608     }
609   }
610 #endif
611 
612   bool NonRenamableLocal = isNonRenamableLocal(F);
613   bool NotEligibleForImport = NonRenamableLocal ||
614                               HasInlineAsmMaybeReferencingInternal ||
615                               HasIndirBranchToBlockAddress || HasIFuncCall;
616   GlobalValueSummary::GVFlags Flags(
617       F.getLinkage(), F.getVisibility(), NotEligibleForImport,
618       /* Live = */ false, F.isDSOLocal(), F.canBeOmittedFromSymbolTable());
619   FunctionSummary::FFlags FunFlags{
620       F.doesNotAccessMemory(), F.onlyReadsMemory() && !F.doesNotAccessMemory(),
621       F.hasFnAttribute(Attribute::NoRecurse), F.returnDoesNotAlias(),
622       // FIXME: refactor this to use the same code that inliner is using.
623       // Don't try to import functions with noinline attribute.
624       F.getAttributes().hasFnAttr(Attribute::NoInline),
625       F.hasFnAttribute(Attribute::AlwaysInline),
626       F.hasFnAttribute(Attribute::NoUnwind), MayThrow, HasUnknownCall,
627       mustBeUnreachableFunction(F)};
628   std::vector<FunctionSummary::ParamAccess> ParamAccesses;
629   if (auto *SSI = GetSSICallback(F))
630     ParamAccesses = SSI->getParamAccesses(Index);
631   auto FuncSummary = std::make_unique<FunctionSummary>(
632       Flags, NumInsts, FunFlags, /*EntryCount=*/0, std::move(Refs),
633       CallGraphEdges.takeVector(), TypeTests.takeVector(),
634       TypeTestAssumeVCalls.takeVector(), TypeCheckedLoadVCalls.takeVector(),
635       TypeTestAssumeConstVCalls.takeVector(),
636       TypeCheckedLoadConstVCalls.takeVector(), std::move(ParamAccesses),
637       std::move(Callsites), std::move(Allocs));
638   if (NonRenamableLocal)
639     CantBePromoted.insert(F.getGUID());
640   Index.addGlobalValueSummary(F, std::move(FuncSummary));
641 }
642 
643 /// Find function pointers referenced within the given vtable initializer
644 /// (or subset of an initializer) \p I. The starting offset of \p I within
645 /// the vtable initializer is \p StartingOffset. Any discovered function
646 /// pointers are added to \p VTableFuncs along with their cumulative offset
647 /// within the initializer.
findFuncPointers(const Constant * I,uint64_t StartingOffset,const Module & M,ModuleSummaryIndex & Index,VTableFuncList & VTableFuncs)648 static void findFuncPointers(const Constant *I, uint64_t StartingOffset,
649                              const Module &M, ModuleSummaryIndex &Index,
650                              VTableFuncList &VTableFuncs) {
651   // First check if this is a function pointer.
652   if (I->getType()->isPointerTy()) {
653     auto C = I->stripPointerCasts();
654     auto A = dyn_cast<GlobalAlias>(C);
655     if (isa<Function>(C) || (A && isa<Function>(A->getAliasee()))) {
656       auto GV = dyn_cast<GlobalValue>(C);
657       assert(GV);
658       // We can disregard __cxa_pure_virtual as a possible call target, as
659       // calls to pure virtuals are UB.
660       if (GV && GV->getName() != "__cxa_pure_virtual")
661         VTableFuncs.push_back({Index.getOrInsertValueInfo(GV), StartingOffset});
662       return;
663     }
664   }
665 
666   // Walk through the elements in the constant struct or array and recursively
667   // look for virtual function pointers.
668   const DataLayout &DL = M.getDataLayout();
669   if (auto *C = dyn_cast<ConstantStruct>(I)) {
670     StructType *STy = dyn_cast<StructType>(C->getType());
671     assert(STy);
672     const StructLayout *SL = DL.getStructLayout(C->getType());
673 
674     for (auto EI : llvm::enumerate(STy->elements())) {
675       auto Offset = SL->getElementOffset(EI.index());
676       unsigned Op = SL->getElementContainingOffset(Offset);
677       findFuncPointers(cast<Constant>(I->getOperand(Op)),
678                        StartingOffset + Offset, M, Index, VTableFuncs);
679     }
680   } else if (auto *C = dyn_cast<ConstantArray>(I)) {
681     ArrayType *ATy = C->getType();
682     Type *EltTy = ATy->getElementType();
683     uint64_t EltSize = DL.getTypeAllocSize(EltTy);
684     for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i) {
685       findFuncPointers(cast<Constant>(I->getOperand(i)),
686                        StartingOffset + i * EltSize, M, Index, VTableFuncs);
687     }
688   }
689 }
690 
691 // Identify the function pointers referenced by vtable definition \p V.
computeVTableFuncs(ModuleSummaryIndex & Index,const GlobalVariable & V,const Module & M,VTableFuncList & VTableFuncs)692 static void computeVTableFuncs(ModuleSummaryIndex &Index,
693                                const GlobalVariable &V, const Module &M,
694                                VTableFuncList &VTableFuncs) {
695   if (!V.isConstant())
696     return;
697 
698   findFuncPointers(V.getInitializer(), /*StartingOffset=*/0, M, Index,
699                    VTableFuncs);
700 
701 #ifndef NDEBUG
702   // Validate that the VTableFuncs list is ordered by offset.
703   uint64_t PrevOffset = 0;
704   for (auto &P : VTableFuncs) {
705     // The findVFuncPointers traversal should have encountered the
706     // functions in offset order. We need to use ">=" since PrevOffset
707     // starts at 0.
708     assert(P.VTableOffset >= PrevOffset);
709     PrevOffset = P.VTableOffset;
710   }
711 #endif
712 }
713 
714 /// Record vtable definition \p V for each type metadata it references.
715 static void
recordTypeIdCompatibleVtableReferences(ModuleSummaryIndex & Index,const GlobalVariable & V,SmallVectorImpl<MDNode * > & Types)716 recordTypeIdCompatibleVtableReferences(ModuleSummaryIndex &Index,
717                                        const GlobalVariable &V,
718                                        SmallVectorImpl<MDNode *> &Types) {
719   for (MDNode *Type : Types) {
720     auto TypeID = Type->getOperand(1).get();
721 
722     uint64_t Offset =
723         cast<ConstantInt>(
724             cast<ConstantAsMetadata>(Type->getOperand(0))->getValue())
725             ->getZExtValue();
726 
727     if (auto *TypeId = dyn_cast<MDString>(TypeID))
728       Index.getOrInsertTypeIdCompatibleVtableSummary(TypeId->getString())
729           .push_back({Offset, Index.getOrInsertValueInfo(&V)});
730   }
731 }
732 
computeVariableSummary(ModuleSummaryIndex & Index,const GlobalVariable & V,DenseSet<GlobalValue::GUID> & CantBePromoted,const Module & M,SmallVectorImpl<MDNode * > & Types)733 static void computeVariableSummary(ModuleSummaryIndex &Index,
734                                    const GlobalVariable &V,
735                                    DenseSet<GlobalValue::GUID> &CantBePromoted,
736                                    const Module &M,
737                                    SmallVectorImpl<MDNode *> &Types) {
738   SetVector<ValueInfo, std::vector<ValueInfo>> RefEdges;
739   SmallPtrSet<const User *, 8> Visited;
740   bool HasBlockAddress = findRefEdges(Index, &V, RefEdges, Visited);
741   bool NonRenamableLocal = isNonRenamableLocal(V);
742   GlobalValueSummary::GVFlags Flags(
743       V.getLinkage(), V.getVisibility(), NonRenamableLocal,
744       /* Live = */ false, V.isDSOLocal(), V.canBeOmittedFromSymbolTable());
745 
746   VTableFuncList VTableFuncs;
747   // If splitting is not enabled, then we compute the summary information
748   // necessary for index-based whole program devirtualization.
749   if (!Index.enableSplitLTOUnit()) {
750     Types.clear();
751     V.getMetadata(LLVMContext::MD_type, Types);
752     if (!Types.empty()) {
753       // Identify the function pointers referenced by this vtable definition.
754       computeVTableFuncs(Index, V, M, VTableFuncs);
755 
756       // Record this vtable definition for each type metadata it references.
757       recordTypeIdCompatibleVtableReferences(Index, V, Types);
758     }
759   }
760 
761   // Don't mark variables we won't be able to internalize as read/write-only.
762   bool CanBeInternalized =
763       !V.hasComdat() && !V.hasAppendingLinkage() && !V.isInterposable() &&
764       !V.hasAvailableExternallyLinkage() && !V.hasDLLExportStorageClass();
765   bool Constant = V.isConstant();
766   GlobalVarSummary::GVarFlags VarFlags(CanBeInternalized,
767                                        Constant ? false : CanBeInternalized,
768                                        Constant, V.getVCallVisibility());
769   auto GVarSummary = std::make_unique<GlobalVarSummary>(Flags, VarFlags,
770                                                          RefEdges.takeVector());
771   if (NonRenamableLocal)
772     CantBePromoted.insert(V.getGUID());
773   if (HasBlockAddress)
774     GVarSummary->setNotEligibleToImport();
775   if (!VTableFuncs.empty())
776     GVarSummary->setVTableFuncs(VTableFuncs);
777   Index.addGlobalValueSummary(V, std::move(GVarSummary));
778 }
779 
computeAliasSummary(ModuleSummaryIndex & Index,const GlobalAlias & A,DenseSet<GlobalValue::GUID> & CantBePromoted)780 static void computeAliasSummary(ModuleSummaryIndex &Index, const GlobalAlias &A,
781                                 DenseSet<GlobalValue::GUID> &CantBePromoted) {
782   // Skip summary for indirect function aliases as summary for aliasee will not
783   // be emitted.
784   const GlobalObject *Aliasee = A.getAliaseeObject();
785   if (isa<GlobalIFunc>(Aliasee))
786     return;
787   bool NonRenamableLocal = isNonRenamableLocal(A);
788   GlobalValueSummary::GVFlags Flags(
789       A.getLinkage(), A.getVisibility(), NonRenamableLocal,
790       /* Live = */ false, A.isDSOLocal(), A.canBeOmittedFromSymbolTable());
791   auto AS = std::make_unique<AliasSummary>(Flags);
792   auto AliaseeVI = Index.getValueInfo(Aliasee->getGUID());
793   assert(AliaseeVI && "Alias expects aliasee summary to be available");
794   assert(AliaseeVI.getSummaryList().size() == 1 &&
795          "Expected a single entry per aliasee in per-module index");
796   AS->setAliasee(AliaseeVI, AliaseeVI.getSummaryList()[0].get());
797   if (NonRenamableLocal)
798     CantBePromoted.insert(A.getGUID());
799   Index.addGlobalValueSummary(A, std::move(AS));
800 }
801 
802 // Set LiveRoot flag on entries matching the given value name.
setLiveRoot(ModuleSummaryIndex & Index,StringRef Name)803 static void setLiveRoot(ModuleSummaryIndex &Index, StringRef Name) {
804   if (ValueInfo VI = Index.getValueInfo(GlobalValue::getGUID(Name)))
805     for (const auto &Summary : VI.getSummaryList())
806       Summary->setLive(true);
807 }
808 
buildModuleSummaryIndex(const Module & M,std::function<BlockFrequencyInfo * (const Function & F)> GetBFICallback,ProfileSummaryInfo * PSI,std::function<const StackSafetyInfo * (const Function & F)> GetSSICallback)809 ModuleSummaryIndex llvm::buildModuleSummaryIndex(
810     const Module &M,
811     std::function<BlockFrequencyInfo *(const Function &F)> GetBFICallback,
812     ProfileSummaryInfo *PSI,
813     std::function<const StackSafetyInfo *(const Function &F)> GetSSICallback) {
814   assert(PSI);
815   bool EnableSplitLTOUnit = false;
816   bool UnifiedLTO = false;
817   if (auto *MD = mdconst::extract_or_null<ConstantInt>(
818           M.getModuleFlag("EnableSplitLTOUnit")))
819     EnableSplitLTOUnit = MD->getZExtValue();
820   if (auto *MD =
821           mdconst::extract_or_null<ConstantInt>(M.getModuleFlag("UnifiedLTO")))
822     UnifiedLTO = MD->getZExtValue();
823   ModuleSummaryIndex Index(/*HaveGVs=*/true, EnableSplitLTOUnit, UnifiedLTO);
824 
825   // Identify the local values in the llvm.used and llvm.compiler.used sets,
826   // which should not be exported as they would then require renaming and
827   // promotion, but we may have opaque uses e.g. in inline asm. We collect them
828   // here because we use this information to mark functions containing inline
829   // assembly calls as not importable.
830   SmallPtrSet<GlobalValue *, 4> LocalsUsed;
831   SmallVector<GlobalValue *, 4> Used;
832   // First collect those in the llvm.used set.
833   collectUsedGlobalVariables(M, Used, /*CompilerUsed=*/false);
834   // Next collect those in the llvm.compiler.used set.
835   collectUsedGlobalVariables(M, Used, /*CompilerUsed=*/true);
836   DenseSet<GlobalValue::GUID> CantBePromoted;
837   for (auto *V : Used) {
838     if (V->hasLocalLinkage()) {
839       LocalsUsed.insert(V);
840       CantBePromoted.insert(V->getGUID());
841     }
842   }
843 
844   bool HasLocalInlineAsmSymbol = false;
845   if (!M.getModuleInlineAsm().empty()) {
846     // Collect the local values defined by module level asm, and set up
847     // summaries for these symbols so that they can be marked as NoRename,
848     // to prevent export of any use of them in regular IR that would require
849     // renaming within the module level asm. Note we don't need to create a
850     // summary for weak or global defs, as they don't need to be flagged as
851     // NoRename, and defs in module level asm can't be imported anyway.
852     // Also, any values used but not defined within module level asm should
853     // be listed on the llvm.used or llvm.compiler.used global and marked as
854     // referenced from there.
855     ModuleSymbolTable::CollectAsmSymbols(
856         M, [&](StringRef Name, object::BasicSymbolRef::Flags Flags) {
857           // Symbols not marked as Weak or Global are local definitions.
858           if (Flags & (object::BasicSymbolRef::SF_Weak |
859                        object::BasicSymbolRef::SF_Global))
860             return;
861           HasLocalInlineAsmSymbol = true;
862           GlobalValue *GV = M.getNamedValue(Name);
863           if (!GV)
864             return;
865           assert(GV->isDeclaration() && "Def in module asm already has definition");
866           GlobalValueSummary::GVFlags GVFlags(
867               GlobalValue::InternalLinkage, GlobalValue::DefaultVisibility,
868               /* NotEligibleToImport = */ true,
869               /* Live = */ true,
870               /* Local */ GV->isDSOLocal(), GV->canBeOmittedFromSymbolTable());
871           CantBePromoted.insert(GV->getGUID());
872           // Create the appropriate summary type.
873           if (Function *F = dyn_cast<Function>(GV)) {
874             std::unique_ptr<FunctionSummary> Summary =
875                 std::make_unique<FunctionSummary>(
876                     GVFlags, /*InstCount=*/0,
877                     FunctionSummary::FFlags{
878                         F->hasFnAttribute(Attribute::ReadNone),
879                         F->hasFnAttribute(Attribute::ReadOnly),
880                         F->hasFnAttribute(Attribute::NoRecurse),
881                         F->returnDoesNotAlias(),
882                         /* NoInline = */ false,
883                         F->hasFnAttribute(Attribute::AlwaysInline),
884                         F->hasFnAttribute(Attribute::NoUnwind),
885                         /* MayThrow */ true,
886                         /* HasUnknownCall */ true,
887                         /* MustBeUnreachable */ false},
888                     /*EntryCount=*/0, ArrayRef<ValueInfo>{},
889                     ArrayRef<FunctionSummary::EdgeTy>{},
890                     ArrayRef<GlobalValue::GUID>{},
891                     ArrayRef<FunctionSummary::VFuncId>{},
892                     ArrayRef<FunctionSummary::VFuncId>{},
893                     ArrayRef<FunctionSummary::ConstVCall>{},
894                     ArrayRef<FunctionSummary::ConstVCall>{},
895                     ArrayRef<FunctionSummary::ParamAccess>{},
896                     ArrayRef<CallsiteInfo>{}, ArrayRef<AllocInfo>{});
897             Index.addGlobalValueSummary(*GV, std::move(Summary));
898           } else {
899             std::unique_ptr<GlobalVarSummary> Summary =
900                 std::make_unique<GlobalVarSummary>(
901                     GVFlags,
902                     GlobalVarSummary::GVarFlags(
903                         false, false, cast<GlobalVariable>(GV)->isConstant(),
904                         GlobalObject::VCallVisibilityPublic),
905                     ArrayRef<ValueInfo>{});
906             Index.addGlobalValueSummary(*GV, std::move(Summary));
907           }
908         });
909   }
910 
911   bool IsThinLTO = true;
912   if (auto *MD =
913           mdconst::extract_or_null<ConstantInt>(M.getModuleFlag("ThinLTO")))
914     IsThinLTO = MD->getZExtValue();
915 
916   // Compute summaries for all functions defined in module, and save in the
917   // index.
918   for (const auto &F : M) {
919     if (F.isDeclaration())
920       continue;
921 
922     DominatorTree DT(const_cast<Function &>(F));
923     BlockFrequencyInfo *BFI = nullptr;
924     std::unique_ptr<BlockFrequencyInfo> BFIPtr;
925     if (GetBFICallback)
926       BFI = GetBFICallback(F);
927     else if (F.hasProfileData()) {
928       LoopInfo LI{DT};
929       BranchProbabilityInfo BPI{F, LI};
930       BFIPtr = std::make_unique<BlockFrequencyInfo>(F, BPI, LI);
931       BFI = BFIPtr.get();
932     }
933 
934     computeFunctionSummary(Index, M, F, BFI, PSI, DT,
935                            !LocalsUsed.empty() || HasLocalInlineAsmSymbol,
936                            CantBePromoted, IsThinLTO, GetSSICallback);
937   }
938 
939   // Compute summaries for all variables defined in module, and save in the
940   // index.
941   SmallVector<MDNode *, 2> Types;
942   for (const GlobalVariable &G : M.globals()) {
943     if (G.isDeclaration())
944       continue;
945     computeVariableSummary(Index, G, CantBePromoted, M, Types);
946   }
947 
948   // Compute summaries for all aliases defined in module, and save in the
949   // index.
950   for (const GlobalAlias &A : M.aliases())
951     computeAliasSummary(Index, A, CantBePromoted);
952 
953   // Iterate through ifuncs, set their resolvers all alive.
954   for (const GlobalIFunc &I : M.ifuncs()) {
955     I.applyAlongResolverPath([&Index](const GlobalValue &GV) {
956       Index.getGlobalValueSummary(GV)->setLive(true);
957     });
958   }
959 
960   for (auto *V : LocalsUsed) {
961     auto *Summary = Index.getGlobalValueSummary(*V);
962     assert(Summary && "Missing summary for global value");
963     Summary->setNotEligibleToImport();
964   }
965 
966   // The linker doesn't know about these LLVM produced values, so we need
967   // to flag them as live in the index to ensure index-based dead value
968   // analysis treats them as live roots of the analysis.
969   setLiveRoot(Index, "llvm.used");
970   setLiveRoot(Index, "llvm.compiler.used");
971   setLiveRoot(Index, "llvm.global_ctors");
972   setLiveRoot(Index, "llvm.global_dtors");
973   setLiveRoot(Index, "llvm.global.annotations");
974 
975   for (auto &GlobalList : Index) {
976     // Ignore entries for references that are undefined in the current module.
977     if (GlobalList.second.SummaryList.empty())
978       continue;
979 
980     assert(GlobalList.second.SummaryList.size() == 1 &&
981            "Expected module's index to have one summary per GUID");
982     auto &Summary = GlobalList.second.SummaryList[0];
983     if (!IsThinLTO) {
984       Summary->setNotEligibleToImport();
985       continue;
986     }
987 
988     bool AllRefsCanBeExternallyReferenced =
989         llvm::all_of(Summary->refs(), [&](const ValueInfo &VI) {
990           return !CantBePromoted.count(VI.getGUID());
991         });
992     if (!AllRefsCanBeExternallyReferenced) {
993       Summary->setNotEligibleToImport();
994       continue;
995     }
996 
997     if (auto *FuncSummary = dyn_cast<FunctionSummary>(Summary.get())) {
998       bool AllCallsCanBeExternallyReferenced = llvm::all_of(
999           FuncSummary->calls(), [&](const FunctionSummary::EdgeTy &Edge) {
1000             return !CantBePromoted.count(Edge.first.getGUID());
1001           });
1002       if (!AllCallsCanBeExternallyReferenced)
1003         Summary->setNotEligibleToImport();
1004     }
1005   }
1006 
1007   if (!ModuleSummaryDotFile.empty()) {
1008     std::error_code EC;
1009     raw_fd_ostream OSDot(ModuleSummaryDotFile, EC, sys::fs::OpenFlags::OF_None);
1010     if (EC)
1011       report_fatal_error(Twine("Failed to open dot file ") +
1012                          ModuleSummaryDotFile + ": " + EC.message() + "\n");
1013     Index.exportToDot(OSDot, {});
1014   }
1015 
1016   return Index;
1017 }
1018 
1019 AnalysisKey ModuleSummaryIndexAnalysis::Key;
1020 
1021 ModuleSummaryIndex
run(Module & M,ModuleAnalysisManager & AM)1022 ModuleSummaryIndexAnalysis::run(Module &M, ModuleAnalysisManager &AM) {
1023   ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M);
1024   auto &FAM = AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
1025   bool NeedSSI = needsParamAccessSummary(M);
1026   return buildModuleSummaryIndex(
1027       M,
1028       [&FAM](const Function &F) {
1029         return &FAM.getResult<BlockFrequencyAnalysis>(
1030             *const_cast<Function *>(&F));
1031       },
1032       &PSI,
1033       [&FAM, NeedSSI](const Function &F) -> const StackSafetyInfo * {
1034         return NeedSSI ? &FAM.getResult<StackSafetyAnalysis>(
1035                              const_cast<Function &>(F))
1036                        : nullptr;
1037       });
1038 }
1039 
1040 char ModuleSummaryIndexWrapperPass::ID = 0;
1041 
1042 INITIALIZE_PASS_BEGIN(ModuleSummaryIndexWrapperPass, "module-summary-analysis",
1043                       "Module Summary Analysis", false, true)
INITIALIZE_PASS_DEPENDENCY(BlockFrequencyInfoWrapperPass)1044 INITIALIZE_PASS_DEPENDENCY(BlockFrequencyInfoWrapperPass)
1045 INITIALIZE_PASS_DEPENDENCY(ProfileSummaryInfoWrapperPass)
1046 INITIALIZE_PASS_DEPENDENCY(StackSafetyInfoWrapperPass)
1047 INITIALIZE_PASS_END(ModuleSummaryIndexWrapperPass, "module-summary-analysis",
1048                     "Module Summary Analysis", false, true)
1049 
1050 ModulePass *llvm::createModuleSummaryIndexWrapperPass() {
1051   return new ModuleSummaryIndexWrapperPass();
1052 }
1053 
ModuleSummaryIndexWrapperPass()1054 ModuleSummaryIndexWrapperPass::ModuleSummaryIndexWrapperPass()
1055     : ModulePass(ID) {
1056   initializeModuleSummaryIndexWrapperPassPass(*PassRegistry::getPassRegistry());
1057 }
1058 
runOnModule(Module & M)1059 bool ModuleSummaryIndexWrapperPass::runOnModule(Module &M) {
1060   auto *PSI = &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI();
1061   bool NeedSSI = needsParamAccessSummary(M);
1062   Index.emplace(buildModuleSummaryIndex(
1063       M,
1064       [this](const Function &F) {
1065         return &(this->getAnalysis<BlockFrequencyInfoWrapperPass>(
1066                          *const_cast<Function *>(&F))
1067                      .getBFI());
1068       },
1069       PSI,
1070       [&](const Function &F) -> const StackSafetyInfo * {
1071         return NeedSSI ? &getAnalysis<StackSafetyInfoWrapperPass>(
1072                               const_cast<Function &>(F))
1073                               .getResult()
1074                        : nullptr;
1075       }));
1076   return false;
1077 }
1078 
doFinalization(Module & M)1079 bool ModuleSummaryIndexWrapperPass::doFinalization(Module &M) {
1080   Index.reset();
1081   return false;
1082 }
1083 
getAnalysisUsage(AnalysisUsage & AU) const1084 void ModuleSummaryIndexWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const {
1085   AU.setPreservesAll();
1086   AU.addRequired<BlockFrequencyInfoWrapperPass>();
1087   AU.addRequired<ProfileSummaryInfoWrapperPass>();
1088   AU.addRequired<StackSafetyInfoWrapperPass>();
1089 }
1090 
1091 char ImmutableModuleSummaryIndexWrapperPass::ID = 0;
1092 
ImmutableModuleSummaryIndexWrapperPass(const ModuleSummaryIndex * Index)1093 ImmutableModuleSummaryIndexWrapperPass::ImmutableModuleSummaryIndexWrapperPass(
1094     const ModuleSummaryIndex *Index)
1095     : ImmutablePass(ID), Index(Index) {
1096   initializeImmutableModuleSummaryIndexWrapperPassPass(
1097       *PassRegistry::getPassRegistry());
1098 }
1099 
getAnalysisUsage(AnalysisUsage & AU) const1100 void ImmutableModuleSummaryIndexWrapperPass::getAnalysisUsage(
1101     AnalysisUsage &AU) const {
1102   AU.setPreservesAll();
1103 }
1104 
createImmutableModuleSummaryIndexWrapperPass(const ModuleSummaryIndex * Index)1105 ImmutablePass *llvm::createImmutableModuleSummaryIndexWrapperPass(
1106     const ModuleSummaryIndex *Index) {
1107   return new ImmutableModuleSummaryIndexWrapperPass(Index);
1108 }
1109 
1110 INITIALIZE_PASS(ImmutableModuleSummaryIndexWrapperPass, "module-summary-info",
1111                 "Module summary info", false, true)
1112 
mayHaveMemprofSummary(const CallBase * CB)1113 bool llvm::mayHaveMemprofSummary(const CallBase *CB) {
1114   if (!CB)
1115     return false;
1116   if (CB->isDebugOrPseudoInst())
1117     return false;
1118   auto *CI = dyn_cast<CallInst>(CB);
1119   auto *CalledValue = CB->getCalledOperand();
1120   auto *CalledFunction = CB->getCalledFunction();
1121   if (CalledValue && !CalledFunction) {
1122     CalledValue = CalledValue->stripPointerCasts();
1123     // Stripping pointer casts can reveal a called function.
1124     CalledFunction = dyn_cast<Function>(CalledValue);
1125   }
1126   // Check if this is an alias to a function. If so, get the
1127   // called aliasee for the checks below.
1128   if (auto *GA = dyn_cast<GlobalAlias>(CalledValue)) {
1129     assert(!CalledFunction &&
1130            "Expected null called function in callsite for alias");
1131     CalledFunction = dyn_cast<Function>(GA->getAliaseeObject());
1132   }
1133   // Check if this is a direct call to a known function or a known
1134   // intrinsic, or an indirect call with profile data.
1135   if (CalledFunction) {
1136     if (CI && CalledFunction->isIntrinsic())
1137       return false;
1138   } else {
1139     // TODO: For now skip indirect calls. See comments in
1140     // computeFunctionSummary for what is needed to handle this.
1141     return false;
1142   }
1143   return true;
1144 }
1145