1////===- SampleProfileLoadBaseImpl.h - Profile loader base impl --*- C++-*-===//
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/// \file
10/// This file provides the interface for the sampled PGO profile loader base
11/// implementation.
12//
13//===----------------------------------------------------------------------===//
14
15#ifndef LLVM_TRANSFORMS_UTILS_SAMPLEPROFILELOADERBASEIMPL_H
16#define LLVM_TRANSFORMS_UTILS_SAMPLEPROFILELOADERBASEIMPL_H
17
18#include "llvm/ADT/ArrayRef.h"
19#include "llvm/ADT/DenseMap.h"
20#include "llvm/ADT/DenseSet.h"
21#include "llvm/ADT/IntrusiveRefCntPtr.h"
22#include "llvm/ADT/SmallPtrSet.h"
23#include "llvm/ADT/SmallSet.h"
24#include "llvm/ADT/SmallVector.h"
25#include "llvm/Analysis/LazyCallGraph.h"
26#include "llvm/Analysis/LoopInfo.h"
27#include "llvm/Analysis/OptimizationRemarkEmitter.h"
28#include "llvm/Analysis/PostDominators.h"
29#include "llvm/IR/BasicBlock.h"
30#include "llvm/IR/CFG.h"
31#include "llvm/IR/DebugInfoMetadata.h"
32#include "llvm/IR/DebugLoc.h"
33#include "llvm/IR/Dominators.h"
34#include "llvm/IR/Function.h"
35#include "llvm/IR/Instruction.h"
36#include "llvm/IR/Instructions.h"
37#include "llvm/IR/Module.h"
38#include "llvm/IR/PseudoProbe.h"
39#include "llvm/ProfileData/SampleProf.h"
40#include "llvm/ProfileData/SampleProfReader.h"
41#include "llvm/Support/CommandLine.h"
42#include "llvm/Support/GenericDomTree.h"
43#include "llvm/Support/raw_ostream.h"
44#include "llvm/Transforms/Utils/SampleProfileInference.h"
45#include "llvm/Transforms/Utils/SampleProfileLoaderBaseUtil.h"
46
47namespace llvm {
48using namespace sampleprof;
49using namespace sampleprofutil;
50using ProfileCount = Function::ProfileCount;
51
52namespace vfs {
53class FileSystem;
54} // namespace vfs
55
56#define DEBUG_TYPE "sample-profile-impl"
57
58namespace afdo_detail {
59
60template <typename BlockT> struct IRTraits;
61template <> struct IRTraits<BasicBlock> {
62 using InstructionT = Instruction;
63 using BasicBlockT = BasicBlock;
64 using FunctionT = Function;
65 using BlockFrequencyInfoT = BlockFrequencyInfo;
66 using LoopT = Loop;
67 using LoopInfoPtrT = std::unique_ptr<LoopInfo>;
68 using DominatorTreePtrT = std::unique_ptr<DominatorTree>;
69 using PostDominatorTreeT = PostDominatorTree;
70 using PostDominatorTreePtrT = std::unique_ptr<PostDominatorTree>;
71 using OptRemarkEmitterT = OptimizationRemarkEmitter;
72 using OptRemarkAnalysisT = OptimizationRemarkAnalysis;
73 using PredRangeT = pred_range;
74 using SuccRangeT = succ_range;
75 static Function &getFunction(Function &F) { return F; }
76 static const BasicBlock *getEntryBB(const Function *F) {
77 return &F->getEntryBlock();
78 }
79 static pred_range getPredecessors(BasicBlock *BB) { return predecessors(BB); }
80 static succ_range getSuccessors(BasicBlock *BB) { return successors(BB); }
81};
82
83} // end namespace afdo_detail
84
85// This class serves sample counts correlation for SampleProfileLoader by
86// analyzing pseudo probes and their function descriptors injected by
87// SampleProfileProber.
88class PseudoProbeManager {
89 DenseMap<uint64_t, PseudoProbeDescriptor> GUIDToProbeDescMap;
90
91public:
92 PseudoProbeManager(const Module &M) {
93 if (NamedMDNode *FuncInfo =
94 M.getNamedMetadata(Name: PseudoProbeDescMetadataName)) {
95 for (const auto *Operand : FuncInfo->operands()) {
96 const auto *MD = cast<MDNode>(Val: Operand);
97 auto GUID = mdconst::dyn_extract<ConstantInt>(MD: MD->getOperand(I: 0))
98 ->getZExtValue();
99 auto Hash = mdconst::dyn_extract<ConstantInt>(MD: MD->getOperand(I: 1))
100 ->getZExtValue();
101 GUIDToProbeDescMap.try_emplace(Key: GUID, Args: PseudoProbeDescriptor(GUID, Hash));
102 }
103 }
104 }
105
106 const PseudoProbeDescriptor *getDesc(uint64_t GUID) const {
107 auto I = GUIDToProbeDescMap.find(Val: GUID);
108 return I == GUIDToProbeDescMap.end() ? nullptr : &I->second;
109 }
110
111 const PseudoProbeDescriptor *getDesc(StringRef FProfileName) const {
112 return getDesc(GUID: Function::getGUIDAssumingExternalLinkage(GlobalName: FProfileName));
113 }
114
115 const PseudoProbeDescriptor *getDesc(const Function &F) const {
116 return getDesc(GUID: Function::getGUIDAssumingExternalLinkage(
117 GlobalName: FunctionSamples::getCanonicalFnName(F)));
118 }
119
120 bool profileIsHashMismatched(const PseudoProbeDescriptor &FuncDesc,
121 const FunctionSamples &Samples) const {
122 return FuncDesc.getFunctionHash() != Samples.getFunctionHash();
123 }
124
125 bool moduleIsProbed(const Module &M) const {
126 return M.getNamedMetadata(Name: PseudoProbeDescMetadataName);
127 }
128
129 bool profileIsValid(const Function &F, const FunctionSamples &Samples) const {
130 const auto *Desc = getDesc(F);
131 bool IsAvailableExternallyLinkage =
132 GlobalValue::isAvailableExternallyLinkage(Linkage: F.getLinkage());
133 // Always check the function attribute to determine checksum mismatch for
134 // `available_externally` functions even if their desc are available. This
135 // is because the desc is computed based on the original internal function
136 // and it's substituted by the `available_externally` function during link
137 // time. However, when unstable IR or ODR violation issue occurs, the
138 // definitions of the same function across different translation units could
139 // be different and result in different checksums. So we should use the
140 // state from the new (available_externally) function, which is saved in its
141 // attribute.
142 // TODO: If the function's profile only exists as nested inlinee profile in
143 // a different module, we don't have the attr mismatch state(unknown), we
144 // need to fix it later.
145 if (IsAvailableExternallyLinkage || !Desc)
146 return !F.hasFnAttribute(Kind: "profile-checksum-mismatch");
147
148 return Desc && !profileIsHashMismatched(FuncDesc: *Desc, Samples);
149 }
150};
151
152
153
154extern cl::opt<bool> SampleProfileUseProfi;
155
156static inline bool skipProfileForFunction(const Function &F) {
157 return F.isDeclaration() || !F.hasFnAttribute(Kind: "use-sample-profile");
158}
159
160static inline void
161buildTopDownFuncOrder(LazyCallGraph &CG,
162 std::vector<Function *> &FunctionOrderList) {
163 CG.buildRefSCCs();
164 for (LazyCallGraph::RefSCC &RC : CG.postorder_ref_sccs()) {
165 for (LazyCallGraph::SCC &C : RC) {
166 for (LazyCallGraph::Node &N : C) {
167 Function &F = N.getFunction();
168 if (!skipProfileForFunction(F))
169 FunctionOrderList.push_back(x: &F);
170 }
171 }
172 }
173 std::reverse(first: FunctionOrderList.begin(), last: FunctionOrderList.end());
174}
175
176template <typename FT> class SampleProfileLoaderBaseImpl {
177public:
178 SampleProfileLoaderBaseImpl(std::string Name, std::string RemapName,
179 IntrusiveRefCntPtr<vfs::FileSystem> FS)
180 : Filename(Name), RemappingFilename(RemapName), FS(std::move(FS)) {}
181 void dump() { Reader->dump(); }
182
183 using NodeRef = typename GraphTraits<FT *>::NodeRef;
184 using BT = std::remove_pointer_t<NodeRef>;
185 using InstructionT = typename afdo_detail::IRTraits<BT>::InstructionT;
186 using BasicBlockT = typename afdo_detail::IRTraits<BT>::BasicBlockT;
187 using BlockFrequencyInfoT =
188 typename afdo_detail::IRTraits<BT>::BlockFrequencyInfoT;
189 using FunctionT = typename afdo_detail::IRTraits<BT>::FunctionT;
190 using LoopT = typename afdo_detail::IRTraits<BT>::LoopT;
191 using LoopInfoPtrT = typename afdo_detail::IRTraits<BT>::LoopInfoPtrT;
192 using DominatorTreePtrT =
193 typename afdo_detail::IRTraits<BT>::DominatorTreePtrT;
194 using PostDominatorTreePtrT =
195 typename afdo_detail::IRTraits<BT>::PostDominatorTreePtrT;
196 using PostDominatorTreeT =
197 typename afdo_detail::IRTraits<BT>::PostDominatorTreeT;
198 using OptRemarkEmitterT =
199 typename afdo_detail::IRTraits<BT>::OptRemarkEmitterT;
200 using OptRemarkAnalysisT =
201 typename afdo_detail::IRTraits<BT>::OptRemarkAnalysisT;
202 using PredRangeT = typename afdo_detail::IRTraits<BT>::PredRangeT;
203 using SuccRangeT = typename afdo_detail::IRTraits<BT>::SuccRangeT;
204
205 using BlockWeightMap = DenseMap<const BasicBlockT *, uint64_t>;
206 using EquivalenceClassMap =
207 DenseMap<const BasicBlockT *, const BasicBlockT *>;
208 using Edge = std::pair<const BasicBlockT *, const BasicBlockT *>;
209 using EdgeWeightMap = DenseMap<Edge, uint64_t>;
210 using BlockEdgeMap =
211 DenseMap<const BasicBlockT *, SmallVector<const BasicBlockT *, 8>>;
212
213protected:
214 ~SampleProfileLoaderBaseImpl() = default;
215 friend class SampleCoverageTracker;
216
217 Function &getFunction(FunctionT &F) {
218 return afdo_detail::IRTraits<BT>::getFunction(F);
219 }
220 const BasicBlockT *getEntryBB(const FunctionT *F) {
221 return afdo_detail::IRTraits<BT>::getEntryBB(F);
222 }
223 PredRangeT getPredecessors(BasicBlockT *BB) {
224 return afdo_detail::IRTraits<BT>::getPredecessors(BB);
225 }
226 SuccRangeT getSuccessors(BasicBlockT *BB) {
227 return afdo_detail::IRTraits<BT>::getSuccessors(BB);
228 }
229
230 unsigned getFunctionLoc(FunctionT &Func);
231 virtual ErrorOr<uint64_t> getInstWeight(const InstructionT &Inst);
232 ErrorOr<uint64_t> getInstWeightImpl(const InstructionT &Inst);
233 virtual ErrorOr<uint64_t> getProbeWeight(const InstructionT &Inst);
234 ErrorOr<uint64_t> getBlockWeight(const BasicBlockT *BB);
235 mutable DenseMap<const DILocation *, const FunctionSamples *>
236 DILocation2SampleMap;
237 virtual const FunctionSamples *
238 findFunctionSamples(const InstructionT &I) const;
239 void printEdgeWeight(raw_ostream &OS, Edge E);
240 void printBlockWeight(raw_ostream &OS, const BasicBlockT *BB) const;
241 void printBlockEquivalence(raw_ostream &OS, const BasicBlockT *BB);
242 bool computeBlockWeights(FunctionT &F);
243 void findEquivalenceClasses(FunctionT &F);
244 void findEquivalencesFor(BasicBlockT *BB1,
245 ArrayRef<BasicBlockT *> Descendants,
246 PostDominatorTreeT *DomTree);
247 void propagateWeights(FunctionT &F);
248 void applyProfi(FunctionT &F, BlockEdgeMap &Successors,
249 BlockWeightMap &SampleBlockWeights,
250 BlockWeightMap &BlockWeights, EdgeWeightMap &EdgeWeights);
251 uint64_t visitEdge(Edge E, unsigned *NumUnknownEdges, Edge *UnknownEdge);
252 void buildEdges(FunctionT &F);
253 bool propagateThroughEdges(FunctionT &F, bool UpdateBlockCount);
254 void clearFunctionData(bool ResetDT = true);
255 void computeDominanceAndLoopInfo(FunctionT &F);
256 bool
257 computeAndPropagateWeights(FunctionT &F,
258 const DenseSet<GlobalValue::GUID> &InlinedGUIDs);
259 void initWeightPropagation(FunctionT &F,
260 const DenseSet<GlobalValue::GUID> &InlinedGUIDs);
261 void
262 finalizeWeightPropagation(FunctionT &F,
263 const DenseSet<GlobalValue::GUID> &InlinedGUIDs);
264 void emitCoverageRemarks(FunctionT &F);
265
266 /// Map basic blocks to their computed weights.
267 ///
268 /// The weight of a basic block is defined to be the maximum
269 /// of all the instruction weights in that block.
270 BlockWeightMap BlockWeights;
271
272 /// Map edges to their computed weights.
273 ///
274 /// Edge weights are computed by propagating basic block weights in
275 /// SampleProfile::propagateWeights.
276 EdgeWeightMap EdgeWeights;
277
278 /// Set of visited blocks during propagation.
279 SmallPtrSet<const BasicBlockT *, 32> VisitedBlocks;
280
281 /// Set of visited edges during propagation.
282 SmallSet<Edge, 32> VisitedEdges;
283
284 /// Equivalence classes for block weights.
285 ///
286 /// Two blocks BB1 and BB2 are in the same equivalence class if they
287 /// dominate and post-dominate each other, and they are in the same loop
288 /// nest. When this happens, the two blocks are guaranteed to execute
289 /// the same number of times.
290 EquivalenceClassMap EquivalenceClass;
291
292 /// Dominance, post-dominance and loop information.
293 DominatorTreePtrT DT;
294 PostDominatorTreePtrT PDT;
295 LoopInfoPtrT LI;
296
297 /// Predecessors for each basic block in the CFG.
298 BlockEdgeMap Predecessors;
299
300 /// Successors for each basic block in the CFG.
301 BlockEdgeMap Successors;
302
303 /// Profile coverage tracker.
304 SampleCoverageTracker CoverageTracker;
305
306 /// Profile reader object.
307 std::unique_ptr<SampleProfileReader> Reader;
308
309 /// Synthetic samples created by duplicating the samples of inlined functions
310 /// from the original profile as if they were top level sample profiles.
311 /// Use std::map because insertion may happen while its content is referenced.
312 std::map<SampleContext, FunctionSamples> OutlineFunctionSamples;
313
314 // A pseudo probe helper to correlate the imported sample counts.
315 std::unique_ptr<PseudoProbeManager> ProbeManager;
316
317 /// Samples collected for the body of this function.
318 FunctionSamples *Samples = nullptr;
319
320 /// Name of the profile file to load.
321 std::string Filename;
322
323 /// Name of the profile remapping file to load.
324 std::string RemappingFilename;
325
326 /// VirtualFileSystem to load profile files from.
327 IntrusiveRefCntPtr<vfs::FileSystem> FS;
328
329 /// Profile Summary Info computed from sample profile.
330 ProfileSummaryInfo *PSI = nullptr;
331
332 /// Optimization Remark Emitter used to emit diagnostic remarks.
333 OptRemarkEmitterT *ORE = nullptr;
334};
335
336/// Clear all the per-function data used to load samples and propagate weights.
337template <typename BT>
338void SampleProfileLoaderBaseImpl<BT>::clearFunctionData(bool ResetDT) {
339 BlockWeights.clear();
340 EdgeWeights.clear();
341 VisitedBlocks.clear();
342 VisitedEdges.clear();
343 EquivalenceClass.clear();
344 if (ResetDT) {
345 DT = nullptr;
346 PDT = nullptr;
347 LI = nullptr;
348 }
349 Predecessors.clear();
350 Successors.clear();
351 CoverageTracker.clear();
352}
353
354#ifndef NDEBUG
355/// Print the weight of edge \p E on stream \p OS.
356///
357/// \param OS Stream to emit the output to.
358/// \param E Edge to print.
359template <typename BT>
360void SampleProfileLoaderBaseImpl<BT>::printEdgeWeight(raw_ostream &OS, Edge E) {
361 OS << "weight[" << E.first->getName() << "->" << E.second->getName()
362 << "]: " << EdgeWeights[E] << "\n";
363}
364
365/// Print the equivalence class of block \p BB on stream \p OS.
366///
367/// \param OS Stream to emit the output to.
368/// \param BB Block to print.
369template <typename BT>
370void SampleProfileLoaderBaseImpl<BT>::printBlockEquivalence(
371 raw_ostream &OS, const BasicBlockT *BB) {
372 const BasicBlockT *Equiv = EquivalenceClass[BB];
373 OS << "equivalence[" << BB->getName()
374 << "]: " << ((Equiv) ? EquivalenceClass[BB]->getName() : "NONE") << "\n";
375}
376
377/// Print the weight of block \p BB on stream \p OS.
378///
379/// \param OS Stream to emit the output to.
380/// \param BB Block to print.
381template <typename BT>
382void SampleProfileLoaderBaseImpl<BT>::printBlockWeight(
383 raw_ostream &OS, const BasicBlockT *BB) const {
384 const auto &I = BlockWeights.find(BB);
385 uint64_t W = (I == BlockWeights.end() ? 0 : I->second);
386 OS << "weight[" << BB->getName() << "]: " << W << "\n";
387}
388#endif
389
390/// Get the weight for an instruction.
391///
392/// The "weight" of an instruction \p Inst is the number of samples
393/// collected on that instruction at runtime. To retrieve it, we
394/// need to compute the line number of \p Inst relative to the start of its
395/// function. We use HeaderLineno to compute the offset. We then
396/// look up the samples collected for \p Inst using BodySamples.
397///
398/// \param Inst Instruction to query.
399///
400/// \returns the weight of \p Inst.
401template <typename BT>
402ErrorOr<uint64_t>
403SampleProfileLoaderBaseImpl<BT>::getInstWeight(const InstructionT &Inst) {
404 if (FunctionSamples::ProfileIsProbeBased)
405 return getProbeWeight(Inst);
406 return getInstWeightImpl(Inst);
407}
408
409template <typename BT>
410ErrorOr<uint64_t>
411SampleProfileLoaderBaseImpl<BT>::getInstWeightImpl(const InstructionT &Inst) {
412 const FunctionSamples *FS = findFunctionSamples(I: Inst);
413 if (!FS)
414 return std::error_code();
415
416 const DebugLoc &DLoc = Inst.getDebugLoc();
417 if (!DLoc)
418 return std::error_code();
419
420 const DILocation *DIL = DLoc;
421 uint32_t LineOffset = FunctionSamples::getOffset(DIL);
422 uint32_t Discriminator;
423 if (EnableFSDiscriminator)
424 Discriminator = DIL->getDiscriminator();
425 else
426 Discriminator = DIL->getBaseDiscriminator();
427
428 ErrorOr<uint64_t> R = FS->findSamplesAt(LineOffset, Discriminator);
429 if (R) {
430 bool FirstMark =
431 CoverageTracker.markSamplesUsed(FS, LineOffset, Discriminator, Samples: R.get());
432 if (FirstMark) {
433 ORE->emit([&]() {
434 OptRemarkAnalysisT Remark(DEBUG_TYPE, "AppliedSamples", &Inst);
435 Remark << "Applied " << ore::NV("NumSamples", *R);
436 Remark << " samples from profile (offset: ";
437 Remark << ore::NV("LineOffset", LineOffset);
438 if (Discriminator) {
439 Remark << ".";
440 Remark << ore::NV("Discriminator", Discriminator);
441 }
442 Remark << ")";
443 return Remark;
444 });
445 }
446 LLVM_DEBUG(dbgs() << " " << DLoc.getLine() << "." << Discriminator << ":"
447 << Inst << " (line offset: " << LineOffset << "."
448 << Discriminator << " - weight: " << R.get() << ")\n");
449 }
450 return R;
451}
452
453template <typename BT>
454ErrorOr<uint64_t>
455SampleProfileLoaderBaseImpl<BT>::getProbeWeight(const InstructionT &Inst) {
456 assert(FunctionSamples::ProfileIsProbeBased &&
457 "Profile is not pseudo probe based");
458 std::optional<PseudoProbe> Probe = extractProbe(Inst);
459 // Ignore the non-probe instruction. If none of the instruction in the BB is
460 // probe, we choose to infer the BB's weight.
461 if (!Probe)
462 return std::error_code();
463
464 const FunctionSamples *FS = findFunctionSamples(I: Inst);
465 if (!FS) {
466 // If we can't find the function samples for a probe, it could be due to the
467 // probe is later optimized away or the inlining context is mismatced. We
468 // treat it as unknown, leaving it to profile inference instead of forcing a
469 // zero count.
470 return std::error_code();
471 }
472
473 auto R = FS->findSamplesAt(LineOffset: Probe->Id, Discriminator: Probe->Discriminator);
474 if (R) {
475 uint64_t Samples = R.get() * Probe->Factor;
476 bool FirstMark = CoverageTracker.markSamplesUsed(FS, LineOffset: Probe->Id, Discriminator: 0, Samples);
477 if (FirstMark) {
478 ORE->emit([&]() {
479 OptRemarkAnalysisT Remark(DEBUG_TYPE, "AppliedSamples", &Inst);
480 Remark << "Applied " << ore::NV("NumSamples", Samples);
481 Remark << " samples from profile (ProbeId=";
482 Remark << ore::NV("ProbeId", Probe->Id);
483 if (Probe->Discriminator) {
484 Remark << ".";
485 Remark << ore::NV("Discriminator", Probe->Discriminator);
486 }
487 Remark << ", Factor=";
488 Remark << ore::NV("Factor", Probe->Factor);
489 Remark << ", OriginalSamples=";
490 Remark << ore::NV("OriginalSamples", R.get());
491 Remark << ")";
492 return Remark;
493 });
494 }
495 LLVM_DEBUG({dbgs() << " " << Probe->Id;
496 if (Probe->Discriminator)
497 dbgs() << "." << Probe->Discriminator;
498 dbgs() << ":" << Inst << " - weight: " << R.get()
499 << " - factor: " << format("%0.2f", Probe->Factor) << ")\n";});
500 return Samples;
501 }
502 return R;
503}
504
505/// Compute the weight of a basic block.
506///
507/// The weight of basic block \p BB is the maximum weight of all the
508/// instructions in BB.
509///
510/// \param BB The basic block to query.
511///
512/// \returns the weight for \p BB.
513template <typename BT>
514ErrorOr<uint64_t>
515SampleProfileLoaderBaseImpl<BT>::getBlockWeight(const BasicBlockT *BB) {
516 uint64_t Max = 0;
517 bool HasWeight = false;
518 for (auto &I : *BB) {
519 const ErrorOr<uint64_t> &R = getInstWeight(Inst: I);
520 if (R) {
521 Max = std::max(a: Max, b: R.get());
522 HasWeight = true;
523 }
524 }
525 return HasWeight ? ErrorOr<uint64_t>(Max) : std::error_code();
526}
527
528/// Compute and store the weights of every basic block.
529///
530/// This populates the BlockWeights map by computing
531/// the weights of every basic block in the CFG.
532///
533/// \param F The function to query.
534template <typename BT>
535bool SampleProfileLoaderBaseImpl<BT>::computeBlockWeights(FunctionT &F) {
536 bool Changed = false;
537 LLVM_DEBUG(dbgs() << "Block weights\n");
538 for (const auto &BB : F) {
539 ErrorOr<uint64_t> Weight = getBlockWeight(BB: &BB);
540 if (Weight) {
541 BlockWeights[&BB] = Weight.get();
542 VisitedBlocks.insert(&BB);
543 Changed = true;
544 }
545 LLVM_DEBUG(printBlockWeight(dbgs(), &BB));
546 }
547
548 return Changed;
549}
550
551/// Get the FunctionSamples for an instruction.
552///
553/// The FunctionSamples of an instruction \p Inst is the inlined instance
554/// in which that instruction is coming from. We traverse the inline stack
555/// of that instruction, and match it with the tree nodes in the profile.
556///
557/// \param Inst Instruction to query.
558///
559/// \returns the FunctionSamples pointer to the inlined instance.
560template <typename BT>
561const FunctionSamples *SampleProfileLoaderBaseImpl<BT>::findFunctionSamples(
562 const InstructionT &Inst) const {
563 const DILocation *DIL = Inst.getDebugLoc();
564 if (!DIL)
565 return Samples;
566
567 auto it = DILocation2SampleMap.try_emplace(Key: DIL, Args: nullptr);
568 if (it.second) {
569 it.first->second = Samples->findFunctionSamples(DIL, Remapper: Reader->getRemapper());
570 }
571 return it.first->second;
572}
573
574/// Find equivalence classes for the given block.
575///
576/// This finds all the blocks that are guaranteed to execute the same
577/// number of times as \p BB1. To do this, it traverses all the
578/// descendants of \p BB1 in the dominator or post-dominator tree.
579///
580/// A block BB2 will be in the same equivalence class as \p BB1 if
581/// the following holds:
582///
583/// 1- \p BB1 is a descendant of BB2 in the opposite tree. So, if BB2
584/// is a descendant of \p BB1 in the dominator tree, then BB2 should
585/// dominate BB1 in the post-dominator tree.
586///
587/// 2- Both BB2 and \p BB1 must be in the same loop.
588///
589/// For every block BB2 that meets those two requirements, we set BB2's
590/// equivalence class to \p BB1.
591///
592/// \param BB1 Block to check.
593/// \param Descendants Descendants of \p BB1 in either the dom or pdom tree.
594/// \param DomTree Opposite dominator tree. If \p Descendants is filled
595/// with blocks from \p BB1's dominator tree, then
596/// this is the post-dominator tree, and vice versa.
597template <typename BT>
598void SampleProfileLoaderBaseImpl<BT>::findEquivalencesFor(
599 BasicBlockT *BB1, ArrayRef<BasicBlockT *> Descendants,
600 PostDominatorTreeT *DomTree) {
601 const BasicBlockT *EC = EquivalenceClass[BB1];
602 uint64_t Weight = BlockWeights[EC];
603 for (const auto *BB2 : Descendants) {
604 bool IsDomParent = DomTree->dominates(BB2, BB1);
605 bool IsInSameLoop = LI->getLoopFor(BB1) == LI->getLoopFor(BB2);
606 if (BB1 != BB2 && IsDomParent && IsInSameLoop) {
607 EquivalenceClass[BB2] = EC;
608 // If BB2 is visited, then the entire EC should be marked as visited.
609 if (VisitedBlocks.count(BB2)) {
610 VisitedBlocks.insert(EC);
611 }
612
613 // If BB2 is heavier than BB1, make BB2 have the same weight
614 // as BB1.
615 //
616 // Note that we don't worry about the opposite situation here
617 // (when BB2 is lighter than BB1). We will deal with this
618 // during the propagation phase. Right now, we just want to
619 // make sure that BB1 has the largest weight of all the
620 // members of its equivalence set.
621 Weight = std::max(Weight, BlockWeights[BB2]);
622 }
623 }
624 const BasicBlockT *EntryBB = getEntryBB(F: EC->getParent());
625 if (EC == EntryBB) {
626 BlockWeights[EC] = Samples->getHeadSamples() + 1;
627 } else {
628 BlockWeights[EC] = Weight;
629 }
630}
631
632/// Find equivalence classes.
633///
634/// Since samples may be missing from blocks, we can fill in the gaps by setting
635/// the weights of all the blocks in the same equivalence class to the same
636/// weight. To compute the concept of equivalence, we use dominance and loop
637/// information. Two blocks B1 and B2 are in the same equivalence class if B1
638/// dominates B2, B2 post-dominates B1 and both are in the same loop.
639///
640/// \param F The function to query.
641template <typename BT>
642void SampleProfileLoaderBaseImpl<BT>::findEquivalenceClasses(FunctionT &F) {
643 SmallVector<BasicBlockT *, 8> DominatedBBs;
644 LLVM_DEBUG(dbgs() << "\nBlock equivalence classes\n");
645 // Find equivalence sets based on dominance and post-dominance information.
646 for (auto &BB : F) {
647 BasicBlockT *BB1 = &BB;
648
649 // Compute BB1's equivalence class once.
650 // By default, blocks are in their own equivalence class.
651 auto [It, Inserted] = EquivalenceClass.try_emplace(BB1, BB1);
652 if (!Inserted) {
653 LLVM_DEBUG(printBlockEquivalence(dbgs(), BB1));
654 continue;
655 }
656
657 // Traverse all the blocks dominated by BB1. We are looking for
658 // every basic block BB2 such that:
659 //
660 // 1- BB1 dominates BB2.
661 // 2- BB2 post-dominates BB1.
662 // 3- BB1 and BB2 are in the same loop nest.
663 //
664 // If all those conditions hold, it means that BB2 is executed
665 // as many times as BB1, so they are placed in the same equivalence
666 // class by making BB2's equivalence class be BB1.
667 DominatedBBs.clear();
668 DT->getDescendants(BB1, DominatedBBs);
669 findEquivalencesFor(BB1, Descendants: DominatedBBs, DomTree: &*PDT);
670
671 LLVM_DEBUG(printBlockEquivalence(dbgs(), BB1));
672 }
673
674 // Assign weights to equivalence classes.
675 //
676 // All the basic blocks in the same equivalence class will execute
677 // the same number of times. Since we know that the head block in
678 // each equivalence class has the largest weight, assign that weight
679 // to all the blocks in that equivalence class.
680 LLVM_DEBUG(
681 dbgs() << "\nAssign the same weight to all blocks in the same class\n");
682 for (auto &BI : F) {
683 const BasicBlockT *BB = &BI;
684 const BasicBlockT *EquivBB = EquivalenceClass[BB];
685 if (BB != EquivBB)
686 BlockWeights[BB] = BlockWeights[EquivBB];
687 LLVM_DEBUG(printBlockWeight(dbgs(), BB));
688 }
689}
690
691/// Visit the given edge to decide if it has a valid weight.
692///
693/// If \p E has not been visited before, we copy to \p UnknownEdge
694/// and increment the count of unknown edges.
695///
696/// \param E Edge to visit.
697/// \param NumUnknownEdges Current number of unknown edges.
698/// \param UnknownEdge Set if E has not been visited before.
699///
700/// \returns E's weight, if known. Otherwise, return 0.
701template <typename BT>
702uint64_t SampleProfileLoaderBaseImpl<BT>::visitEdge(Edge E,
703 unsigned *NumUnknownEdges,
704 Edge *UnknownEdge) {
705 if (!VisitedEdges.count(E)) {
706 (*NumUnknownEdges)++;
707 *UnknownEdge = E;
708 return 0;
709 }
710
711 return EdgeWeights[E];
712}
713
714/// Propagate weights through incoming/outgoing edges.
715///
716/// If the weight of a basic block is known, and there is only one edge
717/// with an unknown weight, we can calculate the weight of that edge.
718///
719/// Similarly, if all the edges have a known count, we can calculate the
720/// count of the basic block, if needed.
721///
722/// \param F Function to process.
723/// \param UpdateBlockCount Whether we should update basic block counts that
724/// has already been annotated.
725///
726/// \returns True if new weights were assigned to edges or blocks.
727template <typename BT>
728bool SampleProfileLoaderBaseImpl<BT>::propagateThroughEdges(
729 FunctionT &F, bool UpdateBlockCount) {
730 bool Changed = false;
731 LLVM_DEBUG(dbgs() << "\nPropagation through edges\n");
732 for (const auto &BI : F) {
733 const BasicBlockT *BB = &BI;
734 const BasicBlockT *EC = EquivalenceClass[BB];
735
736 // Visit all the predecessor and successor edges to determine
737 // which ones have a weight assigned already. Note that it doesn't
738 // matter that we only keep track of a single unknown edge. The
739 // only case we are interested in handling is when only a single
740 // edge is unknown (see setEdgeOrBlockWeight).
741 for (unsigned i = 0; i < 2; i++) {
742 uint64_t TotalWeight = 0;
743 unsigned NumUnknownEdges = 0, NumTotalEdges = 0;
744 Edge UnknownEdge, SelfReferentialEdge, SingleEdge;
745
746 if (i == 0) {
747 // First, visit all predecessor edges.
748 auto &Preds = Predecessors[BB];
749 NumTotalEdges = Preds.size();
750 for (auto *Pred : Preds) {
751 Edge E = std::make_pair(Pred, BB);
752 TotalWeight += visitEdge(E, NumUnknownEdges: &NumUnknownEdges, UnknownEdge: &UnknownEdge);
753 if (E.first == E.second)
754 SelfReferentialEdge = E;
755 }
756 if (NumTotalEdges == 1) {
757 SingleEdge = std::make_pair(Predecessors[BB][0], BB);
758 }
759 } else {
760 // On the second round, visit all successor edges.
761 auto &Succs = Successors[BB];
762 NumTotalEdges = Succs.size();
763 for (auto *Succ : Succs) {
764 Edge E = std::make_pair(BB, Succ);
765 TotalWeight += visitEdge(E, NumUnknownEdges: &NumUnknownEdges, UnknownEdge: &UnknownEdge);
766 }
767 if (NumTotalEdges == 1) {
768 SingleEdge = std::make_pair(BB, Successors[BB][0]);
769 }
770 }
771
772 // After visiting all the edges, there are three cases that we
773 // can handle immediately:
774 //
775 // - All the edge weights are known (i.e., NumUnknownEdges == 0).
776 // In this case, we simply check that the sum of all the edges
777 // is the same as BB's weight. If not, we change BB's weight
778 // to match. Additionally, if BB had not been visited before,
779 // we mark it visited.
780 //
781 // - Only one edge is unknown and BB has already been visited.
782 // In this case, we can compute the weight of the edge by
783 // subtracting the total block weight from all the known
784 // edge weights. If the edges weight more than BB, then the
785 // edge of the last remaining edge is set to zero.
786 //
787 // - There exists a self-referential edge and the weight of BB is
788 // known. In this case, this edge can be based on BB's weight.
789 // We add up all the other known edges and set the weight on
790 // the self-referential edge as we did in the previous case.
791 //
792 // In any other case, we must continue iterating. Eventually,
793 // all edges will get a weight, or iteration will stop when
794 // it reaches SampleProfileMaxPropagateIterations.
795 if (NumUnknownEdges <= 1) {
796 uint64_t &BBWeight = BlockWeights[EC];
797 if (NumUnknownEdges == 0) {
798 if (!VisitedBlocks.count(EC)) {
799 // If we already know the weight of all edges, the weight of the
800 // basic block can be computed. It should be no larger than the sum
801 // of all edge weights.
802 if (TotalWeight > BBWeight) {
803 BBWeight = TotalWeight;
804 Changed = true;
805 LLVM_DEBUG(dbgs() << "All edge weights for " << BB->getName()
806 << " known. Set weight for block: ";
807 printBlockWeight(dbgs(), BB););
808 }
809 } else if (NumTotalEdges == 1 &&
810 EdgeWeights[SingleEdge] < BlockWeights[EC]) {
811 // If there is only one edge for the visited basic block, use the
812 // block weight to adjust edge weight if edge weight is smaller.
813 EdgeWeights[SingleEdge] = BlockWeights[EC];
814 Changed = true;
815 }
816 } else if (NumUnknownEdges == 1 && VisitedBlocks.count(EC)) {
817 // If there is a single unknown edge and the block has been
818 // visited, then we can compute E's weight.
819 if (BBWeight >= TotalWeight)
820 EdgeWeights[UnknownEdge] = BBWeight - TotalWeight;
821 else
822 EdgeWeights[UnknownEdge] = 0;
823 const BasicBlockT *OtherEC;
824 if (i == 0)
825 OtherEC = EquivalenceClass[UnknownEdge.first];
826 else
827 OtherEC = EquivalenceClass[UnknownEdge.second];
828 // Edge weights should never exceed the BB weights it connects.
829 if (VisitedBlocks.count(OtherEC) &&
830 EdgeWeights[UnknownEdge] > BlockWeights[OtherEC])
831 EdgeWeights[UnknownEdge] = BlockWeights[OtherEC];
832 VisitedEdges.insert(UnknownEdge);
833 Changed = true;
834 LLVM_DEBUG(dbgs() << "Set weight for edge: ";
835 printEdgeWeight(dbgs(), UnknownEdge));
836 }
837 } else if (VisitedBlocks.count(EC) && BlockWeights[EC] == 0) {
838 // If a block Weights 0, all its in/out edges should weight 0.
839 if (i == 0) {
840 for (auto *Pred : Predecessors[BB]) {
841 Edge E = std::make_pair(Pred, BB);
842 EdgeWeights[E] = 0;
843 VisitedEdges.insert(E);
844 }
845 } else {
846 for (auto *Succ : Successors[BB]) {
847 Edge E = std::make_pair(BB, Succ);
848 EdgeWeights[E] = 0;
849 VisitedEdges.insert(E);
850 }
851 }
852 } else if (SelfReferentialEdge.first && VisitedBlocks.count(EC)) {
853 uint64_t &BBWeight = BlockWeights[BB];
854 // We have a self-referential edge and the weight of BB is known.
855 if (BBWeight >= TotalWeight)
856 EdgeWeights[SelfReferentialEdge] = BBWeight - TotalWeight;
857 else
858 EdgeWeights[SelfReferentialEdge] = 0;
859 VisitedEdges.insert(SelfReferentialEdge);
860 Changed = true;
861 LLVM_DEBUG(dbgs() << "Set self-referential edge weight to: ";
862 printEdgeWeight(dbgs(), SelfReferentialEdge));
863 }
864 if (UpdateBlockCount && TotalWeight > 0 &&
865 VisitedBlocks.insert(EC).second) {
866 BlockWeights[EC] = TotalWeight;
867 Changed = true;
868 }
869 }
870 }
871
872 return Changed;
873}
874
875/// Build in/out edge lists for each basic block in the CFG.
876///
877/// We are interested in unique edges. If a block B1 has multiple
878/// edges to another block B2, we only add a single B1->B2 edge.
879template <typename BT>
880void SampleProfileLoaderBaseImpl<BT>::buildEdges(FunctionT &F) {
881 for (auto &BI : F) {
882 BasicBlockT *B1 = &BI;
883
884 // Add predecessors for B1.
885 SmallPtrSet<BasicBlockT *, 16> Visited;
886 auto &Preds = Predecessors[B1];
887 if (!Preds.empty())
888 llvm_unreachable("Found a stale predecessors list in a basic block.");
889 for (auto *B2 : getPredecessors(BB: B1))
890 if (Visited.insert(B2).second)
891 Preds.push_back(B2);
892
893 // Add successors for B1.
894 Visited.clear();
895 auto &Succs = Successors[B1];
896 if (!Succs.empty())
897 llvm_unreachable("Found a stale successors list in a basic block.");
898 for (auto *B2 : getSuccessors(BB: B1))
899 if (Visited.insert(B2).second)
900 Succs.push_back(B2);
901 }
902}
903
904/// Propagate weights into edges
905///
906/// The following rules are applied to every block BB in the CFG:
907///
908/// - If BB has a single predecessor/successor, then the weight
909/// of that edge is the weight of the block.
910///
911/// - If all incoming or outgoing edges are known except one, and the
912/// weight of the block is already known, the weight of the unknown
913/// edge will be the weight of the block minus the sum of all the known
914/// edges. If the sum of all the known edges is larger than BB's weight,
915/// we set the unknown edge weight to zero.
916///
917/// - If there is a self-referential edge, and the weight of the block is
918/// known, the weight for that edge is set to the weight of the block
919/// minus the weight of the other incoming edges to that block (if
920/// known).
921template <typename BT>
922void SampleProfileLoaderBaseImpl<BT>::propagateWeights(FunctionT &F) {
923 // Flow-based profile inference is only usable with BasicBlock instantiation
924 // of SampleProfileLoaderBaseImpl.
925 if (SampleProfileUseProfi) {
926 // Prepare block sample counts for inference.
927 BlockWeightMap SampleBlockWeights;
928 for (const auto &BI : F) {
929 ErrorOr<uint64_t> Weight = getBlockWeight(BB: &BI);
930 if (Weight)
931 SampleBlockWeights[&BI] = Weight.get();
932 }
933 // Fill in BlockWeights and EdgeWeights using an inference algorithm.
934 applyProfi(F, Successors, SampleBlockWeights, BlockWeights, EdgeWeights);
935 } else {
936 bool Changed = true;
937 unsigned I = 0;
938
939 // If BB weight is larger than its corresponding loop's header BB weight,
940 // use the BB weight to replace the loop header BB weight.
941 for (auto &BI : F) {
942 BasicBlockT *BB = &BI;
943 LoopT *L = LI->getLoopFor(BB);
944 if (!L) {
945 continue;
946 }
947 BasicBlockT *Header = L->getHeader();
948 if (Header && BlockWeights[BB] > BlockWeights[Header]) {
949 BlockWeights[Header] = BlockWeights[BB];
950 }
951 }
952
953 // Propagate until we converge or we go past the iteration limit.
954 while (Changed && I++ < SampleProfileMaxPropagateIterations) {
955 Changed = propagateThroughEdges(F, UpdateBlockCount: false);
956 }
957
958 // The first propagation propagates BB counts from annotated BBs to unknown
959 // BBs. The 2nd propagation pass resets edges weights, and use all BB
960 // weights to propagate edge weights.
961 VisitedEdges.clear();
962 Changed = true;
963 while (Changed && I++ < SampleProfileMaxPropagateIterations) {
964 Changed = propagateThroughEdges(F, UpdateBlockCount: false);
965 }
966
967 // The 3rd propagation pass allows adjust annotated BB weights that are
968 // obviously wrong.
969 Changed = true;
970 while (Changed && I++ < SampleProfileMaxPropagateIterations) {
971 Changed = propagateThroughEdges(F, UpdateBlockCount: true);
972 }
973 }
974}
975
976template <typename FT>
977void SampleProfileLoaderBaseImpl<FT>::applyProfi(
978 FunctionT &F, BlockEdgeMap &Successors, BlockWeightMap &SampleBlockWeights,
979 BlockWeightMap &BlockWeights, EdgeWeightMap &EdgeWeights) {
980 auto Infer = SampleProfileInference<FT>(F, Successors, SampleBlockWeights);
981 Infer.apply(BlockWeights, EdgeWeights);
982}
983
984/// Generate branch weight metadata for all branches in \p F.
985///
986/// Branch weights are computed out of instruction samples using a
987/// propagation heuristic. Propagation proceeds in 3 phases:
988///
989/// 1- Assignment of block weights. All the basic blocks in the function
990/// are initial assigned the same weight as their most frequently
991/// executed instruction.
992///
993/// 2- Creation of equivalence classes. Since samples may be missing from
994/// blocks, we can fill in the gaps by setting the weights of all the
995/// blocks in the same equivalence class to the same weight. To compute
996/// the concept of equivalence, we use dominance and loop information.
997/// Two blocks B1 and B2 are in the same equivalence class if B1
998/// dominates B2, B2 post-dominates B1 and both are in the same loop.
999///
1000/// 3- Propagation of block weights into edges. This uses a simple
1001/// propagation heuristic. The following rules are applied to every
1002/// block BB in the CFG:
1003///
1004/// - If BB has a single predecessor/successor, then the weight
1005/// of that edge is the weight of the block.
1006///
1007/// - If all the edges are known except one, and the weight of the
1008/// block is already known, the weight of the unknown edge will
1009/// be the weight of the block minus the sum of all the known
1010/// edges. If the sum of all the known edges is larger than BB's weight,
1011/// we set the unknown edge weight to zero.
1012///
1013/// - If there is a self-referential edge, and the weight of the block is
1014/// known, the weight for that edge is set to the weight of the block
1015/// minus the weight of the other incoming edges to that block (if
1016/// known).
1017///
1018/// Since this propagation is not guaranteed to finalize for every CFG, we
1019/// only allow it to proceed for a limited number of iterations (controlled
1020/// by -sample-profile-max-propagate-iterations).
1021///
1022/// FIXME: Try to replace this propagation heuristic with a scheme
1023/// that is guaranteed to finalize. A work-list approach similar to
1024/// the standard value propagation algorithm used by SSA-CCP might
1025/// work here.
1026///
1027/// \param F The function to query.
1028///
1029/// \returns true if \p F was modified. Returns false, otherwise.
1030template <typename BT>
1031bool SampleProfileLoaderBaseImpl<BT>::computeAndPropagateWeights(
1032 FunctionT &F, const DenseSet<GlobalValue::GUID> &InlinedGUIDs) {
1033 bool Changed = (InlinedGUIDs.size() != 0);
1034
1035 // Compute basic block weights.
1036 Changed |= computeBlockWeights(F);
1037
1038 if (Changed) {
1039 // Initialize propagation.
1040 initWeightPropagation(F, InlinedGUIDs);
1041
1042 // Propagate weights to all edges.
1043 propagateWeights(F);
1044
1045 // Post-process propagated weights.
1046 finalizeWeightPropagation(F, InlinedGUIDs);
1047 }
1048
1049 return Changed;
1050}
1051
1052template <typename BT>
1053void SampleProfileLoaderBaseImpl<BT>::initWeightPropagation(
1054 FunctionT &F, const DenseSet<GlobalValue::GUID> &InlinedGUIDs) {
1055 // Add an entry count to the function using the samples gathered at the
1056 // function entry.
1057 // Sets the GUIDs that are inlined in the profiled binary. This is used
1058 // for ThinLink to make correct liveness analysis, and also make the IR
1059 // match the profiled binary before annotation.
1060 getFunction(F).setEntryCount(
1061 ProfileCount(Samples->getHeadSamples() + 1, Function::PCT_Real),
1062 &InlinedGUIDs);
1063
1064 if (!SampleProfileUseProfi) {
1065 // Compute dominance and loop info needed for propagation.
1066 computeDominanceAndLoopInfo(F);
1067
1068 // Find equivalence classes.
1069 findEquivalenceClasses(F);
1070 }
1071
1072 // Before propagation starts, build, for each block, a list of
1073 // unique predecessors and successors. This is necessary to handle
1074 // identical edges in multiway branches. Since we visit all blocks and all
1075 // edges of the CFG, it is cleaner to build these lists once at the start
1076 // of the pass.
1077 buildEdges(F);
1078}
1079
1080template <typename BT>
1081void SampleProfileLoaderBaseImpl<BT>::finalizeWeightPropagation(
1082 FunctionT &F, const DenseSet<GlobalValue::GUID> &InlinedGUIDs) {
1083 // If we utilize a flow-based count inference, then we trust the computed
1084 // counts and set the entry count as computed by the algorithm. This is
1085 // primarily done to sync the counts produced by profi and BFI inference,
1086 // which uses the entry count for mass propagation.
1087 // If profi produces a zero-value for the entry count, we fallback to
1088 // Samples->getHeadSamples() + 1 to avoid functions with zero count.
1089 if (SampleProfileUseProfi) {
1090 const BasicBlockT *EntryBB = getEntryBB(F: &F);
1091 ErrorOr<uint64_t> EntryWeight = getBlockWeight(BB: EntryBB);
1092 if (BlockWeights[EntryBB] > 0) {
1093 getFunction(F).setEntryCount(
1094 ProfileCount(BlockWeights[EntryBB], Function::PCT_Real),
1095 &InlinedGUIDs);
1096 }
1097 }
1098}
1099
1100template <typename BT>
1101void SampleProfileLoaderBaseImpl<BT>::emitCoverageRemarks(FunctionT &F) {
1102 // If coverage checking was requested, compute it now.
1103 const Function &Func = getFunction(F);
1104 if (SampleProfileRecordCoverage) {
1105 unsigned Used = CoverageTracker.countUsedRecords(FS: Samples, PSI);
1106 unsigned Total = CoverageTracker.countBodyRecords(FS: Samples, PSI);
1107 unsigned Coverage = CoverageTracker.computeCoverage(Used, Total);
1108 if (Coverage < SampleProfileRecordCoverage) {
1109 Func.getContext().diagnose(DI: DiagnosticInfoSampleProfile(
1110 Func.getSubprogram()->getFilename(), getFunctionLoc(Func&: F),
1111 Twine(Used) + " of " + Twine(Total) + " available profile records (" +
1112 Twine(Coverage) + "%) were applied",
1113 DS_Warning));
1114 }
1115 }
1116
1117 if (SampleProfileSampleCoverage) {
1118 uint64_t Used = CoverageTracker.getTotalUsedSamples();
1119 uint64_t Total = CoverageTracker.countBodySamples(FS: Samples, PSI);
1120 unsigned Coverage = CoverageTracker.computeCoverage(Used, Total);
1121 if (Coverage < SampleProfileSampleCoverage) {
1122 Func.getContext().diagnose(DI: DiagnosticInfoSampleProfile(
1123 Func.getSubprogram()->getFilename(), getFunctionLoc(Func&: F),
1124 Twine(Used) + " of " + Twine(Total) + " available profile samples (" +
1125 Twine(Coverage) + "%) were applied",
1126 DS_Warning));
1127 }
1128 }
1129}
1130
1131/// Get the line number for the function header.
1132///
1133/// This looks up function \p F in the current compilation unit and
1134/// retrieves the line number where the function is defined. This is
1135/// line 0 for all the samples read from the profile file. Every line
1136/// number is relative to this line.
1137///
1138/// \param F Function object to query.
1139///
1140/// \returns the line number where \p F is defined. If it returns 0,
1141/// it means that there is no debug information available for \p F.
1142template <typename BT>
1143unsigned SampleProfileLoaderBaseImpl<BT>::getFunctionLoc(FunctionT &F) {
1144 const Function &Func = getFunction(F);
1145 if (DISubprogram *S = Func.getSubprogram())
1146 return S->getLine();
1147
1148 if (NoWarnSampleUnused)
1149 return 0;
1150
1151 // If the start of \p F is missing, emit a diagnostic to inform the user
1152 // about the missed opportunity.
1153 Func.getContext().diagnose(DI: DiagnosticInfoSampleProfile(
1154 "No debug information found in function " + Func.getName() +
1155 ": Function profile not used",
1156 DS_Warning));
1157 return 0;
1158}
1159
1160#undef DEBUG_TYPE
1161
1162} // namespace llvm
1163#endif // LLVM_TRANSFORMS_UTILS_SAMPLEPROFILELOADERBASEIMPL_H
1164