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

source code of llvm/include/llvm/Transforms/Utils/SampleProfileLoaderBaseImpl.h