Bug Summary

File:lib/Transforms/IPO/SampleProfile.cpp
Warning:line 727, column 10
Called C++ object pointer is null

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name SampleProfile.cpp -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=cplusplus -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -analyzer-config-compatibility-mode=true -mrelocation-model pic -pic-level 2 -mthread-model posix -fmath-errno -masm-verbose -mconstructor-aliases -munwind-tables -fuse-init-array -target-cpu x86-64 -dwarf-column-info -debugger-tuning=gdb -momit-leaf-frame-pointer -ffunction-sections -fdata-sections -resource-dir /usr/lib/llvm-8/lib/clang/8.0.0 -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-8~svn350071/build-llvm/lib/Transforms/IPO -I /build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO -I /build/llvm-toolchain-snapshot-8~svn350071/build-llvm/include -I /build/llvm-toolchain-snapshot-8~svn350071/include -U NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/x86_64-linux-gnu/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/x86_64-linux-gnu/c++/6.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/6.3.0/../../../../include/c++/6.3.0/backward -internal-isystem /usr/include/clang/8.0.0/include/ -internal-isystem /usr/local/include -internal-isystem /usr/lib/llvm-8/lib/clang/8.0.0/include -internal-externc-isystem /usr/include/x86_64-linux-gnu -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -Wno-unused-parameter -Wwrite-strings -Wno-missing-field-initializers -Wno-long-long -Wno-maybe-uninitialized -Wno-comment -std=c++11 -fdeprecated-macro -fdebug-compilation-dir /build/llvm-toolchain-snapshot-8~svn350071/build-llvm/lib/Transforms/IPO -fdebug-prefix-map=/build/llvm-toolchain-snapshot-8~svn350071=. -ferror-limit 19 -fmessage-length 0 -fvisibility-inlines-hidden -stack-protector 2 -fobjc-runtime=gcc -fdiagnostics-show-option -vectorize-loops -vectorize-slp -analyzer-output=html -analyzer-config stable-report-filename=true -o /tmp/scan-build-2018-12-27-042839-1215-1 -x c++ /build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO/SampleProfile.cpp -faddrsig
1//===- SampleProfile.cpp - Incorporate sample profiles into the IR --------===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file implements the SampleProfileLoader transformation. This pass
11// reads a profile file generated by a sampling profiler (e.g. Linux Perf -
12// http://perf.wiki.kernel.org/) and generates IR metadata to reflect the
13// profile information in the given profile.
14//
15// This pass generates branch weight annotations on the IR:
16//
17// - prof: Represents branch weights. This annotation is added to branches
18// to indicate the weights of each edge coming out of the branch.
19// The weight of each edge is the weight of the target block for
20// that edge. The weight of a block B is computed as the maximum
21// number of samples found in B.
22//
23//===----------------------------------------------------------------------===//
24
25#include "llvm/Transforms/IPO/SampleProfile.h"
26#include "llvm/ADT/ArrayRef.h"
27#include "llvm/ADT/DenseMap.h"
28#include "llvm/ADT/DenseSet.h"
29#include "llvm/ADT/None.h"
30#include "llvm/ADT/SmallPtrSet.h"
31#include "llvm/ADT/SmallSet.h"
32#include "llvm/ADT/SmallVector.h"
33#include "llvm/ADT/StringMap.h"
34#include "llvm/ADT/StringRef.h"
35#include "llvm/ADT/Twine.h"
36#include "llvm/Analysis/AssumptionCache.h"
37#include "llvm/Analysis/InlineCost.h"
38#include "llvm/Analysis/LoopInfo.h"
39#include "llvm/Analysis/OptimizationRemarkEmitter.h"
40#include "llvm/Analysis/PostDominators.h"
41#include "llvm/Analysis/ProfileSummaryInfo.h"
42#include "llvm/Analysis/TargetTransformInfo.h"
43#include "llvm/IR/BasicBlock.h"
44#include "llvm/IR/CFG.h"
45#include "llvm/IR/CallSite.h"
46#include "llvm/IR/DebugInfoMetadata.h"
47#include "llvm/IR/DebugLoc.h"
48#include "llvm/IR/DiagnosticInfo.h"
49#include "llvm/IR/Dominators.h"
50#include "llvm/IR/Function.h"
51#include "llvm/IR/GlobalValue.h"
52#include "llvm/IR/InstrTypes.h"
53#include "llvm/IR/Instruction.h"
54#include "llvm/IR/Instructions.h"
55#include "llvm/IR/IntrinsicInst.h"
56#include "llvm/IR/LLVMContext.h"
57#include "llvm/IR/MDBuilder.h"
58#include "llvm/IR/Module.h"
59#include "llvm/IR/PassManager.h"
60#include "llvm/IR/ValueSymbolTable.h"
61#include "llvm/Pass.h"
62#include "llvm/ProfileData/InstrProf.h"
63#include "llvm/ProfileData/SampleProf.h"
64#include "llvm/ProfileData/SampleProfReader.h"
65#include "llvm/Support/Casting.h"
66#include "llvm/Support/CommandLine.h"
67#include "llvm/Support/Debug.h"
68#include "llvm/Support/ErrorHandling.h"
69#include "llvm/Support/ErrorOr.h"
70#include "llvm/Support/GenericDomTree.h"
71#include "llvm/Support/raw_ostream.h"
72#include "llvm/Transforms/IPO.h"
73#include "llvm/Transforms/Instrumentation.h"
74#include "llvm/Transforms/Utils/CallPromotionUtils.h"
75#include "llvm/Transforms/Utils/Cloning.h"
76#include <algorithm>
77#include <cassert>
78#include <cstdint>
79#include <functional>
80#include <limits>
81#include <map>
82#include <memory>
83#include <string>
84#include <system_error>
85#include <utility>
86#include <vector>
87
88using namespace llvm;
89using namespace sampleprof;
90using ProfileCount = Function::ProfileCount;
91#define DEBUG_TYPE"sample-profile" "sample-profile"
92
93// Command line option to specify the file to read samples from. This is
94// mainly used for debugging.
95static cl::opt<std::string> SampleProfileFile(
96 "sample-profile-file", cl::init(""), cl::value_desc("filename"),
97 cl::desc("Profile file loaded by -sample-profile"), cl::Hidden);
98
99// The named file contains a set of transformations that may have been applied
100// to the symbol names between the program from which the sample data was
101// collected and the current program's symbols.
102static cl::opt<std::string> SampleProfileRemappingFile(
103 "sample-profile-remapping-file", cl::init(""), cl::value_desc("filename"),
104 cl::desc("Profile remapping file loaded by -sample-profile"), cl::Hidden);
105
106static cl::opt<unsigned> SampleProfileMaxPropagateIterations(
107 "sample-profile-max-propagate-iterations", cl::init(100),
108 cl::desc("Maximum number of iterations to go through when propagating "
109 "sample block/edge weights through the CFG."));
110
111static cl::opt<unsigned> SampleProfileRecordCoverage(
112 "sample-profile-check-record-coverage", cl::init(0), cl::value_desc("N"),
113 cl::desc("Emit a warning if less than N% of records in the input profile "
114 "are matched to the IR."));
115
116static cl::opt<unsigned> SampleProfileSampleCoverage(
117 "sample-profile-check-sample-coverage", cl::init(0), cl::value_desc("N"),
118 cl::desc("Emit a warning if less than N% of samples in the input profile "
119 "are matched to the IR."));
120
121static cl::opt<bool> NoWarnSampleUnused(
122 "no-warn-sample-unused", cl::init(false), cl::Hidden,
123 cl::desc("Use this option to turn off/on warnings about function with "
124 "samples but without debug information to use those samples. "));
125
126static cl::opt<bool> ProfileSampleAccurate(
127 "profile-sample-accurate", cl::Hidden, cl::init(false),
128 cl::desc("If the sample profile is accurate, we will mark all un-sampled "
129 "callsite and function as having 0 samples. Otherwise, treat "
130 "un-sampled callsites and functions conservatively as unknown. "));
131
132namespace {
133
134using BlockWeightMap = DenseMap<const BasicBlock *, uint64_t>;
135using EquivalenceClassMap = DenseMap<const BasicBlock *, const BasicBlock *>;
136using Edge = std::pair<const BasicBlock *, const BasicBlock *>;
137using EdgeWeightMap = DenseMap<Edge, uint64_t>;
138using BlockEdgeMap =
139 DenseMap<const BasicBlock *, SmallVector<const BasicBlock *, 8>>;
140
141class SampleCoverageTracker {
142public:
143 SampleCoverageTracker() = default;
144
145 bool markSamplesUsed(const FunctionSamples *FS, uint32_t LineOffset,
146 uint32_t Discriminator, uint64_t Samples);
147 unsigned computeCoverage(unsigned Used, unsigned Total) const;
148 unsigned countUsedRecords(const FunctionSamples *FS,
149 ProfileSummaryInfo *PSI) const;
150 unsigned countBodyRecords(const FunctionSamples *FS,
151 ProfileSummaryInfo *PSI) const;
152 uint64_t getTotalUsedSamples() const { return TotalUsedSamples; }
153 uint64_t countBodySamples(const FunctionSamples *FS,
154 ProfileSummaryInfo *PSI) const;
155
156 void clear() {
157 SampleCoverage.clear();
158 TotalUsedSamples = 0;
159 }
160
161private:
162 using BodySampleCoverageMap = std::map<LineLocation, unsigned>;
163 using FunctionSamplesCoverageMap =
164 DenseMap<const FunctionSamples *, BodySampleCoverageMap>;
165
166 /// Coverage map for sampling records.
167 ///
168 /// This map keeps a record of sampling records that have been matched to
169 /// an IR instruction. This is used to detect some form of staleness in
170 /// profiles (see flag -sample-profile-check-coverage).
171 ///
172 /// Each entry in the map corresponds to a FunctionSamples instance. This is
173 /// another map that counts how many times the sample record at the
174 /// given location has been used.
175 FunctionSamplesCoverageMap SampleCoverage;
176
177 /// Number of samples used from the profile.
178 ///
179 /// When a sampling record is used for the first time, the samples from
180 /// that record are added to this accumulator. Coverage is later computed
181 /// based on the total number of samples available in this function and
182 /// its callsites.
183 ///
184 /// Note that this accumulator tracks samples used from a single function
185 /// and all the inlined callsites. Strictly, we should have a map of counters
186 /// keyed by FunctionSamples pointers, but these stats are cleared after
187 /// every function, so we just need to keep a single counter.
188 uint64_t TotalUsedSamples = 0;
189};
190
191/// Sample profile pass.
192///
193/// This pass reads profile data from the file specified by
194/// -sample-profile-file and annotates every affected function with the
195/// profile information found in that file.
196class SampleProfileLoader {
197public:
198 SampleProfileLoader(
199 StringRef Name, StringRef RemapName, bool IsThinLTOPreLink,
200 std::function<AssumptionCache &(Function &)> GetAssumptionCache,
201 std::function<TargetTransformInfo &(Function &)> GetTargetTransformInfo)
202 : GetAC(std::move(GetAssumptionCache)),
203 GetTTI(std::move(GetTargetTransformInfo)), Filename(Name),
204 RemappingFilename(RemapName), IsThinLTOPreLink(IsThinLTOPreLink) {}
205
206 bool doInitialization(Module &M);
207 bool runOnModule(Module &M, ModuleAnalysisManager *AM,
208 ProfileSummaryInfo *_PSI);
209
210 void dump() { Reader->dump(); }
211
212protected:
213 bool runOnFunction(Function &F, ModuleAnalysisManager *AM);
214 unsigned getFunctionLoc(Function &F);
215 bool emitAnnotations(Function &F);
216 ErrorOr<uint64_t> getInstWeight(const Instruction &I);
217 ErrorOr<uint64_t> getBlockWeight(const BasicBlock *BB);
218 const FunctionSamples *findCalleeFunctionSamples(const Instruction &I) const;
219 std::vector<const FunctionSamples *>
220 findIndirectCallFunctionSamples(const Instruction &I, uint64_t &Sum) const;
221 const FunctionSamples *findFunctionSamples(const Instruction &I) const;
222 bool inlineCallInstruction(Instruction *I);
223 bool inlineHotFunctions(Function &F,
224 DenseSet<GlobalValue::GUID> &InlinedGUIDs);
225 void printEdgeWeight(raw_ostream &OS, Edge E);
226 void printBlockWeight(raw_ostream &OS, const BasicBlock *BB) const;
227 void printBlockEquivalence(raw_ostream &OS, const BasicBlock *BB);
228 bool computeBlockWeights(Function &F);
229 void findEquivalenceClasses(Function &F);
230 template <bool IsPostDom>
231 void findEquivalencesFor(BasicBlock *BB1, ArrayRef<BasicBlock *> Descendants,
232 DominatorTreeBase<BasicBlock, IsPostDom> *DomTree);
233
234 void propagateWeights(Function &F);
235 uint64_t visitEdge(Edge E, unsigned *NumUnknownEdges, Edge *UnknownEdge);
236 void buildEdges(Function &F);
237 bool propagateThroughEdges(Function &F, bool UpdateBlockCount);
238 void computeDominanceAndLoopInfo(Function &F);
239 void clearFunctionData();
240
241 /// Map basic blocks to their computed weights.
242 ///
243 /// The weight of a basic block is defined to be the maximum
244 /// of all the instruction weights in that block.
245 BlockWeightMap BlockWeights;
246
247 /// Map edges to their computed weights.
248 ///
249 /// Edge weights are computed by propagating basic block weights in
250 /// SampleProfile::propagateWeights.
251 EdgeWeightMap EdgeWeights;
252
253 /// Set of visited blocks during propagation.
254 SmallPtrSet<const BasicBlock *, 32> VisitedBlocks;
255
256 /// Set of visited edges during propagation.
257 SmallSet<Edge, 32> VisitedEdges;
258
259 /// Equivalence classes for block weights.
260 ///
261 /// Two blocks BB1 and BB2 are in the same equivalence class if they
262 /// dominate and post-dominate each other, and they are in the same loop
263 /// nest. When this happens, the two blocks are guaranteed to execute
264 /// the same number of times.
265 EquivalenceClassMap EquivalenceClass;
266
267 /// Map from function name to Function *. Used to find the function from
268 /// the function name. If the function name contains suffix, additional
269 /// entry is added to map from the stripped name to the function if there
270 /// is one-to-one mapping.
271 StringMap<Function *> SymbolMap;
272
273 /// Dominance, post-dominance and loop information.
274 std::unique_ptr<DominatorTree> DT;
275 std::unique_ptr<PostDominatorTree> PDT;
276 std::unique_ptr<LoopInfo> LI;
277
278 std::function<AssumptionCache &(Function &)> GetAC;
279 std::function<TargetTransformInfo &(Function &)> GetTTI;
280
281 /// Predecessors for each basic block in the CFG.
282 BlockEdgeMap Predecessors;
283
284 /// Successors for each basic block in the CFG.
285 BlockEdgeMap Successors;
286
287 SampleCoverageTracker CoverageTracker;
288
289 /// Profile reader object.
290 std::unique_ptr<SampleProfileReader> Reader;
291
292 /// Samples collected for the body of this function.
293 FunctionSamples *Samples = nullptr;
294
295 /// Name of the profile file to load.
296 std::string Filename;
297
298 /// Name of the profile remapping file to load.
299 std::string RemappingFilename;
300
301 /// Flag indicating whether the profile input loaded successfully.
302 bool ProfileIsValid = false;
303
304 /// Flag indicating if the pass is invoked in ThinLTO compile phase.
305 ///
306 /// In this phase, in annotation, we should not promote indirect calls.
307 /// Instead, we will mark GUIDs that needs to be annotated to the function.
308 bool IsThinLTOPreLink;
309
310 /// Profile Summary Info computed from sample profile.
311 ProfileSummaryInfo *PSI = nullptr;
312
313 /// Total number of samples collected in this profile.
314 ///
315 /// This is the sum of all the samples collected in all the functions executed
316 /// at runtime.
317 uint64_t TotalCollectedSamples = 0;
318
319 /// Optimization Remark Emitter used to emit diagnostic remarks.
320 OptimizationRemarkEmitter *ORE = nullptr;
321};
322
323class SampleProfileLoaderLegacyPass : public ModulePass {
324public:
325 // Class identification, replacement for typeinfo
326 static char ID;
327
328 SampleProfileLoaderLegacyPass(StringRef Name = SampleProfileFile,
329 bool IsThinLTOPreLink = false)
330 : ModulePass(ID),
331 SampleLoader(Name, SampleProfileRemappingFile, IsThinLTOPreLink,
332 [&](Function &F) -> AssumptionCache & {
333 return ACT->getAssumptionCache(F);
334 },
335 [&](Function &F) -> TargetTransformInfo & {
336 return TTIWP->getTTI(F);
337 }) {
338 initializeSampleProfileLoaderLegacyPassPass(
339 *PassRegistry::getPassRegistry());
340 }
341
342 void dump() { SampleLoader.dump(); }
343
344 bool doInitialization(Module &M) override {
345 return SampleLoader.doInitialization(M);
346 }
347
348 StringRef getPassName() const override { return "Sample profile pass"; }
349 bool runOnModule(Module &M) override;
350
351 void getAnalysisUsage(AnalysisUsage &AU) const override {
352 AU.addRequired<AssumptionCacheTracker>();
353 AU.addRequired<TargetTransformInfoWrapperPass>();
354 AU.addRequired<ProfileSummaryInfoWrapperPass>();
355 }
356
357private:
358 SampleProfileLoader SampleLoader;
359 AssumptionCacheTracker *ACT = nullptr;
360 TargetTransformInfoWrapperPass *TTIWP = nullptr;
361};
362
363} // end anonymous namespace
364
365/// Return true if the given callsite is hot wrt to hot cutoff threshold.
366///
367/// Functions that were inlined in the original binary will be represented
368/// in the inline stack in the sample profile. If the profile shows that
369/// the original inline decision was "good" (i.e., the callsite is executed
370/// frequently), then we will recreate the inline decision and apply the
371/// profile from the inlined callsite.
372///
373/// To decide whether an inlined callsite is hot, we compare the callsite
374/// sample count with the hot cutoff computed by ProfileSummaryInfo, it is
375/// regarded as hot if the count is above the cutoff value.
376static bool callsiteIsHot(const FunctionSamples *CallsiteFS,
377 ProfileSummaryInfo *PSI) {
378 if (!CallsiteFS)
379 return false; // The callsite was not inlined in the original binary.
380
381 assert(PSI && "PSI is expected to be non null")((PSI && "PSI is expected to be non null") ? static_cast
<void> (0) : __assert_fail ("PSI && \"PSI is expected to be non null\""
, "/build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO/SampleProfile.cpp"
, 381, __PRETTY_FUNCTION__));
382 uint64_t CallsiteTotalSamples = CallsiteFS->getTotalSamples();
383 return PSI->isHotCount(CallsiteTotalSamples);
384}
385
386/// Mark as used the sample record for the given function samples at
387/// (LineOffset, Discriminator).
388///
389/// \returns true if this is the first time we mark the given record.
390bool SampleCoverageTracker::markSamplesUsed(const FunctionSamples *FS,
391 uint32_t LineOffset,
392 uint32_t Discriminator,
393 uint64_t Samples) {
394 LineLocation Loc(LineOffset, Discriminator);
395 unsigned &Count = SampleCoverage[FS][Loc];
396 bool FirstTime = (++Count == 1);
397 if (FirstTime)
398 TotalUsedSamples += Samples;
399 return FirstTime;
400}
401
402/// Return the number of sample records that were applied from this profile.
403///
404/// This count does not include records from cold inlined callsites.
405unsigned
406SampleCoverageTracker::countUsedRecords(const FunctionSamples *FS,
407 ProfileSummaryInfo *PSI) const {
408 auto I = SampleCoverage.find(FS);
409
410 // The size of the coverage map for FS represents the number of records
411 // that were marked used at least once.
412 unsigned Count = (I != SampleCoverage.end()) ? I->second.size() : 0;
413
414 // If there are inlined callsites in this function, count the samples found
415 // in the respective bodies. However, do not bother counting callees with 0
416 // total samples, these are callees that were never invoked at runtime.
417 for (const auto &I : FS->getCallsiteSamples())
418 for (const auto &J : I.second) {
419 const FunctionSamples *CalleeSamples = &J.second;
420 if (callsiteIsHot(CalleeSamples, PSI))
421 Count += countUsedRecords(CalleeSamples, PSI);
422 }
423
424 return Count;
425}
426
427/// Return the number of sample records in the body of this profile.
428///
429/// This count does not include records from cold inlined callsites.
430unsigned
431SampleCoverageTracker::countBodyRecords(const FunctionSamples *FS,
432 ProfileSummaryInfo *PSI) const {
433 unsigned Count = FS->getBodySamples().size();
434
435 // Only count records in hot callsites.
436 for (const auto &I : FS->getCallsiteSamples())
437 for (const auto &J : I.second) {
438 const FunctionSamples *CalleeSamples = &J.second;
439 if (callsiteIsHot(CalleeSamples, PSI))
440 Count += countBodyRecords(CalleeSamples, PSI);
441 }
442
443 return Count;
444}
445
446/// Return the number of samples collected in the body of this profile.
447///
448/// This count does not include samples from cold inlined callsites.
449uint64_t
450SampleCoverageTracker::countBodySamples(const FunctionSamples *FS,
451 ProfileSummaryInfo *PSI) const {
452 uint64_t Total = 0;
453 for (const auto &I : FS->getBodySamples())
454 Total += I.second.getSamples();
455
456 // Only count samples in hot callsites.
457 for (const auto &I : FS->getCallsiteSamples())
458 for (const auto &J : I.second) {
459 const FunctionSamples *CalleeSamples = &J.second;
460 if (callsiteIsHot(CalleeSamples, PSI))
461 Total += countBodySamples(CalleeSamples, PSI);
462 }
463
464 return Total;
465}
466
467/// Return the fraction of sample records used in this profile.
468///
469/// The returned value is an unsigned integer in the range 0-100 indicating
470/// the percentage of sample records that were used while applying this
471/// profile to the associated function.
472unsigned SampleCoverageTracker::computeCoverage(unsigned Used,
473 unsigned Total) const {
474 assert(Used <= Total &&((Used <= Total && "number of used records cannot exceed the total number of records"
) ? static_cast<void> (0) : __assert_fail ("Used <= Total && \"number of used records cannot exceed the total number of records\""
, "/build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO/SampleProfile.cpp"
, 475, __PRETTY_FUNCTION__))
475 "number of used records cannot exceed the total number of records")((Used <= Total && "number of used records cannot exceed the total number of records"
) ? static_cast<void> (0) : __assert_fail ("Used <= Total && \"number of used records cannot exceed the total number of records\""
, "/build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO/SampleProfile.cpp"
, 475, __PRETTY_FUNCTION__));
476 return Total > 0 ? Used * 100 / Total : 100;
477}
478
479/// Clear all the per-function data used to load samples and propagate weights.
480void SampleProfileLoader::clearFunctionData() {
481 BlockWeights.clear();
482 EdgeWeights.clear();
483 VisitedBlocks.clear();
484 VisitedEdges.clear();
485 EquivalenceClass.clear();
486 DT = nullptr;
487 PDT = nullptr;
488 LI = nullptr;
489 Predecessors.clear();
490 Successors.clear();
491 CoverageTracker.clear();
492}
493
494#ifndef NDEBUG
495/// Print the weight of edge \p E on stream \p OS.
496///
497/// \param OS Stream to emit the output to.
498/// \param E Edge to print.
499void SampleProfileLoader::printEdgeWeight(raw_ostream &OS, Edge E) {
500 OS << "weight[" << E.first->getName() << "->" << E.second->getName()
501 << "]: " << EdgeWeights[E] << "\n";
502}
503
504/// Print the equivalence class of block \p BB on stream \p OS.
505///
506/// \param OS Stream to emit the output to.
507/// \param BB Block to print.
508void SampleProfileLoader::printBlockEquivalence(raw_ostream &OS,
509 const BasicBlock *BB) {
510 const BasicBlock *Equiv = EquivalenceClass[BB];
511 OS << "equivalence[" << BB->getName()
512 << "]: " << ((Equiv) ? EquivalenceClass[BB]->getName() : "NONE") << "\n";
513}
514
515/// Print the weight of block \p BB on stream \p OS.
516///
517/// \param OS Stream to emit the output to.
518/// \param BB Block to print.
519void SampleProfileLoader::printBlockWeight(raw_ostream &OS,
520 const BasicBlock *BB) const {
521 const auto &I = BlockWeights.find(BB);
522 uint64_t W = (I == BlockWeights.end() ? 0 : I->second);
523 OS << "weight[" << BB->getName() << "]: " << W << "\n";
524}
525#endif
526
527/// Get the weight for an instruction.
528///
529/// The "weight" of an instruction \p Inst is the number of samples
530/// collected on that instruction at runtime. To retrieve it, we
531/// need to compute the line number of \p Inst relative to the start of its
532/// function. We use HeaderLineno to compute the offset. We then
533/// look up the samples collected for \p Inst using BodySamples.
534///
535/// \param Inst Instruction to query.
536///
537/// \returns the weight of \p Inst.
538ErrorOr<uint64_t> SampleProfileLoader::getInstWeight(const Instruction &Inst) {
539 const DebugLoc &DLoc = Inst.getDebugLoc();
540 if (!DLoc)
2
←
Assuming the condition is false
→
3
←
Taking false branch
→
9
←
Assuming the condition is false
→
10
←
Taking false branch
→
541 return std::error_code();
542
543 const FunctionSamples *FS = findFunctionSamples(Inst);
11
←
Calling 'SampleProfileLoader::findFunctionSamples'
→
544 if (!FS)
4
←
Assuming 'FS' is null
→
5
←
Taking true branch
→
545 return std::error_code();
546
547 // Ignore all intrinsics and branch instructions.
548 // Branch instruction usually contains debug info from sources outside of
549 // the residing basic block, thus we ignore them during annotation.
550 if (isa<BranchInst>(Inst) || isa<IntrinsicInst>(Inst))
551 return std::error_code();
552
553 // If a direct call/invoke instruction is inlined in profile
554 // (findCalleeFunctionSamples returns non-empty result), but not inlined here,
555 // it means that the inlined callsite has no sample, thus the call
556 // instruction should have 0 count.
557 if ((isa<CallInst>(Inst) || isa<InvokeInst>(Inst)) &&
558 !ImmutableCallSite(&Inst).isIndirectCall() &&
559 findCalleeFunctionSamples(Inst))
560 return 0;
561
562 const DILocation *DIL = DLoc;
563 uint32_t LineOffset = FunctionSamples::getOffset(DIL);
564 uint32_t Discriminator = DIL->getBaseDiscriminator();
565 ErrorOr<uint64_t> R = FS->findSamplesAt(LineOffset, Discriminator);
566 if (R) {
567 bool FirstMark =
568 CoverageTracker.markSamplesUsed(FS, LineOffset, Discriminator, R.get());
569 if (FirstMark) {
570 ORE->emit([&]() {
571 OptimizationRemarkAnalysis Remark(DEBUG_TYPE"sample-profile", "AppliedSamples", &Inst);
572 Remark << "Applied " << ore::NV("NumSamples", *R);
573 Remark << " samples from profile (offset: ";
574 Remark << ore::NV("LineOffset", LineOffset);
575 if (Discriminator) {
576 Remark << ".";
577 Remark << ore::NV("Discriminator", Discriminator);
578 }
579 Remark << ")";
580 return Remark;
581 });
582 }
583 LLVM_DEBUG(dbgs() << " " << DLoc.getLine() << "."do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << " " << DLoc.getLine
() << "." << DIL->getBaseDiscriminator() <<
":" << Inst << " (line offset: " << LineOffset
<< "." << DIL->getBaseDiscriminator() <<
" - weight: " << R.get() << ")\n"; } } while (false
)
584 << DIL->getBaseDiscriminator() << ":" << Instdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << " " << DLoc.getLine
() << "." << DIL->getBaseDiscriminator() <<
":" << Inst << " (line offset: " << LineOffset
<< "." << DIL->getBaseDiscriminator() <<
" - weight: " << R.get() << ")\n"; } } while (false
)
585 << " (line offset: " << LineOffset << "."do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << " " << DLoc.getLine
() << "." << DIL->getBaseDiscriminator() <<
":" << Inst << " (line offset: " << LineOffset
<< "." << DIL->getBaseDiscriminator() <<
" - weight: " << R.get() << ")\n"; } } while (false
)
586 << DIL->getBaseDiscriminator() << " - weight: " << R.get()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << " " << DLoc.getLine
() << "." << DIL->getBaseDiscriminator() <<
":" << Inst << " (line offset: " << LineOffset
<< "." << DIL->getBaseDiscriminator() <<
" - weight: " << R.get() << ")\n"; } } while (false
)
587 << ")\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << " " << DLoc.getLine
() << "." << DIL->getBaseDiscriminator() <<
":" << Inst << " (line offset: " << LineOffset
<< "." << DIL->getBaseDiscriminator() <<
" - weight: " << R.get() << ")\n"; } } while (false
);
588 }
589 return R;
590}
591
592/// Compute the weight of a basic block.
593///
594/// The weight of basic block \p BB is the maximum weight of all the
595/// instructions in BB.
596///
597/// \param BB The basic block to query.
598///
599/// \returns the weight for \p BB.
600ErrorOr<uint64_t> SampleProfileLoader::getBlockWeight(const BasicBlock *BB) {
601 uint64_t Max = 0;
602 bool HasWeight = false;
603 for (auto &I : BB->getInstList()) {
604 const ErrorOr<uint64_t> &R = getInstWeight(I);
1
Calling 'SampleProfileLoader::getInstWeight'
→
6
←
Returning from 'SampleProfileLoader::getInstWeight'
→
8
←
Calling 'SampleProfileLoader::getInstWeight'
→
605 if (R) {
7
←
Taking false branch
→
606 Max = std::max(Max, R.get());
607 HasWeight = true;
608 }
609 }
610 return HasWeight ? ErrorOr<uint64_t>(Max) : std::error_code();
611}
612
613/// Compute and store the weights of every basic block.
614///
615/// This populates the BlockWeights map by computing
616/// the weights of every basic block in the CFG.
617///
618/// \param F The function to query.
619bool SampleProfileLoader::computeBlockWeights(Function &F) {
620 bool Changed = false;
621 LLVM_DEBUG(dbgs() << "Block weights\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "Block weights\n"; } } while
(false);
622 for (const auto &BB : F) {
623 ErrorOr<uint64_t> Weight = getBlockWeight(&BB);
624 if (Weight) {
625 BlockWeights[&BB] = Weight.get();
626 VisitedBlocks.insert(&BB);
627 Changed = true;
628 }
629 LLVM_DEBUG(printBlockWeight(dbgs(), &BB))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { printBlockWeight(dbgs(), &BB); } } while
(false);
630 }
631
632 return Changed;
633}
634
635/// Get the FunctionSamples for a call instruction.
636///
637/// The FunctionSamples of a call/invoke instruction \p Inst is the inlined
638/// instance in which that call instruction is calling to. It contains
639/// all samples that resides in the inlined instance. We first find the
640/// inlined instance in which the call instruction is from, then we
641/// traverse its children to find the callsite with the matching
642/// location.
643///
644/// \param Inst Call/Invoke instruction to query.
645///
646/// \returns The FunctionSamples pointer to the inlined instance.
647const FunctionSamples *
648SampleProfileLoader::findCalleeFunctionSamples(const Instruction &Inst) const {
649 const DILocation *DIL = Inst.getDebugLoc();
650 if (!DIL) {
651 return nullptr;
652 }
653
654 StringRef CalleeName;
655 if (const CallInst *CI = dyn_cast<CallInst>(&Inst))
656 if (Function *Callee = CI->getCalledFunction())
657 CalleeName = Callee->getName();
658
659 const FunctionSamples *FS = findFunctionSamples(Inst);
660 if (FS == nullptr)
661 return nullptr;
662
663 return FS->findFunctionSamplesAt(LineLocation(FunctionSamples::getOffset(DIL),
664 DIL->getBaseDiscriminator()),
665 CalleeName);
666}
667
668/// Returns a vector of FunctionSamples that are the indirect call targets
669/// of \p Inst. The vector is sorted by the total number of samples. Stores
670/// the total call count of the indirect call in \p Sum.
671std::vector<const FunctionSamples *>
672SampleProfileLoader::findIndirectCallFunctionSamples(
673 const Instruction &Inst, uint64_t &Sum) const {
674 const DILocation *DIL = Inst.getDebugLoc();
675 std::vector<const FunctionSamples *> R;
676
677 if (!DIL) {
678 return R;
679 }
680
681 const FunctionSamples *FS = findFunctionSamples(Inst);
682 if (FS == nullptr)
683 return R;
684
685 uint32_t LineOffset = FunctionSamples::getOffset(DIL);
686 uint32_t Discriminator = DIL->getBaseDiscriminator();
687
688 auto T = FS->findCallTargetMapAt(LineOffset, Discriminator);
689 Sum = 0;
690 if (T)
691 for (const auto &T_C : T.get())
692 Sum += T_C.second;
693 if (const FunctionSamplesMap *M = FS->findFunctionSamplesMapAt(LineLocation(
694 FunctionSamples::getOffset(DIL), DIL->getBaseDiscriminator()))) {
695 if (M->empty())
696 return R;
697 for (const auto &NameFS : *M) {
698 Sum += NameFS.second.getEntrySamples();
699 R.push_back(&NameFS.second);
700 }
701 llvm::sort(R, [](const FunctionSamples *L, const FunctionSamples *R) {
702 if (L->getEntrySamples() != R->getEntrySamples())
703 return L->getEntrySamples() > R->getEntrySamples();
704 return FunctionSamples::getGUID(L->getName()) <
705 FunctionSamples::getGUID(R->getName());
706 });
707 }
708 return R;
709}
710
711/// Get the FunctionSamples for an instruction.
712///
713/// The FunctionSamples of an instruction \p Inst is the inlined instance
714/// in which that instruction is coming from. We traverse the inline stack
715/// of that instruction, and match it with the tree nodes in the profile.
716///
717/// \param Inst Instruction to query.
718///
719/// \returns the FunctionSamples pointer to the inlined instance.
720const FunctionSamples *
721SampleProfileLoader::findFunctionSamples(const Instruction &Inst) const {
722 SmallVector<std::pair<LineLocation, StringRef>, 10> S;
723 const DILocation *DIL = Inst.getDebugLoc();
724 if (!DIL)
12
←
Assuming 'DIL' is non-null
→
13
←
Taking false branch
→
725 return Samples;
726
727 return Samples->findFunctionSamples(DIL);
14
←
Called C++ object pointer is null
728}
729
730bool SampleProfileLoader::inlineCallInstruction(Instruction *I) {
731 assert(isa<CallInst>(I) || isa<InvokeInst>(I))((isa<CallInst>(I) || isa<InvokeInst>(I)) ? static_cast
<void> (0) : __assert_fail ("isa<CallInst>(I) || isa<InvokeInst>(I)"
, "/build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO/SampleProfile.cpp"
, 731, __PRETTY_FUNCTION__));
732 CallSite CS(I);
733 Function *CalledFunction = CS.getCalledFunction();
734 assert(CalledFunction)((CalledFunction) ? static_cast<void> (0) : __assert_fail
("CalledFunction", "/build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO/SampleProfile.cpp"
, 734, __PRETTY_FUNCTION__));
735 DebugLoc DLoc = I->getDebugLoc();
736 BasicBlock *BB = I->getParent();
737 InlineParams Params = getInlineParams();
738 Params.ComputeFullInlineCost = true;
739 // Checks if there is anything in the reachable portion of the callee at
740 // this callsite that makes this inlining potentially illegal. Need to
741 // set ComputeFullInlineCost, otherwise getInlineCost may return early
742 // when cost exceeds threshold without checking all IRs in the callee.
743 // The acutal cost does not matter because we only checks isNever() to
744 // see if it is legal to inline the callsite.
745 InlineCost Cost = getInlineCost(CS, Params, GetTTI(*CalledFunction), GetAC,
746 None, nullptr, nullptr);
747 if (Cost.isNever()) {
748 ORE->emit(OptimizationRemark(DEBUG_TYPE"sample-profile", "Not inline", DLoc, BB)
749 << "incompatible inlining");
750 return false;
751 }
752 InlineFunctionInfo IFI(nullptr, &GetAC);
753 if (InlineFunction(CS, IFI)) {
754 // The call to InlineFunction erases I, so we can't pass it here.
755 ORE->emit(OptimizationRemark(DEBUG_TYPE"sample-profile", "HotInline", DLoc, BB)
756 << "inlined hot callee '" << ore::NV("Callee", CalledFunction)
757 << "' into '" << ore::NV("Caller", BB->getParent()) << "'");
758 return true;
759 }
760 return false;
761}
762
763/// Iteratively inline hot callsites of a function.
764///
765/// Iteratively traverse all callsites of the function \p F, and find if
766/// the corresponding inlined instance exists and is hot in profile. If
767/// it is hot enough, inline the callsites and adds new callsites of the
768/// callee into the caller. If the call is an indirect call, first promote
769/// it to direct call. Each indirect call is limited with a single target.
770///
771/// \param F function to perform iterative inlining.
772/// \param InlinedGUIDs a set to be updated to include all GUIDs that are
773/// inlined in the profiled binary.
774///
775/// \returns True if there is any inline happened.
776bool SampleProfileLoader::inlineHotFunctions(
777 Function &F, DenseSet<GlobalValue::GUID> &InlinedGUIDs) {
778 DenseSet<Instruction *> PromotedInsns;
779 bool Changed = false;
780 while (true) {
781 bool LocalChanged = false;
782 SmallVector<Instruction *, 10> CIS;
783 for (auto &BB : F) {
784 bool Hot = false;
785 SmallVector<Instruction *, 10> Candidates;
786 for (auto &I : BB.getInstList()) {
787 const FunctionSamples *FS = nullptr;
788 if ((isa<CallInst>(I) || isa<InvokeInst>(I)) &&
789 !isa<IntrinsicInst>(I) && (FS = findCalleeFunctionSamples(I))) {
790 Candidates.push_back(&I);
791 if (callsiteIsHot(FS, PSI))
792 Hot = true;
793 }
794 }
795 if (Hot) {
796 CIS.insert(CIS.begin(), Candidates.begin(), Candidates.end());
797 }
798 }
799 for (auto I : CIS) {
800 Function *CalledFunction = CallSite(I).getCalledFunction();
801 // Do not inline recursive calls.
802 if (CalledFunction == &F)
803 continue;
804 if (CallSite(I).isIndirectCall()) {
805 if (PromotedInsns.count(I))
806 continue;
807 uint64_t Sum;
808 for (const auto *FS : findIndirectCallFunctionSamples(*I, Sum)) {
809 if (IsThinLTOPreLink) {
810 FS->findInlinedFunctions(InlinedGUIDs, F.getParent(),
811 PSI->getOrCompHotCountThreshold());
812 continue;
813 }
814 auto CalleeFunctionName = FS->getFuncNameInModule(F.getParent());
815 // If it is a recursive call, we do not inline it as it could bloat
816 // the code exponentially. There is way to better handle this, e.g.
817 // clone the caller first, and inline the cloned caller if it is
818 // recursive. As llvm does not inline recursive calls, we will
819 // simply ignore it instead of handling it explicitly.
820 if (CalleeFunctionName == F.getName())
821 continue;
822
823 const char *Reason = "Callee function not available";
824 auto R = SymbolMap.find(CalleeFunctionName);
825 if (R != SymbolMap.end() && R->getValue() &&
826 !R->getValue()->isDeclaration() &&
827 R->getValue()->getSubprogram() &&
828 isLegalToPromote(CallSite(I), R->getValue(), &Reason)) {
829 uint64_t C = FS->getEntrySamples();
830 Instruction *DI =
831 pgo::promoteIndirectCall(I, R->getValue(), C, Sum, false, ORE);
832 Sum -= C;
833 PromotedInsns.insert(I);
834 // If profile mismatches, we should not attempt to inline DI.
835 if ((isa<CallInst>(DI) || isa<InvokeInst>(DI)) &&
836 inlineCallInstruction(DI))
837 LocalChanged = true;
838 } else {
839 LLVM_DEBUG(dbgs()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nFailed to promote indirect call to "
<< CalleeFunctionName << " because " << Reason
<< "\n"; } } while (false)
840 << "\nFailed to promote indirect call to "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nFailed to promote indirect call to "
<< CalleeFunctionName << " because " << Reason
<< "\n"; } } while (false)
841 << CalleeFunctionName << " because " << Reason << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nFailed to promote indirect call to "
<< CalleeFunctionName << " because " << Reason
<< "\n"; } } while (false);
842 }
843 }
844 } else if (CalledFunction && CalledFunction->getSubprogram() &&
845 !CalledFunction->isDeclaration()) {
846 if (inlineCallInstruction(I))
847 LocalChanged = true;
848 } else if (IsThinLTOPreLink) {
849 findCalleeFunctionSamples(*I)->findInlinedFunctions(
850 InlinedGUIDs, F.getParent(), PSI->getOrCompHotCountThreshold());
851 }
852 }
853 if (LocalChanged) {
854 Changed = true;
855 } else {
856 break;
857 }
858 }
859 return Changed;
860}
861
862/// Find equivalence classes for the given block.
863///
864/// This finds all the blocks that are guaranteed to execute the same
865/// number of times as \p BB1. To do this, it traverses all the
866/// descendants of \p BB1 in the dominator or post-dominator tree.
867///
868/// A block BB2 will be in the same equivalence class as \p BB1 if
869/// the following holds:
870///
871/// 1- \p BB1 is a descendant of BB2 in the opposite tree. So, if BB2
872/// is a descendant of \p BB1 in the dominator tree, then BB2 should
873/// dominate BB1 in the post-dominator tree.
874///
875/// 2- Both BB2 and \p BB1 must be in the same loop.
876///
877/// For every block BB2 that meets those two requirements, we set BB2's
878/// equivalence class to \p BB1.
879///
880/// \param BB1 Block to check.
881/// \param Descendants Descendants of \p BB1 in either the dom or pdom tree.
882/// \param DomTree Opposite dominator tree. If \p Descendants is filled
883/// with blocks from \p BB1's dominator tree, then
884/// this is the post-dominator tree, and vice versa.
885template <bool IsPostDom>
886void SampleProfileLoader::findEquivalencesFor(
887 BasicBlock *BB1, ArrayRef<BasicBlock *> Descendants,
888 DominatorTreeBase<BasicBlock, IsPostDom> *DomTree) {
889 const BasicBlock *EC = EquivalenceClass[BB1];
890 uint64_t Weight = BlockWeights[EC];
891 for (const auto *BB2 : Descendants) {
892 bool IsDomParent = DomTree->dominates(BB2, BB1);
893 bool IsInSameLoop = LI->getLoopFor(BB1) == LI->getLoopFor(BB2);
894 if (BB1 != BB2 && IsDomParent && IsInSameLoop) {
895 EquivalenceClass[BB2] = EC;
896 // If BB2 is visited, then the entire EC should be marked as visited.
897 if (VisitedBlocks.count(BB2)) {
898 VisitedBlocks.insert(EC);
899 }
900
901 // If BB2 is heavier than BB1, make BB2 have the same weight
902 // as BB1.
903 //
904 // Note that we don't worry about the opposite situation here
905 // (when BB2 is lighter than BB1). We will deal with this
906 // during the propagation phase. Right now, we just want to
907 // make sure that BB1 has the largest weight of all the
908 // members of its equivalence set.
909 Weight = std::max(Weight, BlockWeights[BB2]);
910 }
911 }
912 if (EC == &EC->getParent()->getEntryBlock()) {
913 BlockWeights[EC] = Samples->getHeadSamples() + 1;
914 } else {
915 BlockWeights[EC] = Weight;
916 }
917}
918
919/// Find equivalence classes.
920///
921/// Since samples may be missing from blocks, we can fill in the gaps by setting
922/// the weights of all the blocks in the same equivalence class to the same
923/// weight. To compute the concept of equivalence, we use dominance and loop
924/// information. Two blocks B1 and B2 are in the same equivalence class if B1
925/// dominates B2, B2 post-dominates B1 and both are in the same loop.
926///
927/// \param F The function to query.
928void SampleProfileLoader::findEquivalenceClasses(Function &F) {
929 SmallVector<BasicBlock *, 8> DominatedBBs;
930 LLVM_DEBUG(dbgs() << "\nBlock equivalence classes\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nBlock equivalence classes\n"
; } } while (false);
931 // Find equivalence sets based on dominance and post-dominance information.
932 for (auto &BB : F) {
933 BasicBlock *BB1 = &BB;
934
935 // Compute BB1's equivalence class once.
936 if (EquivalenceClass.count(BB1)) {
937 LLVM_DEBUG(printBlockEquivalence(dbgs(), BB1))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { printBlockEquivalence(dbgs(), BB1); } }
while (false);
938 continue;
939 }
940
941 // By default, blocks are in their own equivalence class.
942 EquivalenceClass[BB1] = BB1;
943
944 // Traverse all the blocks dominated by BB1. We are looking for
945 // every basic block BB2 such that:
946 //
947 // 1- BB1 dominates BB2.
948 // 2- BB2 post-dominates BB1.
949 // 3- BB1 and BB2 are in the same loop nest.
950 //
951 // If all those conditions hold, it means that BB2 is executed
952 // as many times as BB1, so they are placed in the same equivalence
953 // class by making BB2's equivalence class be BB1.
954 DominatedBBs.clear();
955 DT->getDescendants(BB1, DominatedBBs);
956 findEquivalencesFor(BB1, DominatedBBs, PDT.get());
957
958 LLVM_DEBUG(printBlockEquivalence(dbgs(), BB1))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { printBlockEquivalence(dbgs(), BB1); } }
while (false);
959 }
960
961 // Assign weights to equivalence classes.
962 //
963 // All the basic blocks in the same equivalence class will execute
964 // the same number of times. Since we know that the head block in
965 // each equivalence class has the largest weight, assign that weight
966 // to all the blocks in that equivalence class.
967 LLVM_DEBUG(do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nAssign the same weight to all blocks in the same class\n"
; } } while (false)
968 dbgs() << "\nAssign the same weight to all blocks in the same class\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nAssign the same weight to all blocks in the same class\n"
; } } while (false);
969 for (auto &BI : F) {
970 const BasicBlock *BB = &BI;
971 const BasicBlock *EquivBB = EquivalenceClass[BB];
972 if (BB != EquivBB)
973 BlockWeights[BB] = BlockWeights[EquivBB];
974 LLVM_DEBUG(printBlockWeight(dbgs(), BB))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { printBlockWeight(dbgs(), BB); } } while
(false);
975 }
976}
977
978/// Visit the given edge to decide if it has a valid weight.
979///
980/// If \p E has not been visited before, we copy to \p UnknownEdge
981/// and increment the count of unknown edges.
982///
983/// \param E Edge to visit.
984/// \param NumUnknownEdges Current number of unknown edges.
985/// \param UnknownEdge Set if E has not been visited before.
986///
987/// \returns E's weight, if known. Otherwise, return 0.
988uint64_t SampleProfileLoader::visitEdge(Edge E, unsigned *NumUnknownEdges,
989 Edge *UnknownEdge) {
990 if (!VisitedEdges.count(E)) {
991 (*NumUnknownEdges)++;
992 *UnknownEdge = E;
993 return 0;
994 }
995
996 return EdgeWeights[E];
997}
998
999/// Propagate weights through incoming/outgoing edges.
1000///
1001/// If the weight of a basic block is known, and there is only one edge
1002/// with an unknown weight, we can calculate the weight of that edge.
1003///
1004/// Similarly, if all the edges have a known count, we can calculate the
1005/// count of the basic block, if needed.
1006///
1007/// \param F Function to process.
1008/// \param UpdateBlockCount Whether we should update basic block counts that
1009/// has already been annotated.
1010///
1011/// \returns True if new weights were assigned to edges or blocks.
1012bool SampleProfileLoader::propagateThroughEdges(Function &F,
1013 bool UpdateBlockCount) {
1014 bool Changed = false;
1015 LLVM_DEBUG(dbgs() << "\nPropagation through edges\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nPropagation through edges\n"
; } } while (false);
1016 for (const auto &BI : F) {
1017 const BasicBlock *BB = &BI;
1018 const BasicBlock *EC = EquivalenceClass[BB];
1019
1020 // Visit all the predecessor and successor edges to determine
1021 // which ones have a weight assigned already. Note that it doesn't
1022 // matter that we only keep track of a single unknown edge. The
1023 // only case we are interested in handling is when only a single
1024 // edge is unknown (see setEdgeOrBlockWeight).
1025 for (unsigned i = 0; i < 2; i++) {
1026 uint64_t TotalWeight = 0;
1027 unsigned NumUnknownEdges = 0, NumTotalEdges = 0;
1028 Edge UnknownEdge, SelfReferentialEdge, SingleEdge;
1029
1030 if (i == 0) {
1031 // First, visit all predecessor edges.
1032 NumTotalEdges = Predecessors[BB].size();
1033 for (auto *Pred : Predecessors[BB]) {
1034 Edge E = std::make_pair(Pred, BB);
1035 TotalWeight += visitEdge(E, &NumUnknownEdges, &UnknownEdge);
1036 if (E.first == E.second)
1037 SelfReferentialEdge = E;
1038 }
1039 if (NumTotalEdges == 1) {
1040 SingleEdge = std::make_pair(Predecessors[BB][0], BB);
1041 }
1042 } else {
1043 // On the second round, visit all successor edges.
1044 NumTotalEdges = Successors[BB].size();
1045 for (auto *Succ : Successors[BB]) {
1046 Edge E = std::make_pair(BB, Succ);
1047 TotalWeight += visitEdge(E, &NumUnknownEdges, &UnknownEdge);
1048 }
1049 if (NumTotalEdges == 1) {
1050 SingleEdge = std::make_pair(BB, Successors[BB][0]);
1051 }
1052 }
1053
1054 // After visiting all the edges, there are three cases that we
1055 // can handle immediately:
1056 //
1057 // - All the edge weights are known (i.e., NumUnknownEdges == 0).
1058 // In this case, we simply check that the sum of all the edges
1059 // is the same as BB's weight. If not, we change BB's weight
1060 // to match. Additionally, if BB had not been visited before,
1061 // we mark it visited.
1062 //
1063 // - Only one edge is unknown and BB has already been visited.
1064 // In this case, we can compute the weight of the edge by
1065 // subtracting the total block weight from all the known
1066 // edge weights. If the edges weight more than BB, then the
1067 // edge of the last remaining edge is set to zero.
1068 //
1069 // - There exists a self-referential edge and the weight of BB is
1070 // known. In this case, this edge can be based on BB's weight.
1071 // We add up all the other known edges and set the weight on
1072 // the self-referential edge as we did in the previous case.
1073 //
1074 // In any other case, we must continue iterating. Eventually,
1075 // all edges will get a weight, or iteration will stop when
1076 // it reaches SampleProfileMaxPropagateIterations.
1077 if (NumUnknownEdges <= 1) {
1078 uint64_t &BBWeight = BlockWeights[EC];
1079 if (NumUnknownEdges == 0) {
1080 if (!VisitedBlocks.count(EC)) {
1081 // If we already know the weight of all edges, the weight of the
1082 // basic block can be computed. It should be no larger than the sum
1083 // of all edge weights.
1084 if (TotalWeight > BBWeight) {
1085 BBWeight = TotalWeight;
1086 Changed = true;
1087 LLVM_DEBUG(dbgs() << "All edge weights for " << BB->getName()do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "All edge weights for "
<< BB->getName() << " known. Set weight for block: "
; printBlockWeight(dbgs(), BB);; } } while (false)
1088 << " known. Set weight for block: ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "All edge weights for "
<< BB->getName() << " known. Set weight for block: "
; printBlockWeight(dbgs(), BB);; } } while (false)
1089 printBlockWeight(dbgs(), BB);)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "All edge weights for "
<< BB->getName() << " known. Set weight for block: "
; printBlockWeight(dbgs(), BB);; } } while (false);
1090 }
1091 } else if (NumTotalEdges == 1 &&
1092 EdgeWeights[SingleEdge] < BlockWeights[EC]) {
1093 // If there is only one edge for the visited basic block, use the
1094 // block weight to adjust edge weight if edge weight is smaller.
1095 EdgeWeights[SingleEdge] = BlockWeights[EC];
1096 Changed = true;
1097 }
1098 } else if (NumUnknownEdges == 1 && VisitedBlocks.count(EC)) {
1099 // If there is a single unknown edge and the block has been
1100 // visited, then we can compute E's weight.
1101 if (BBWeight >= TotalWeight)
1102 EdgeWeights[UnknownEdge] = BBWeight - TotalWeight;
1103 else
1104 EdgeWeights[UnknownEdge] = 0;
1105 const BasicBlock *OtherEC;
1106 if (i == 0)
1107 OtherEC = EquivalenceClass[UnknownEdge.first];
1108 else
1109 OtherEC = EquivalenceClass[UnknownEdge.second];
1110 // Edge weights should never exceed the BB weights it connects.
1111 if (VisitedBlocks.count(OtherEC) &&
1112 EdgeWeights[UnknownEdge] > BlockWeights[OtherEC])
1113 EdgeWeights[UnknownEdge] = BlockWeights[OtherEC];
1114 VisitedEdges.insert(UnknownEdge);
1115 Changed = true;
1116 LLVM_DEBUG(dbgs() << "Set weight for edge: ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "Set weight for edge: "
; printEdgeWeight(dbgs(), UnknownEdge); } } while (false)
1117 printEdgeWeight(dbgs(), UnknownEdge))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "Set weight for edge: "
; printEdgeWeight(dbgs(), UnknownEdge); } } while (false);
1118 }
1119 } else if (VisitedBlocks.count(EC) && BlockWeights[EC] == 0) {
1120 // If a block Weights 0, all its in/out edges should weight 0.
1121 if (i == 0) {
1122 for (auto *Pred : Predecessors[BB]) {
1123 Edge E = std::make_pair(Pred, BB);
1124 EdgeWeights[E] = 0;
1125 VisitedEdges.insert(E);
1126 }
1127 } else {
1128 for (auto *Succ : Successors[BB]) {
1129 Edge E = std::make_pair(BB, Succ);
1130 EdgeWeights[E] = 0;
1131 VisitedEdges.insert(E);
1132 }
1133 }
1134 } else if (SelfReferentialEdge.first && VisitedBlocks.count(EC)) {
1135 uint64_t &BBWeight = BlockWeights[BB];
1136 // We have a self-referential edge and the weight of BB is known.
1137 if (BBWeight >= TotalWeight)
1138 EdgeWeights[SelfReferentialEdge] = BBWeight - TotalWeight;
1139 else
1140 EdgeWeights[SelfReferentialEdge] = 0;
1141 VisitedEdges.insert(SelfReferentialEdge);
1142 Changed = true;
1143 LLVM_DEBUG(dbgs() << "Set self-referential edge weight to: ";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "Set self-referential edge weight to: "
; printEdgeWeight(dbgs(), SelfReferentialEdge); } } while (false
)
1144 printEdgeWeight(dbgs(), SelfReferentialEdge))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "Set self-referential edge weight to: "
; printEdgeWeight(dbgs(), SelfReferentialEdge); } } while (false
);
1145 }
1146 if (UpdateBlockCount && !VisitedBlocks.count(EC) && TotalWeight > 0) {
1147 BlockWeights[EC] = TotalWeight;
1148 VisitedBlocks.insert(EC);
1149 Changed = true;
1150 }
1151 }
1152 }
1153
1154 return Changed;
1155}
1156
1157/// Build in/out edge lists for each basic block in the CFG.
1158///
1159/// We are interested in unique edges. If a block B1 has multiple
1160/// edges to another block B2, we only add a single B1->B2 edge.
1161void SampleProfileLoader::buildEdges(Function &F) {
1162 for (auto &BI : F) {
1163 BasicBlock *B1 = &BI;
1164
1165 // Add predecessors for B1.
1166 SmallPtrSet<BasicBlock *, 16> Visited;
1167 if (!Predecessors[B1].empty())
1168 llvm_unreachable("Found a stale predecessors list in a basic block.")::llvm::llvm_unreachable_internal("Found a stale predecessors list in a basic block."
, "/build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO/SampleProfile.cpp"
, 1168);
1169 for (pred_iterator PI = pred_begin(B1), PE = pred_end(B1); PI != PE; ++PI) {
1170 BasicBlock *B2 = *PI;
1171 if (Visited.insert(B2).second)
1172 Predecessors[B1].push_back(B2);
1173 }
1174
1175 // Add successors for B1.
1176 Visited.clear();
1177 if (!Successors[B1].empty())
1178 llvm_unreachable("Found a stale successors list in a basic block.")::llvm::llvm_unreachable_internal("Found a stale successors list in a basic block."
, "/build/llvm-toolchain-snapshot-8~svn350071/lib/Transforms/IPO/SampleProfile.cpp"
, 1178);
1179 for (succ_iterator SI = succ_begin(B1), SE = succ_end(B1); SI != SE; ++SI) {
1180 BasicBlock *B2 = *SI;
1181 if (Visited.insert(B2).second)
1182 Successors[B1].push_back(B2);
1183 }
1184 }
1185}
1186
1187/// Returns the sorted CallTargetMap \p M by count in descending order.
1188static SmallVector<InstrProfValueData, 2> SortCallTargets(
1189 const SampleRecord::CallTargetMap &M) {
1190 SmallVector<InstrProfValueData, 2> R;
1191 for (auto I = M.begin(); I != M.end(); ++I)
1192 R.push_back({FunctionSamples::getGUID(I->getKey()), I->getValue()});
1193 llvm::sort(R, [](const InstrProfValueData &L, const InstrProfValueData &R) {
1194 if (L.Count == R.Count)
1195 return L.Value > R.Value;
1196 else
1197 return L.Count > R.Count;
1198 });
1199 return R;
1200}
1201
1202/// Propagate weights into edges
1203///
1204/// The following rules are applied to every block BB in the CFG:
1205///
1206/// - If BB has a single predecessor/successor, then the weight
1207/// of that edge is the weight of the block.
1208///
1209/// - If all incoming or outgoing edges are known except one, and the
1210/// weight of the block is already known, the weight of the unknown
1211/// edge will be the weight of the block minus the sum of all the known
1212/// edges. If the sum of all the known edges is larger than BB's weight,
1213/// we set the unknown edge weight to zero.
1214///
1215/// - If there is a self-referential edge, and the weight of the block is
1216/// known, the weight for that edge is set to the weight of the block
1217/// minus the weight of the other incoming edges to that block (if
1218/// known).
1219void SampleProfileLoader::propagateWeights(Function &F) {
1220 bool Changed = true;
1221 unsigned I = 0;
1222
1223 // If BB weight is larger than its corresponding loop's header BB weight,
1224 // use the BB weight to replace the loop header BB weight.
1225 for (auto &BI : F) {
1226 BasicBlock *BB = &BI;
1227 Loop *L = LI->getLoopFor(BB);
1228 if (!L) {
1229 continue;
1230 }
1231 BasicBlock *Header = L->getHeader();
1232 if (Header && BlockWeights[BB] > BlockWeights[Header]) {
1233 BlockWeights[Header] = BlockWeights[BB];
1234 }
1235 }
1236
1237 // Before propagation starts, build, for each block, a list of
1238 // unique predecessors and successors. This is necessary to handle
1239 // identical edges in multiway branches. Since we visit all blocks and all
1240 // edges of the CFG, it is cleaner to build these lists once at the start
1241 // of the pass.
1242 buildEdges(F);
1243
1244 // Propagate until we converge or we go past the iteration limit.
1245 while (Changed && I++ < SampleProfileMaxPropagateIterations) {
1246 Changed = propagateThroughEdges(F, false);
1247 }
1248
1249 // The first propagation propagates BB counts from annotated BBs to unknown
1250 // BBs. The 2nd propagation pass resets edges weights, and use all BB weights
1251 // to propagate edge weights.
1252 VisitedEdges.clear();
1253 Changed = true;
1254 while (Changed && I++ < SampleProfileMaxPropagateIterations) {
1255 Changed = propagateThroughEdges(F, false);
1256 }
1257
1258 // The 3rd propagation pass allows adjust annotated BB weights that are
1259 // obviously wrong.
1260 Changed = true;
1261 while (Changed && I++ < SampleProfileMaxPropagateIterations) {
1262 Changed = propagateThroughEdges(F, true);
1263 }
1264
1265 // Generate MD_prof metadata for every branch instruction using the
1266 // edge weights computed during propagation.
1267 LLVM_DEBUG(dbgs() << "\nPropagation complete. Setting branch weights\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nPropagation complete. Setting branch weights\n"
; } } while (false);
1268 LLVMContext &Ctx = F.getContext();
1269 MDBuilder MDB(Ctx);
1270 for (auto &BI : F) {
1271 BasicBlock *BB = &BI;
1272
1273 if (BlockWeights[BB]) {
1274 for (auto &I : BB->getInstList()) {
1275 if (!isa<CallInst>(I) && !isa<InvokeInst>(I))
1276 continue;
1277 CallSite CS(&I);
1278 if (!CS.getCalledFunction()) {
1279 const DebugLoc &DLoc = I.getDebugLoc();
1280 if (!DLoc)
1281 continue;
1282 const DILocation *DIL = DLoc;
1283 uint32_t LineOffset = FunctionSamples::getOffset(DIL);
1284 uint32_t Discriminator = DIL->getBaseDiscriminator();
1285
1286 const FunctionSamples *FS = findFunctionSamples(I);
1287 if (!FS)
1288 continue;
1289 auto T = FS->findCallTargetMapAt(LineOffset, Discriminator);
1290 if (!T || T.get().empty())
1291 continue;
1292 SmallVector<InstrProfValueData, 2> SortedCallTargets =
1293 SortCallTargets(T.get());
1294 uint64_t Sum;
1295 findIndirectCallFunctionSamples(I, Sum);
1296 annotateValueSite(*I.getParent()->getParent()->getParent(), I,
1297 SortedCallTargets, Sum, IPVK_IndirectCallTarget,
1298 SortedCallTargets.size());
1299 } else if (!dyn_cast<IntrinsicInst>(&I)) {
1300 SmallVector<uint32_t, 1> Weights;
1301 Weights.push_back(BlockWeights[BB]);
1302 I.setMetadata(LLVMContext::MD_prof, MDB.createBranchWeights(Weights));
1303 }
1304 }
1305 }
1306 Instruction *TI = BB->getTerminator();
1307 if (TI->getNumSuccessors() == 1)
1308 continue;
1309 if (!isa<BranchInst>(TI) && !isa<SwitchInst>(TI))
1310 continue;
1311
1312 DebugLoc BranchLoc = TI->getDebugLoc();
1313 LLVM_DEBUG(dbgs() << "\nGetting weights for branch at line "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nGetting weights for branch at line "
<< ((BranchLoc) ? Twine(BranchLoc.getLine()) : Twine("<UNKNOWN LOCATION>"
)) << ".\n"; } } while (false)
1314 << ((BranchLoc) ? Twine(BranchLoc.getLine())do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nGetting weights for branch at line "
<< ((BranchLoc) ? Twine(BranchLoc.getLine()) : Twine("<UNKNOWN LOCATION>"
)) << ".\n"; } } while (false)
1315 : Twine("<UNKNOWN LOCATION>"))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nGetting weights for branch at line "
<< ((BranchLoc) ? Twine(BranchLoc.getLine()) : Twine("<UNKNOWN LOCATION>"
)) << ".\n"; } } while (false)
1316 << ".\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\nGetting weights for branch at line "
<< ((BranchLoc) ? Twine(BranchLoc.getLine()) : Twine("<UNKNOWN LOCATION>"
)) << ".\n"; } } while (false);
1317 SmallVector<uint32_t, 4> Weights;
1318 uint32_t MaxWeight = 0;
1319 Instruction *MaxDestInst;
1320 for (unsigned I = 0; I < TI->getNumSuccessors(); ++I) {
1321 BasicBlock *Succ = TI->getSuccessor(I);
1322 Edge E = std::make_pair(BB, Succ);
1323 uint64_t Weight = EdgeWeights[E];
1324 LLVM_DEBUG(dbgs() << "\t"; printEdgeWeight(dbgs(), E))do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "\t"; printEdgeWeight(dbgs
(), E); } } while (false);
1325 // Use uint32_t saturated arithmetic to adjust the incoming weights,
1326 // if needed. Sample counts in profiles are 64-bit unsigned values,
1327 // but internally branch weights are expressed as 32-bit values.
1328 if (Weight > std::numeric_limits<uint32_t>::max()) {
1329 LLVM_DEBUG(dbgs() << " (saturated due to uint32_t overflow)")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << " (saturated due to uint32_t overflow)"
; } } while (false);
1330 Weight = std::numeric_limits<uint32_t>::max();
1331 }
1332 // Weight is added by one to avoid propagation errors introduced by
1333 // 0 weights.
1334 Weights.push_back(static_cast<uint32_t>(Weight + 1));
1335 if (Weight != 0) {
1336 if (Weight > MaxWeight) {
1337 MaxWeight = Weight;
1338 MaxDestInst = Succ->getFirstNonPHIOrDbgOrLifetime();
1339 }
1340 }
1341 }
1342
1343 uint64_t TempWeight;
1344 // Only set weights if there is at least one non-zero weight.
1345 // In any other case, let the analyzer set weights.
1346 // Do not set weights if the weights are present. In ThinLTO, the profile
1347 // annotation is done twice. If the first annotation already set the
1348 // weights, the second pass does not need to set it.
1349 if (MaxWeight > 0 && !TI->extractProfTotalWeight(TempWeight)) {
1350 LLVM_DEBUG(dbgs() << "SUCCESS. Found non-zero weights.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "SUCCESS. Found non-zero weights.\n"
; } } while (false);
1351 TI->setMetadata(LLVMContext::MD_prof,
1352 MDB.createBranchWeights(Weights));
1353 ORE->emit([&]() {
1354 return OptimizationRemark(DEBUG_TYPE"sample-profile", "PopularDest", MaxDestInst)
1355 << "most popular destination for conditional branches at "
1356 << ore::NV("CondBranchesLoc", BranchLoc);
1357 });
1358 } else {
1359 LLVM_DEBUG(dbgs() << "SKIPPED. All branch weights are zero.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "SKIPPED. All branch weights are zero.\n"
; } } while (false);
1360 }
1361 }
1362}
1363
1364/// Get the line number for the function header.
1365///
1366/// This looks up function \p F in the current compilation unit and
1367/// retrieves the line number where the function is defined. This is
1368/// line 0 for all the samples read from the profile file. Every line
1369/// number is relative to this line.
1370///
1371/// \param F Function object to query.
1372///
1373/// \returns the line number where \p F is defined. If it returns 0,
1374/// it means that there is no debug information available for \p F.
1375unsigned SampleProfileLoader::getFunctionLoc(Function &F) {
1376 if (DISubprogram *S = F.getSubprogram())
1377 return S->getLine();
1378
1379 if (NoWarnSampleUnused)
1380 return 0;
1381
1382 // If the start of \p F is missing, emit a diagnostic to inform the user
1383 // about the missed opportunity.
1384 F.getContext().diagnose(DiagnosticInfoSampleProfile(
1385 "No debug information found in function " + F.getName() +
1386 ": Function profile not used",
1387 DS_Warning));
1388 return 0;
1389}
1390
1391void SampleProfileLoader::computeDominanceAndLoopInfo(Function &F) {
1392 DT.reset(new DominatorTree);
1393 DT->recalculate(F);
1394
1395 PDT.reset(new PostDominatorTree(F));
1396
1397 LI.reset(new LoopInfo);
1398 LI->analyze(*DT);
1399}
1400
1401/// Generate branch weight metadata for all branches in \p F.
1402///
1403/// Branch weights are computed out of instruction samples using a
1404/// propagation heuristic. Propagation proceeds in 3 phases:
1405///
1406/// 1- Assignment of block weights. All the basic blocks in the function
1407/// are initial assigned the same weight as their most frequently
1408/// executed instruction.
1409///
1410/// 2- Creation of equivalence classes. Since samples may be missing from
1411/// blocks, we can fill in the gaps by setting the weights of all the
1412/// blocks in the same equivalence class to the same weight. To compute
1413/// the concept of equivalence, we use dominance and loop information.
1414/// Two blocks B1 and B2 are in the same equivalence class if B1
1415/// dominates B2, B2 post-dominates B1 and both are in the same loop.
1416///
1417/// 3- Propagation of block weights into edges. This uses a simple
1418/// propagation heuristic. The following rules are applied to every
1419/// block BB in the CFG:
1420///
1421/// - If BB has a single predecessor/successor, then the weight
1422/// of that edge is the weight of the block.
1423///
1424/// - If all the edges are known except one, and the weight of the
1425/// block is already known, the weight of the unknown edge will
1426/// be the weight of the block minus the sum of all the known
1427/// edges. If the sum of all the known edges is larger than BB's weight,
1428/// we set the unknown edge weight to zero.
1429///
1430/// - If there is a self-referential edge, and the weight of the block is
1431/// known, the weight for that edge is set to the weight of the block
1432/// minus the weight of the other incoming edges to that block (if
1433/// known).
1434///
1435/// Since this propagation is not guaranteed to finalize for every CFG, we
1436/// only allow it to proceed for a limited number of iterations (controlled
1437/// by -sample-profile-max-propagate-iterations).
1438///
1439/// FIXME: Try to replace this propagation heuristic with a scheme
1440/// that is guaranteed to finalize. A work-list approach similar to
1441/// the standard value propagation algorithm used by SSA-CCP might
1442/// work here.
1443///
1444/// Once all the branch weights are computed, we emit the MD_prof
1445/// metadata on BB using the computed values for each of its branches.
1446///
1447/// \param F The function to query.
1448///
1449/// \returns true if \p F was modified. Returns false, otherwise.
1450bool SampleProfileLoader::emitAnnotations(Function &F) {
1451 bool Changed = false;
1452
1453 if (getFunctionLoc(F) == 0)
1454 return false;
1455
1456 LLVM_DEBUG(dbgs() << "Line number for the first instruction in "do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "Line number for the first instruction in "
<< F.getName() << ": " << getFunctionLoc(F
) << "\n"; } } while (false)
1457 << F.getName() << ": " << getFunctionLoc(F) << "\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("sample-profile")) { dbgs() << "Line number for the first instruction in "
<< F.getName() << ": " << getFunctionLoc(F
) << "\n"; } } while (false);
1458
1459 DenseSet<GlobalValue::GUID> InlinedGUIDs;
1460 Changed |= inlineHotFunctions(F, InlinedGUIDs);
1461
1462 // Compute basic block weights.
1463 Changed |= computeBlockWeights(F);
1464
1465 if (Changed) {
1466 // Add an entry count to the function using the samples gathered at the
1467 // function entry.
1468 // Sets the GUIDs that are inlined in the profiled binary. This is used
1469 // for ThinLink to make correct liveness analysis, and also make the IR
1470 // match the profiled binary before annotation.
1471 F.setEntryCount(
1472 ProfileCount(Samples->getHeadSamples() + 1, Function::PCT_Real),
1473 &InlinedGUIDs);
1474
1475 // Compute dominance and loop info needed for propagation.
1476 computeDominanceAndLoopInfo(F);
1477
1478 // Find equivalence classes.
1479 findEquivalenceClasses(F);
1480
1481 // Propagate weights to all edges.
1482 propagateWeights(F);
1483 }
1484
1485 // If coverage checking was requested, compute it now.
1486 if (SampleProfileRecordCoverage) {
1487 unsigned Used = CoverageTracker.countUsedRecords(Samples, PSI);
1488 unsigned Total = CoverageTracker.countBodyRecords(Samples, PSI);
1489 unsigned Coverage = CoverageTracker.computeCoverage(Used, Total);
1490 if (Coverage < SampleProfileRecordCoverage) {
1491 F.getContext().diagnose(DiagnosticInfoSampleProfile(
1492 F.getSubprogram()->getFilename(), getFunctionLoc(F),
1493 Twine(Used) + " of " + Twine(Total) + " available profile records (" +
1494 Twine(Coverage) + "%) were applied",
1495 DS_Warning));
1496 }
1497 }
1498
1499 if (SampleProfileSampleCoverage) {
1500 uint64_t Used = CoverageTracker.getTotalUsedSamples();
1501 uint64_t Total = CoverageTracker.countBodySamples(Samples, PSI);
1502 unsigned Coverage = CoverageTracker.computeCoverage(Used, Total);
1503 if (Coverage < SampleProfileSampleCoverage) {
1504 F.getContext().diagnose(DiagnosticInfoSampleProfile(
1505 F.getSubprogram()->getFilename(), getFunctionLoc(F),
1506 Twine(Used) + " of " + Twine(Total) + " available profile samples (" +
1507 Twine(Coverage) + "%) were applied",
1508 DS_Warning));
1509 }
1510 }
1511 return Changed;
1512}
1513
1514char SampleProfileLoaderLegacyPass::ID = 0;
1515
1516INITIALIZE_PASS_BEGIN(SampleProfileLoaderLegacyPass, "sample-profile",static void *initializeSampleProfileLoaderLegacyPassPassOnce(
PassRegistry &Registry) {
1517 "Sample Profile loader", false, false)static void *initializeSampleProfileLoaderLegacyPassPassOnce(
PassRegistry &Registry) {
1518INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)initializeAssumptionCacheTrackerPass(Registry);
1519INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)initializeTargetTransformInfoWrapperPassPass(Registry);
1520INITIALIZE_PASS_DEPENDENCY(ProfileSummaryInfoWrapperPass)initializeProfileSummaryInfoWrapperPassPass(Registry);
1521INITIALIZE_PASS_END(SampleProfileLoaderLegacyPass, "sample-profile",PassInfo *PI = new PassInfo( "Sample Profile loader", "sample-profile"
, &SampleProfileLoaderLegacyPass::ID, PassInfo::NormalCtor_t
(callDefaultCtor<SampleProfileLoaderLegacyPass>), false
, false); Registry.registerPass(*PI, true); return PI; } static
llvm::once_flag InitializeSampleProfileLoaderLegacyPassPassFlag
; void llvm::initializeSampleProfileLoaderLegacyPassPass(PassRegistry
&Registry) { llvm::call_once(InitializeSampleProfileLoaderLegacyPassPassFlag
, initializeSampleProfileLoaderLegacyPassPassOnce, std::ref(Registry
)); }
1522 "Sample Profile loader", false, false)PassInfo *PI = new PassInfo( "Sample Profile loader", "sample-profile"
, &SampleProfileLoaderLegacyPass::ID, PassInfo::NormalCtor_t
(callDefaultCtor<SampleProfileLoaderLegacyPass>), false
, false); Registry.registerPass(*PI, true); return PI; } static
llvm::once_flag InitializeSampleProfileLoaderLegacyPassPassFlag
; void llvm::initializeSampleProfileLoaderLegacyPassPass(PassRegistry
&Registry) { llvm::call_once(InitializeSampleProfileLoaderLegacyPassPassFlag
, initializeSampleProfileLoaderLegacyPassPassOnce, std::ref(Registry
)); }
1523
1524bool SampleProfileLoader::doInitialization(Module &M) {
1525 auto &Ctx = M.getContext();
1526 auto ReaderOrErr = SampleProfileReader::create(Filename, Ctx);
1527 if (std::error_code EC = ReaderOrErr.getError()) {
1528 std::string Msg = "Could not open profile: " + EC.message();
1529 Ctx.diagnose(DiagnosticInfoSampleProfile(Filename, Msg));
1530 return false;
1531 }
1532 Reader = std::move(ReaderOrErr.get());
1533 Reader->collectFuncsToUse(M);
1534 ProfileIsValid = (Reader->read() == sampleprof_error::success);
1535
1536 if (!RemappingFilename.empty()) {
1537 // Apply profile remappings to the loaded profile data if requested.
1538 // For now, we only support remapping symbols encoded using the Itanium
1539 // C++ ABI's name mangling scheme.
1540 ReaderOrErr = SampleProfileReaderItaniumRemapper::create(
1541 RemappingFilename, Ctx, std::move(Reader));
1542 if (std::error_code EC = ReaderOrErr.getError()) {
1543 std::string Msg = "Could not open profile remapping file: " + EC.message();
1544 Ctx.diagnose(DiagnosticInfoSampleProfile(Filename, Msg));
1545 return false;
1546 }
1547 Reader = std::move(ReaderOrErr.get());
1548 ProfileIsValid = (Reader->read() == sampleprof_error::success);
1549 }
1550 return true;
1551}
1552
1553ModulePass *llvm::createSampleProfileLoaderPass() {
1554 return new SampleProfileLoaderLegacyPass();
1555}
1556
1557ModulePass *llvm::createSampleProfileLoaderPass(StringRef Name) {
1558 return new SampleProfileLoaderLegacyPass(Name);
1559}
1560
1561bool SampleProfileLoader::runOnModule(Module &M, ModuleAnalysisManager *AM,
1562 ProfileSummaryInfo *_PSI) {
1563 FunctionSamples::GUIDToFuncNameMapper Mapper(M);
1564 if (!ProfileIsValid)
1565 return false;
1566
1567 PSI = _PSI;
1568 if (M.getProfileSummary() == nullptr)
1569 M.setProfileSummary(Reader->getSummary().getMD(M.getContext()));
1570
1571 // Compute the total number of samples collected in this profile.
1572 for (const auto &I : Reader->getProfiles())
1573 TotalCollectedSamples += I.second.getTotalSamples();
1574
1575 // Populate the symbol map.
1576 for (const auto &N_F : M.getValueSymbolTable()) {
1577 StringRef OrigName = N_F.getKey();
1578 Function *F = dyn_cast<Function>(N_F.getValue());
1579 if (F == nullptr)
1580 continue;
1581 SymbolMap[OrigName] = F;
1582 auto pos = OrigName.find('.');
1583 if (pos != StringRef::npos) {
1584 StringRef NewName = OrigName.substr(0, pos);
1585 auto r = SymbolMap.insert(std::make_pair(NewName, F));
1586 // Failiing to insert means there is already an entry in SymbolMap,
1587 // thus there are multiple functions that are mapped to the same
1588 // stripped name. In this case of name conflicting, set the value
1589 // to nullptr to avoid confusion.
1590 if (!r.second)
1591 r.first->second = nullptr;
1592 }
1593 }
1594
1595 bool retval = false;
1596 for (auto &F : M)
1597 if (!F.isDeclaration()) {
1598 clearFunctionData();
1599 retval |= runOnFunction(F, AM);
1600 }
1601 return retval;
1602}
1603
1604bool SampleProfileLoaderLegacyPass::runOnModule(Module &M) {
1605 ACT = &getAnalysis<AssumptionCacheTracker>();
1606 TTIWP = &getAnalysis<TargetTransformInfoWrapperPass>();
1607 ProfileSummaryInfo *PSI =
1608 &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI();
1609 return SampleLoader.runOnModule(M, nullptr, PSI);
1610}
1611
1612bool SampleProfileLoader::runOnFunction(Function &F, ModuleAnalysisManager *AM) {
1613 // By default the entry count is initialized to -1, which will be treated
1614 // conservatively by getEntryCount as the same as unknown (None). This is
1615 // to avoid newly added code to be treated as cold. If we have samples
1616 // this will be overwritten in emitAnnotations.
1617 // If ProfileSampleAccurate is true or F has profile-sample-accurate
1618 // attribute, initialize the entry count to 0 so callsites or functions
1619 // unsampled will be treated as cold.
1620 uint64_t initialEntryCount =
1621 (ProfileSampleAccurate || F.hasFnAttribute("profile-sample-accurate"))
1622 ? 0
1623 : -1;
1624 F.setEntryCount(ProfileCount(initialEntryCount, Function::PCT_Real));
1625 std::unique_ptr<OptimizationRemarkEmitter> OwnedORE;
1626 if (AM) {
1627 auto &FAM =
1628 AM->getResult<FunctionAnalysisManagerModuleProxy>(*F.getParent())
1629 .getManager();
1630 ORE = &FAM.getResult<OptimizationRemarkEmitterAnalysis>(F);
1631 } else {
1632 OwnedORE = make_unique<OptimizationRemarkEmitter>(&F);
1633 ORE = OwnedORE.get();
1634 }
1635 Samples = Reader->getSamplesFor(F);
1636 if (Samples && !Samples->empty())
1637 return emitAnnotations(F);
1638 return false;
1639}
1640
1641PreservedAnalyses SampleProfileLoaderPass::run(Module &M,
1642 ModuleAnalysisManager &AM) {
1643 FunctionAnalysisManager &FAM =
1644 AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
1645
1646 auto GetAssumptionCache = [&](Function &F) -> AssumptionCache & {
1647 return FAM.getResult<AssumptionAnalysis>(F);
1648 };
1649 auto GetTTI = [&](Function &F) -> TargetTransformInfo & {
1650 return FAM.getResult<TargetIRAnalysis>(F);
1651 };
1652
1653 SampleProfileLoader SampleLoader(
1654 ProfileFileName.empty() ? SampleProfileFile : ProfileFileName,
1655 ProfileRemappingFileName.empty() ? SampleProfileRemappingFile
1656 : ProfileRemappingFileName,
1657 IsThinLTOPreLink, GetAssumptionCache, GetTTI);
1658
1659 SampleLoader.doInitialization(M);
1660
1661 ProfileSummaryInfo *PSI = &AM.getResult<ProfileSummaryAnalysis>(M);
1662 if (!SampleLoader.runOnModule(M, &AM, PSI))
1663 return PreservedAnalyses::all();
1664
1665 return PreservedAnalyses::none();
1666}