1 //===- ProfileSummaryInfo.cpp - Global profile summary information --------===//
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 contains a pass that provides access to the global profile summary
11 // information.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "llvm/Analysis/ProfileSummaryInfo.h"
16 #include "llvm/Analysis/BlockFrequencyInfo.h"
17 #include "llvm/IR/BasicBlock.h"
18 #include "llvm/IR/CallSite.h"
19 #include "llvm/IR/Metadata.h"
20 #include "llvm/IR/Module.h"
21 #include "llvm/IR/ProfileSummary.h"
22 using namespace llvm;
23 
24 // The following two parameters determine the threshold for a count to be
25 // considered hot/cold. These two parameters are percentile values (multiplied
26 // by 10000). If the counts are sorted in descending order, the minimum count to
27 // reach ProfileSummaryCutoffHot gives the threshold to determine a hot count.
28 // Similarly, the minimum count to reach ProfileSummaryCutoffCold gives the
29 // threshold for determining cold count (everything <= this threshold is
30 // considered cold).
31 
32 static cl::opt<int> ProfileSummaryCutoffHot(
33     "profile-summary-cutoff-hot", cl::Hidden, cl::init(999000), cl::ZeroOrMore,
34     cl::desc("A count is hot if it exceeds the minimum count to"
35              " reach this percentile of total counts."));
36 
37 static cl::opt<int> ProfileSummaryCutoffCold(
38     "profile-summary-cutoff-cold", cl::Hidden, cl::init(999999), cl::ZeroOrMore,
39     cl::desc("A count is cold if it is below the minimum count"
40              " to reach this percentile of total counts."));
41 
42 // Find the minimum count to reach a desired percentile of counts.
43 static uint64_t getMinCountForPercentile(SummaryEntryVector &DS,
44                                          uint64_t Percentile) {
45   auto Compare = [](const ProfileSummaryEntry &Entry, uint64_t Percentile) {
46     return Entry.Cutoff < Percentile;
47   };
48   auto It = std::lower_bound(DS.begin(), DS.end(), Percentile, Compare);
49   // The required percentile has to be <= one of the percentiles in the
50   // detailed summary.
51   if (It == DS.end())
52     report_fatal_error("Desired percentile exceeds the maximum cutoff");
53   return It->MinCount;
54 }
55 
56 // The profile summary metadata may be attached either by the frontend or by
57 // any backend passes (IR level instrumentation, for example). This method
58 // checks if the Summary is null and if so checks if the summary metadata is now
59 // available in the module and parses it to get the Summary object. Returns true
60 // if a valid Summary is available.
61 bool ProfileSummaryInfo::computeSummary() {
62   if (Summary)
63     return true;
64   auto *SummaryMD = M.getProfileSummary();
65   if (!SummaryMD)
66     return false;
67   Summary.reset(ProfileSummary::getFromMD(SummaryMD));
68   return true;
69 }
70 
71 /// Returns true if the function's entry is hot. If it returns false, it
72 /// either means it is not hot or it is unknown whether it is hot or not (for
73 /// example, no profile data is available).
74 bool ProfileSummaryInfo::isFunctionEntryHot(const Function *F) {
75   if (!F || !computeSummary())
76     return false;
77   auto FunctionCount = F->getEntryCount();
78   // FIXME: The heuristic used below for determining hotness is based on
79   // preliminary SPEC tuning for inliner. This will eventually be a
80   // convenience method that calls isHotCount.
81   return FunctionCount && isHotCount(FunctionCount.getValue());
82 }
83 
84 /// Returns true if the function's entry is a cold. If it returns false, it
85 /// either means it is not cold or it is unknown whether it is cold or not (for
86 /// example, no profile data is available).
87 bool ProfileSummaryInfo::isFunctionEntryCold(const Function *F) {
88   if (!F)
89     return false;
90   if (F->hasFnAttribute(Attribute::Cold)) {
91     return true;
92   }
93   if (!computeSummary())
94     return false;
95   auto FunctionCount = F->getEntryCount();
96   // FIXME: The heuristic used below for determining coldness is based on
97   // preliminary SPEC tuning for inliner. This will eventually be a
98   // convenience method that calls isHotCount.
99   return FunctionCount && isColdCount(FunctionCount.getValue());
100 }
101 
102 /// Compute the hot and cold thresholds.
103 void ProfileSummaryInfo::computeThresholds() {
104   if (!computeSummary())
105     return;
106   auto &DetailedSummary = Summary->getDetailedSummary();
107   HotCountThreshold =
108       getMinCountForPercentile(DetailedSummary, ProfileSummaryCutoffHot);
109   ColdCountThreshold =
110       getMinCountForPercentile(DetailedSummary, ProfileSummaryCutoffCold);
111 }
112 
113 bool ProfileSummaryInfo::isHotCount(uint64_t C) {
114   if (!HotCountThreshold)
115     computeThresholds();
116   return HotCountThreshold && C >= HotCountThreshold.getValue();
117 }
118 
119 bool ProfileSummaryInfo::isColdCount(uint64_t C) {
120   if (!ColdCountThreshold)
121     computeThresholds();
122   return ColdCountThreshold && C <= ColdCountThreshold.getValue();
123 }
124 
125 bool ProfileSummaryInfo::isHotBB(const BasicBlock *B, BlockFrequencyInfo *BFI) {
126   auto Count = BFI->getBlockProfileCount(B);
127   if (Count && isHotCount(*Count))
128     return true;
129   // Use extractProfTotalWeight to get BB count.
130   // For Sample PGO, BFI may not provide accurate BB count due to errors
131   // magnified during sample count propagation. This serves as a backup plan
132   // to ensure all hot BB will not be missed.
133   // The query currently has false positives as branch instruction cloning does
134   // not update/scale branch weights. Unlike false negatives, this will not cause
135   // performance problem.
136   uint64_t TotalCount;
137   auto *TI = B->getTerminator();
138   return extractProfTotalWeight(TI, TotalCount) && isHotCount(TotalCount);
139 }
140 
141 bool ProfileSummaryInfo::isColdBB(const BasicBlock *B,
142                                   BlockFrequencyInfo *BFI) {
143   auto Count = BFI->getBlockProfileCount(B);
144   return Count && isColdCount(*Count);
145 }
146 
147 bool ProfileSummaryInfo::extractProfTotalWeight(const Instruction *I,
148                                                 uint64_t &TotalCount) {
149   if (!computeSummary())
150     return false;
151   // Use profile weight on metadata only for sample profiling where block counts
152   // could differ from the count of an instruction within the block.
153   if (Summary.get()->getKind() != ProfileSummary::PSK_Sample)
154     return false;
155 
156   return (isa<CallInst>(I) ||
157           (isa<TerminatorInst>(I) && !isa<ReturnInst>(I))) &&
158          I->extractProfTotalWeight(TotalCount);
159 }
160 
161 bool ProfileSummaryInfo::isHotCallSite(const CallSite &CS,
162                                        BlockFrequencyInfo *BFI) {
163   auto *CallInst = CS.getInstruction();
164   if (!CS)
165     return false;
166   // Check if there is a profile metadata on the instruction. If it is present,
167   // determine hotness solely based on that.
168   uint64_t TotalCount;
169   if (extractProfTotalWeight(CallInst, TotalCount))
170     return isHotCount(TotalCount);
171   return BFI && isHotBB(CallInst->getParent(), BFI);
172 }
173 
174 bool ProfileSummaryInfo::isColdCallSite(const CallSite &CS,
175                                         BlockFrequencyInfo *BFI) {
176   auto *CallInst = CS.getInstruction();
177   if (!CS)
178     return false;
179   // Check if there is a profile metadata on the instruction. If it is present,
180   // and tells that the callsite is not cold, then return false;
181   uint64_t TotalCount;
182   if (extractProfTotalWeight(CallInst, TotalCount) && !isColdCount(TotalCount))
183     return false;
184   return BFI && isColdBB(CallInst->getParent(), BFI);
185 }
186 
187 INITIALIZE_PASS(ProfileSummaryInfoWrapperPass, "profile-summary-info",
188                 "Profile summary info", false, true)
189 
190 ProfileSummaryInfoWrapperPass::ProfileSummaryInfoWrapperPass()
191     : ImmutablePass(ID) {
192   initializeProfileSummaryInfoWrapperPassPass(*PassRegistry::getPassRegistry());
193 }
194 
195 bool ProfileSummaryInfoWrapperPass::doInitialization(Module &M) {
196   PSI.reset(new ProfileSummaryInfo(M));
197   return false;
198 }
199 
200 bool ProfileSummaryInfoWrapperPass::doFinalization(Module &M) {
201   PSI.reset();
202   return false;
203 }
204 
205 AnalysisKey ProfileSummaryAnalysis::Key;
206 ProfileSummaryInfo ProfileSummaryAnalysis::run(Module &M,
207                                                ModuleAnalysisManager &) {
208   return ProfileSummaryInfo(M);
209 }
210 
211 PreservedAnalyses ProfileSummaryPrinterPass::run(Module &M,
212                                                  ModuleAnalysisManager &AM) {
213   ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M);
214 
215   OS << "Functions in " << M.getName() << " with hot/cold annotations: \n";
216   for (auto &F : M) {
217     OS << F.getName();
218     if (PSI.isFunctionEntryHot(&F))
219       OS << " :hot entry ";
220     else if (PSI.isFunctionEntryCold(&F))
221       OS << " :cold entry ";
222     OS << "\n";
223   }
224   return PreservedAnalyses::all();
225 }
226 
227 char ProfileSummaryInfoWrapperPass::ID = 0;
228