1 //===- ProfileSummaryInfo.cpp - Global profile summary information --------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file contains a pass that provides access to the global profile summary 10 // information. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/Analysis/ProfileSummaryInfo.h" 15 #include "llvm/Analysis/BlockFrequencyInfo.h" 16 #include "llvm/IR/BasicBlock.h" 17 #include "llvm/IR/Instructions.h" 18 #include "llvm/IR/Module.h" 19 #include "llvm/IR/ProfileSummary.h" 20 #include "llvm/InitializePasses.h" 21 #include "llvm/ProfileData/ProfileCommon.h" 22 #include "llvm/Support/CommandLine.h" 23 #include <optional> 24 using namespace llvm; 25 26 // Knobs for profile summary based thresholds. 27 namespace llvm { 28 extern cl::opt<int> ProfileSummaryCutoffHot; 29 extern cl::opt<int> ProfileSummaryCutoffCold; 30 extern cl::opt<unsigned> ProfileSummaryHugeWorkingSetSizeThreshold; 31 extern cl::opt<unsigned> ProfileSummaryLargeWorkingSetSizeThreshold; 32 extern cl::opt<int> ProfileSummaryHotCount; 33 extern cl::opt<int> ProfileSummaryColdCount; 34 } // namespace llvm 35 36 static cl::opt<bool> PartialProfile( 37 "partial-profile", cl::Hidden, cl::init(false), 38 cl::desc("Specify the current profile is used as a partial profile.")); 39 40 cl::opt<bool> ScalePartialSampleProfileWorkingSetSize( 41 "scale-partial-sample-profile-working-set-size", cl::Hidden, cl::init(true), 42 cl::desc( 43 "If true, scale the working set size of the partial sample profile " 44 "by the partial profile ratio to reflect the size of the program " 45 "being compiled.")); 46 47 static cl::opt<double> PartialSampleProfileWorkingSetSizeScaleFactor( 48 "partial-sample-profile-working-set-size-scale-factor", cl::Hidden, 49 cl::init(0.008), 50 cl::desc("The scale factor used to scale the working set size of the " 51 "partial sample profile along with the partial profile ratio. " 52 "This includes the factor of the profile counter per block " 53 "and the factor to scale the working set size to use the same " 54 "shared thresholds as PGO.")); 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. 60 void ProfileSummaryInfo::refresh() { 61 if (hasProfileSummary()) 62 return; 63 // First try to get context sensitive ProfileSummary. 64 auto *SummaryMD = M->getProfileSummary(/* IsCS */ true); 65 if (SummaryMD) 66 Summary.reset(ProfileSummary::getFromMD(SummaryMD)); 67 68 if (!hasProfileSummary()) { 69 // This will actually return PSK_Instr or PSK_Sample summary. 70 SummaryMD = M->getProfileSummary(/* IsCS */ false); 71 if (SummaryMD) 72 Summary.reset(ProfileSummary::getFromMD(SummaryMD)); 73 } 74 if (!hasProfileSummary()) 75 return; 76 computeThresholds(); 77 } 78 79 std::optional<uint64_t> ProfileSummaryInfo::getProfileCount( 80 const CallBase &Call, BlockFrequencyInfo *BFI, bool AllowSynthetic) const { 81 assert((isa<CallInst>(Call) || isa<InvokeInst>(Call)) && 82 "We can only get profile count for call/invoke instruction."); 83 if (hasSampleProfile()) { 84 // In sample PGO mode, check if there is a profile metadata on the 85 // instruction. If it is present, determine hotness solely based on that, 86 // since the sampled entry count may not be accurate. If there is no 87 // annotated on the instruction, return std::nullopt. 88 uint64_t TotalCount; 89 if (Call.extractProfTotalWeight(TotalCount)) 90 return TotalCount; 91 return std::nullopt; 92 } 93 if (BFI) 94 return BFI->getBlockProfileCount(Call.getParent(), AllowSynthetic); 95 return std::nullopt; 96 } 97 98 bool ProfileSummaryInfo::isFunctionHotnessUnknown(const Function &F) const { 99 assert(hasPartialSampleProfile() && "Expect partial sample profile"); 100 return !F.getEntryCount(); 101 } 102 103 /// Returns true if the function's entry is a cold. If it returns false, it 104 /// either means it is not cold or it is unknown whether it is cold or not (for 105 /// example, no profile data is available). 106 bool ProfileSummaryInfo::isFunctionEntryCold(const Function *F) const { 107 if (!F) 108 return false; 109 if (F->hasFnAttribute(Attribute::Cold)) 110 return true; 111 if (!hasProfileSummary()) 112 return false; 113 auto FunctionCount = F->getEntryCount(); 114 // FIXME: The heuristic used below for determining coldness is based on 115 // preliminary SPEC tuning for inliner. This will eventually be a 116 // convenience method that calls isHotCount. 117 return FunctionCount && isColdCount(FunctionCount->getCount()); 118 } 119 120 /// Compute the hot and cold thresholds. 121 void ProfileSummaryInfo::computeThresholds() { 122 auto &DetailedSummary = Summary->getDetailedSummary(); 123 auto &HotEntry = ProfileSummaryBuilder::getEntryForPercentile( 124 DetailedSummary, ProfileSummaryCutoffHot); 125 HotCountThreshold = 126 ProfileSummaryBuilder::getHotCountThreshold(DetailedSummary); 127 ColdCountThreshold = 128 ProfileSummaryBuilder::getColdCountThreshold(DetailedSummary); 129 assert(ColdCountThreshold <= HotCountThreshold && 130 "Cold count threshold cannot exceed hot count threshold!"); 131 if (!hasPartialSampleProfile() || !ScalePartialSampleProfileWorkingSetSize) { 132 HasHugeWorkingSetSize = 133 HotEntry.NumCounts > ProfileSummaryHugeWorkingSetSizeThreshold; 134 HasLargeWorkingSetSize = 135 HotEntry.NumCounts > ProfileSummaryLargeWorkingSetSizeThreshold; 136 } else { 137 // Scale the working set size of the partial sample profile to reflect the 138 // size of the program being compiled. 139 double PartialProfileRatio = Summary->getPartialProfileRatio(); 140 uint64_t ScaledHotEntryNumCounts = 141 static_cast<uint64_t>(HotEntry.NumCounts * PartialProfileRatio * 142 PartialSampleProfileWorkingSetSizeScaleFactor); 143 HasHugeWorkingSetSize = 144 ScaledHotEntryNumCounts > ProfileSummaryHugeWorkingSetSizeThreshold; 145 HasLargeWorkingSetSize = 146 ScaledHotEntryNumCounts > ProfileSummaryLargeWorkingSetSizeThreshold; 147 } 148 } 149 150 std::optional<uint64_t> 151 ProfileSummaryInfo::computeThreshold(int PercentileCutoff) const { 152 if (!hasProfileSummary()) 153 return std::nullopt; 154 auto iter = ThresholdCache.find(PercentileCutoff); 155 if (iter != ThresholdCache.end()) { 156 return iter->second; 157 } 158 auto &DetailedSummary = Summary->getDetailedSummary(); 159 auto &Entry = ProfileSummaryBuilder::getEntryForPercentile(DetailedSummary, 160 PercentileCutoff); 161 uint64_t CountThreshold = Entry.MinCount; 162 ThresholdCache[PercentileCutoff] = CountThreshold; 163 return CountThreshold; 164 } 165 166 bool ProfileSummaryInfo::hasHugeWorkingSetSize() const { 167 return HasHugeWorkingSetSize && *HasHugeWorkingSetSize; 168 } 169 170 bool ProfileSummaryInfo::hasLargeWorkingSetSize() const { 171 return HasLargeWorkingSetSize && *HasLargeWorkingSetSize; 172 } 173 174 bool ProfileSummaryInfo::isHotCount(uint64_t C) const { 175 return HotCountThreshold && C >= *HotCountThreshold; 176 } 177 178 bool ProfileSummaryInfo::isColdCount(uint64_t C) const { 179 return ColdCountThreshold && C <= *ColdCountThreshold; 180 } 181 182 template <bool isHot> 183 bool ProfileSummaryInfo::isHotOrColdCountNthPercentile(int PercentileCutoff, 184 uint64_t C) const { 185 auto CountThreshold = computeThreshold(PercentileCutoff); 186 if (isHot) 187 return CountThreshold && C >= *CountThreshold; 188 else 189 return CountThreshold && C <= *CountThreshold; 190 } 191 192 bool ProfileSummaryInfo::isHotCountNthPercentile(int PercentileCutoff, 193 uint64_t C) const { 194 return isHotOrColdCountNthPercentile<true>(PercentileCutoff, C); 195 } 196 197 bool ProfileSummaryInfo::isColdCountNthPercentile(int PercentileCutoff, 198 uint64_t C) const { 199 return isHotOrColdCountNthPercentile<false>(PercentileCutoff, C); 200 } 201 202 uint64_t ProfileSummaryInfo::getOrCompHotCountThreshold() const { 203 return HotCountThreshold.value_or(UINT64_MAX); 204 } 205 206 uint64_t ProfileSummaryInfo::getOrCompColdCountThreshold() const { 207 return ColdCountThreshold.value_or(0); 208 } 209 210 bool ProfileSummaryInfo::isHotCallSite(const CallBase &CB, 211 BlockFrequencyInfo *BFI) const { 212 auto C = getProfileCount(CB, BFI); 213 return C && isHotCount(*C); 214 } 215 216 bool ProfileSummaryInfo::isColdCallSite(const CallBase &CB, 217 BlockFrequencyInfo *BFI) const { 218 auto C = getProfileCount(CB, BFI); 219 if (C) 220 return isColdCount(*C); 221 222 // In SamplePGO, if the caller has been sampled, and there is no profile 223 // annotated on the callsite, we consider the callsite as cold. 224 return hasSampleProfile() && CB.getCaller()->hasProfileData(); 225 } 226 227 bool ProfileSummaryInfo::hasPartialSampleProfile() const { 228 return hasProfileSummary() && 229 Summary->getKind() == ProfileSummary::PSK_Sample && 230 (PartialProfile || Summary->isPartialProfile()); 231 } 232 233 INITIALIZE_PASS(ProfileSummaryInfoWrapperPass, "profile-summary-info", 234 "Profile summary info", false, true) 235 236 ProfileSummaryInfoWrapperPass::ProfileSummaryInfoWrapperPass() 237 : ImmutablePass(ID) { 238 initializeProfileSummaryInfoWrapperPassPass(*PassRegistry::getPassRegistry()); 239 } 240 241 bool ProfileSummaryInfoWrapperPass::doInitialization(Module &M) { 242 PSI.reset(new ProfileSummaryInfo(M)); 243 return false; 244 } 245 246 bool ProfileSummaryInfoWrapperPass::doFinalization(Module &M) { 247 PSI.reset(); 248 return false; 249 } 250 251 AnalysisKey ProfileSummaryAnalysis::Key; 252 ProfileSummaryInfo ProfileSummaryAnalysis::run(Module &M, 253 ModuleAnalysisManager &) { 254 return ProfileSummaryInfo(M); 255 } 256 257 PreservedAnalyses ProfileSummaryPrinterPass::run(Module &M, 258 ModuleAnalysisManager &AM) { 259 ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M); 260 261 OS << "Functions in " << M.getName() << " with hot/cold annotations: \n"; 262 for (auto &F : M) { 263 OS << F.getName(); 264 if (PSI.isFunctionEntryHot(&F)) 265 OS << " :hot entry "; 266 else if (PSI.isFunctionEntryCold(&F)) 267 OS << " :cold entry "; 268 OS << "\n"; 269 } 270 return PreservedAnalyses::all(); 271 } 272 273 char ProfileSummaryInfoWrapperPass::ID = 0; 274