lib/Analysis/ProfileSummaryInfo.cpp

   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/CallSite.h"
  18 #include "llvm/IR/Metadata.h"
  19 #include "llvm/IR/Module.h"
  20 #include "llvm/IR/ProfileSummary.h"
  21 using namespace llvm;
  22
  23 // The following two parameters determine the threshold for a count to be
  24 // considered hot/cold. These two parameters are percentile values (multiplied
  25 // by 10000). If the counts are sorted in descending order, the minimum count to
  26 // reach ProfileSummaryCutoffHot gives the threshold to determine a hot count.
  27 // Similarly, the minimum count to reach ProfileSummaryCutoffCold gives the
  28 // threshold for determining cold count (everything <= this threshold is
  29 // considered cold).
  30
  31 static cl::opt<int> ProfileSummaryCutoffHot(
  32     "profile-summary-cutoff-hot", cl::Hidden, cl::init(990000), cl::ZeroOrMore,
  33     cl::desc("A count is hot if it exceeds the minimum count to"
  34              " reach this percentile of total counts."));
  35
  36 static cl::opt<int> ProfileSummaryCutoffCold(
  37     "profile-summary-cutoff-cold", cl::Hidden, cl::init(999999), cl::ZeroOrMore,
  38     cl::desc("A count is cold if it is below the minimum count"
  39              " to reach this percentile of total counts."));
  40
  41 static cl::opt<unsigned> ProfileSummaryHugeWorkingSetSizeThreshold(
  42     "profile-summary-huge-working-set-size-threshold", cl::Hidden,
  43     cl::init(15000), cl::ZeroOrMore,
  44     cl::desc("The code working set size is considered huge if the number of"
  45              " blocks required to reach the -profile-summary-cutoff-hot"
  46              " percentile exceeds this count."));
  47
  48 // The next two options override the counts derived from summary computation and
  49 // are useful for debugging purposes.
  50 static cl::opt<int> ProfileSummaryHotCount(
  51     "profile-summary-hot-count", cl::ReallyHidden, cl::ZeroOrMore,
  52     cl::desc("A fixed hot count that overrides the count derived from"
  53              " profile-summary-cutoff-hot"));
  54
  55 static cl::opt<int> ProfileSummaryColdCount(
  56     "profile-summary-cold-count", cl::ReallyHidden, cl::ZeroOrMore,
  57     cl::desc("A fixed cold count that overrides the count derived from"
  58              " profile-summary-cutoff-cold"));
  59
  60 // Find the summary entry for a desired percentile of counts.
  61 static const ProfileSummaryEntry &getEntryForPercentile(SummaryEntryVector &DS,
  62                                                         uint64_t Percentile) {
  63   auto It = partition_point(DS, [=](const ProfileSummaryEntry &Entry) {
  64     return Entry.Cutoff < Percentile;
  65   });
  66   // The required percentile has to be <= one of the percentiles in the
  67   // detailed summary.
  68   if (It == DS.end())
  69     report_fatal_error("Desired percentile exceeds the maximum cutoff");
  70   return *It;
  71 }
  72
  73 // The profile summary metadata may be attached either by the frontend or by
  74 // any backend passes (IR level instrumentation, for example). This method
  75 // checks if the Summary is null and if so checks if the summary metadata is now
  76 // available in the module and parses it to get the Summary object. Returns true
  77 // if a valid Summary is available.
  78 bool ProfileSummaryInfo::computeSummary() {
  79   if (Summary)
  80     return true;
  81   // First try to get context sensitive ProfileSummary.
  82   auto *SummaryMD = M.getProfileSummary(/* IsCS */ true);
  83   if (SummaryMD) {
  84     Summary.reset(ProfileSummary::getFromMD(SummaryMD));
  85     return true;
  86   }
  87   // This will actually return PSK_Instr or PSK_Sample summary.
  88   SummaryMD = M.getProfileSummary(/* IsCS */ false);
  89   if (!SummaryMD)
  90     return false;
  91   Summary.reset(ProfileSummary::getFromMD(SummaryMD));
  92   return true;
  93 }
  94
  95 Optional<uint64_t>
  96 ProfileSummaryInfo::getProfileCount(const Instruction *Inst,
  97                                     BlockFrequencyInfo *BFI,
  98                                     bool AllowSynthetic) {
  99   if (!Inst)
 100     return None;
 101   assert((isa<CallInst>(Inst) || isa<InvokeInst>(Inst)) &&
 102          "We can only get profile count for call/invoke instruction.");
 103   if (hasSampleProfile()) {
 104     // In sample PGO mode, check if there is a profile metadata on the
 105     // instruction. If it is present, determine hotness solely based on that,
 106     // since the sampled entry count may not be accurate. If there is no
 107     // annotated on the instruction, return None.
 108     uint64_t TotalCount;
 109     if (Inst->extractProfTotalWeight(TotalCount))
 110       return TotalCount;
 111     return None;
 112   }
 113   if (BFI)
 114     return BFI->getBlockProfileCount(Inst->getParent(), AllowSynthetic);
 115   return None;
 116 }
 117
 118 /// Returns true if the function's entry is hot. If it returns false, it
 119 /// either means it is not hot or it is unknown whether it is hot or not (for
 120 /// example, no profile data is available).
 121 bool ProfileSummaryInfo::isFunctionEntryHot(const Function *F) {
 122   if (!F || !computeSummary())
 123     return false;
 124   auto FunctionCount = F->getEntryCount();
 125   // FIXME: The heuristic used below for determining hotness is based on
 126   // preliminary SPEC tuning for inliner. This will eventually be a
 127   // convenience method that calls isHotCount.
 128   return FunctionCount && isHotCount(FunctionCount.getCount());
 129 }
 130
 131 /// Returns true if the function contains hot code. This can include a hot
 132 /// function entry count, hot basic block, or (in the case of Sample PGO)
 133 /// hot total call edge count.
 134 /// If it returns false, it either means it is not hot or it is unknown
 135 /// (for example, no profile data is available).
 136 bool ProfileSummaryInfo::isFunctionHotInCallGraph(const Function *F,
 137                                                   BlockFrequencyInfo &BFI) {
 138   if (!F || !computeSummary())
 139     return false;
 140   if (auto FunctionCount = F->getEntryCount())
 141     if (isHotCount(FunctionCount.getCount()))
 142       return true;
 143
 144   if (hasSampleProfile()) {
 145     uint64_t TotalCallCount = 0;
 146     for (const auto &BB : *F)
 147       for (const auto &I : BB)
 148         if (isa<CallInst>(I) || isa<InvokeInst>(I))
 149           if (auto CallCount = getProfileCount(&I, nullptr))
 150             TotalCallCount += CallCount.getValue();
 151     if (isHotCount(TotalCallCount))
 152       return true;
 153   }
 154   for (const auto &BB : *F)
 155     if (isHotBlock(&BB, &BFI))
 156       return true;
 157   return false;
 158 }
 159
 160 /// Returns true if the function only contains cold code. This means that
 161 /// the function entry and blocks are all cold, and (in the case of Sample PGO)
 162 /// the total call edge count is cold.
 163 /// If it returns false, it either means it is not cold or it is unknown
 164 /// (for example, no profile data is available).
 165 bool ProfileSummaryInfo::isFunctionColdInCallGraph(const Function *F,
 166                                                    BlockFrequencyInfo &BFI) {
 167   if (!F || !computeSummary())
 168     return false;
 169   if (auto FunctionCount = F->getEntryCount())
 170     if (!isColdCount(FunctionCount.getCount()))
 171       return false;
 172
 173   if (hasSampleProfile()) {
 174     uint64_t TotalCallCount = 0;
 175     for (const auto &BB : *F)
 176       for (const auto &I : BB)
 177         if (isa<CallInst>(I) || isa<InvokeInst>(I))
 178           if (auto CallCount = getProfileCount(&I, nullptr))
 179             TotalCallCount += CallCount.getValue();
 180     if (!isColdCount(TotalCallCount))
 181       return false;
 182   }
 183   for (const auto &BB : *F)
 184     if (!isColdBlock(&BB, &BFI))
 185       return false;
 186   return true;
 187 }
 188
 189 /// Returns true if the function's entry is a cold. If it returns false, it
 190 /// either means it is not cold or it is unknown whether it is cold or not (for
 191 /// example, no profile data is available).
 192 bool ProfileSummaryInfo::isFunctionEntryCold(const Function *F) {
 193   if (!F)
 194     return false;
 195   if (F->hasFnAttribute(Attribute::Cold))
 196     return true;
 197   if (!computeSummary())
 198     return false;
 199   auto FunctionCount = F->getEntryCount();
 200   // FIXME: The heuristic used below for determining coldness is based on
 201   // preliminary SPEC tuning for inliner. This will eventually be a
 202   // convenience method that calls isHotCount.
 203   return FunctionCount && isColdCount(FunctionCount.getCount());
 204 }
 205
 206 /// Compute the hot and cold thresholds.
 207 void ProfileSummaryInfo::computeThresholds() {
 208   if (!computeSummary())
 209     return;
 210   auto &DetailedSummary = Summary->getDetailedSummary();
 211   auto &HotEntry =
 212       getEntryForPercentile(DetailedSummary, ProfileSummaryCutoffHot);
 213   HotCountThreshold = HotEntry.MinCount;
 214   if (ProfileSummaryHotCount.getNumOccurrences() > 0)
 215     HotCountThreshold = ProfileSummaryHotCount;
 216   auto &ColdEntry =
 217       getEntryForPercentile(DetailedSummary, ProfileSummaryCutoffCold);
 218   ColdCountThreshold = ColdEntry.MinCount;
 219   if (ProfileSummaryColdCount.getNumOccurrences() > 0)
 220     ColdCountThreshold = ProfileSummaryColdCount;
 221   assert(ColdCountThreshold <= HotCountThreshold &&
 222          "Cold count threshold cannot exceed hot count threshold!");
 223   HasHugeWorkingSetSize =
 224       HotEntry.NumCounts > ProfileSummaryHugeWorkingSetSizeThreshold;
 225 }
 226
 227 bool ProfileSummaryInfo::hasHugeWorkingSetSize() {
 228   if (!HasHugeWorkingSetSize)
 229     computeThresholds();
 230   return HasHugeWorkingSetSize && HasHugeWorkingSetSize.getValue();
 231 }
 232
 233 bool ProfileSummaryInfo::isHotCount(uint64_t C) {
 234   if (!HotCountThreshold)
 235     computeThresholds();
 236   return HotCountThreshold && C >= HotCountThreshold.getValue();
 237 }
 238
 239 bool ProfileSummaryInfo::isColdCount(uint64_t C) {
 240   if (!ColdCountThreshold)
 241     computeThresholds();
 242   return ColdCountThreshold && C <= ColdCountThreshold.getValue();
 243 }
 244
 245 uint64_t ProfileSummaryInfo::getOrCompHotCountThreshold() {
 246   if (!HotCountThreshold)
 247     computeThresholds();
 248   return HotCountThreshold ? HotCountThreshold.getValue() : UINT64_MAX;
 249 }
 250
 251 uint64_t ProfileSummaryInfo::getOrCompColdCountThreshold() {
 252   if (!ColdCountThreshold)
 253     computeThresholds();
 254   return ColdCountThreshold ? ColdCountThreshold.getValue() : 0;
 255 }
 256
 257 bool ProfileSummaryInfo::isHotBlock(const BasicBlock *BB, BlockFrequencyInfo *BFI) {
 258   auto Count = BFI->getBlockProfileCount(BB);
 259   return Count && isHotCount(*Count);
 260 }
 261
 262 bool ProfileSummaryInfo::isColdBlock(const BasicBlock *BB,
 263                                   BlockFrequencyInfo *BFI) {
 264   auto Count = BFI->getBlockProfileCount(BB);
 265   return Count && isColdCount(*Count);
 266 }
 267
 268 bool ProfileSummaryInfo::isHotCallSite(const CallSite &CS,
 269                                        BlockFrequencyInfo *BFI) {
 270   auto C = getProfileCount(CS.getInstruction(), BFI);
 271   return C && isHotCount(*C);
 272 }
 273
 274 bool ProfileSummaryInfo::isColdCallSite(const CallSite &CS,
 275                                         BlockFrequencyInfo *BFI) {
 276   auto C = getProfileCount(CS.getInstruction(), BFI);
 277   if (C)
 278     return isColdCount(*C);
 279
 280   // In SamplePGO, if the caller has been sampled, and there is no profile
 281   // annotated on the callsite, we consider the callsite as cold.
 282   return hasSampleProfile() && CS.getCaller()->hasProfileData();
 283 }
 284
 285 INITIALIZE_PASS(ProfileSummaryInfoWrapperPass, "profile-summary-info",
 286                 "Profile summary info", false, true)
 287
 288 ProfileSummaryInfoWrapperPass::ProfileSummaryInfoWrapperPass()
 289     : ImmutablePass(ID) {
 290   initializeProfileSummaryInfoWrapperPassPass(*PassRegistry::getPassRegistry());
 291 }
 292
 293 bool ProfileSummaryInfoWrapperPass::doInitialization(Module &M) {
 294   PSI.reset(new ProfileSummaryInfo(M));
 295   return false;
 296 }
 297
 298 bool ProfileSummaryInfoWrapperPass::doFinalization(Module &M) {
 299   PSI.reset();
 300   return false;
 301 }
 302
 303 AnalysisKey ProfileSummaryAnalysis::Key;
 304 ProfileSummaryInfo ProfileSummaryAnalysis::run(Module &M,
 305                                                ModuleAnalysisManager &) {
 306   return ProfileSummaryInfo(M);
 307 }
 308
 309 PreservedAnalyses ProfileSummaryPrinterPass::run(Module &M,
 310                                                  ModuleAnalysisManager &AM) {
 311   ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M);
 312
 313   OS << "Functions in " << M.getName() << " with hot/cold annotations: \n";
 314   for (auto &F : M) {
 315     OS << F.getName();
 316     if (PSI.isFunctionEntryHot(&F))
 317       OS << " :hot entry ";
 318     else if (PSI.isFunctionEntryCold(&F))
 319       OS << " :cold entry ";
 320     OS << "\n";
 321   }
 322   return PreservedAnalyses::all();
 323 }
 324
 325 char ProfileSummaryInfoWrapperPass::ID = 0;