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