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/Metadata.h" 19 #include "llvm/IR/Module.h" 20 #include "llvm/IR/ProfileSummary.h" 21 #include "llvm/InitializePasses.h" 22 #include "llvm/ProfileData/ProfileCommon.h" 23 #include "llvm/Support/CommandLine.h" 24 using namespace llvm; 25 26 // The following two parameters determine the threshold for a count to be 27 // considered hot/cold. These two parameters are percentile values (multiplied 28 // by 10000). If the counts are sorted in descending order, the minimum count to 29 // reach ProfileSummaryCutoffHot gives the threshold to determine a hot count. 30 // Similarly, the minimum count to reach ProfileSummaryCutoffCold gives the 31 // threshold for determining cold count (everything <= this threshold is 32 // considered cold). 33 34 static cl::opt<int> ProfileSummaryCutoffHot( 35 "profile-summary-cutoff-hot", cl::Hidden, cl::init(990000), cl::ZeroOrMore, 36 cl::desc("A count is hot if it exceeds the minimum count to" 37 " reach this percentile of total counts.")); 38 39 static cl::opt<int> ProfileSummaryCutoffCold( 40 "profile-summary-cutoff-cold", cl::Hidden, cl::init(999999), cl::ZeroOrMore, 41 cl::desc("A count is cold if it is below the minimum count" 42 " to reach this percentile of total counts.")); 43 44 static cl::opt<unsigned> ProfileSummaryHugeWorkingSetSizeThreshold( 45 "profile-summary-huge-working-set-size-threshold", cl::Hidden, 46 cl::init(15000), cl::ZeroOrMore, 47 cl::desc("The code working set size is considered huge if the number of" 48 " blocks required to reach the -profile-summary-cutoff-hot" 49 " percentile exceeds this count.")); 50 51 static cl::opt<unsigned> ProfileSummaryLargeWorkingSetSizeThreshold( 52 "profile-summary-large-working-set-size-threshold", cl::Hidden, 53 cl::init(12500), cl::ZeroOrMore, 54 cl::desc("The code working set size is considered large if the number of" 55 " blocks required to reach the -profile-summary-cutoff-hot" 56 " percentile exceeds this count.")); 57 58 // The next two options override the counts derived from summary computation and 59 // are useful for debugging purposes. 60 static cl::opt<int> ProfileSummaryHotCount( 61 "profile-summary-hot-count", cl::ReallyHidden, cl::ZeroOrMore, 62 cl::desc("A fixed hot count that overrides the count derived from" 63 " profile-summary-cutoff-hot")); 64 65 static cl::opt<int> ProfileSummaryColdCount( 66 "profile-summary-cold-count", cl::ReallyHidden, cl::ZeroOrMore, 67 cl::desc("A fixed cold count that overrides the count derived from" 68 " profile-summary-cutoff-cold")); 69 70 static cl::opt<bool> PartialProfile( 71 "partial-profile", cl::Hidden, cl::init(false), 72 cl::desc("Specify the current profile is used as a partial profile.")); 73 74 cl::opt<bool> ScalePartialSampleProfileWorkingSetSize( 75 "scale-partial-sample-profile-working-set-size", cl::Hidden, cl::init(true), 76 cl::desc( 77 "If true, scale the working set size of the partial sample profile " 78 "by the partial profile ratio to reflect the size of the program " 79 "being compiled.")); 80 81 static cl::opt<double> PartialSampleProfileWorkingSetSizeScaleFactor( 82 "partial-sample-profile-working-set-size-scale-factor", cl::Hidden, 83 cl::init(0.008), 84 cl::desc("The scale factor used to scale the working set size of the " 85 "partial sample profile along with the partial profile ratio. " 86 "This includes the factor of the profile counter per block " 87 "and the factor to scale the working set size to use the same " 88 "shared thresholds as PGO.")); 89 90 // The profile summary metadata may be attached either by the frontend or by 91 // any backend passes (IR level instrumentation, for example). This method 92 // checks if the Summary is null and if so checks if the summary metadata is now 93 // available in the module and parses it to get the Summary object. 94 void ProfileSummaryInfo::refresh() { 95 if (hasProfileSummary()) 96 return; 97 // First try to get context sensitive ProfileSummary. 98 auto *SummaryMD = M.getProfileSummary(/* IsCS */ true); 99 if (SummaryMD) 100 Summary.reset(ProfileSummary::getFromMD(SummaryMD)); 101 102 if (!hasProfileSummary()) { 103 // This will actually return PSK_Instr or PSK_Sample summary. 104 SummaryMD = M.getProfileSummary(/* IsCS */ false); 105 if (SummaryMD) 106 Summary.reset(ProfileSummary::getFromMD(SummaryMD)); 107 } 108 if (!hasProfileSummary()) 109 return; 110 computeThresholds(); 111 } 112 113 Optional<uint64_t> ProfileSummaryInfo::getProfileCount( 114 const CallBase &Call, BlockFrequencyInfo *BFI, bool AllowSynthetic) const { 115 assert((isa<CallInst>(Call) || isa<InvokeInst>(Call)) && 116 "We can only get profile count for call/invoke instruction."); 117 if (hasSampleProfile()) { 118 // In sample PGO mode, check if there is a profile metadata on the 119 // instruction. If it is present, determine hotness solely based on that, 120 // since the sampled entry count may not be accurate. If there is no 121 // annotated on the instruction, return None. 122 uint64_t TotalCount; 123 if (Call.extractProfTotalWeight(TotalCount)) 124 return TotalCount; 125 return None; 126 } 127 if (BFI) 128 return BFI->getBlockProfileCount(Call.getParent(), AllowSynthetic); 129 return None; 130 } 131 132 /// Returns true if the function's entry is hot. If it returns false, it 133 /// either means it is not hot or it is unknown whether it is hot or not (for 134 /// example, no profile data is available). 135 bool ProfileSummaryInfo::isFunctionEntryHot(const Function *F) const { 136 if (!F || !hasProfileSummary()) 137 return false; 138 auto FunctionCount = F->getEntryCount(); 139 // FIXME: The heuristic used below for determining hotness is based on 140 // preliminary SPEC tuning for inliner. This will eventually be a 141 // convenience method that calls isHotCount. 142 return FunctionCount && isHotCount(FunctionCount.getCount()); 143 } 144 145 /// Returns true if the function contains hot code. This can include a hot 146 /// function entry count, hot basic block, or (in the case of Sample PGO) 147 /// hot total call edge count. 148 /// If it returns false, it either means it is not hot or it is unknown 149 /// (for example, no profile data is available). 150 bool ProfileSummaryInfo::isFunctionHotInCallGraph( 151 const Function *F, BlockFrequencyInfo &BFI) const { 152 if (!F || !hasProfileSummary()) 153 return false; 154 if (auto FunctionCount = F->getEntryCount()) 155 if (isHotCount(FunctionCount.getCount())) 156 return true; 157 158 if (hasSampleProfile()) { 159 uint64_t TotalCallCount = 0; 160 for (const auto &BB : *F) 161 for (const auto &I : BB) 162 if (isa<CallInst>(I) || isa<InvokeInst>(I)) 163 if (auto CallCount = getProfileCount(cast<CallBase>(I), nullptr)) 164 TotalCallCount += CallCount.getValue(); 165 if (isHotCount(TotalCallCount)) 166 return true; 167 } 168 for (const auto &BB : *F) 169 if (isHotBlock(&BB, &BFI)) 170 return true; 171 return false; 172 } 173 174 /// Returns true if the function only contains cold code. This means that 175 /// the function entry and blocks are all cold, and (in the case of Sample PGO) 176 /// the total call edge count is cold. 177 /// If it returns false, it either means it is not cold or it is unknown 178 /// (for example, no profile data is available). 179 bool ProfileSummaryInfo::isFunctionColdInCallGraph( 180 const Function *F, BlockFrequencyInfo &BFI) const { 181 if (!F || !hasProfileSummary()) 182 return false; 183 if (auto FunctionCount = F->getEntryCount()) 184 if (!isColdCount(FunctionCount.getCount())) 185 return false; 186 187 if (hasSampleProfile()) { 188 uint64_t TotalCallCount = 0; 189 for (const auto &BB : *F) 190 for (const auto &I : BB) 191 if (isa<CallInst>(I) || isa<InvokeInst>(I)) 192 if (auto CallCount = getProfileCount(cast<CallBase>(I), nullptr)) 193 TotalCallCount += CallCount.getValue(); 194 if (!isColdCount(TotalCallCount)) 195 return false; 196 } 197 for (const auto &BB : *F) 198 if (!isColdBlock(&BB, &BFI)) 199 return false; 200 return true; 201 } 202 203 bool ProfileSummaryInfo::isFunctionHotnessUnknown(const Function &F) const { 204 assert(hasPartialSampleProfile() && "Expect partial sample profile"); 205 return !F.getEntryCount().hasValue(); 206 } 207 208 template <bool isHot> 209 bool ProfileSummaryInfo::isFunctionHotOrColdInCallGraphNthPercentile( 210 int PercentileCutoff, const Function *F, BlockFrequencyInfo &BFI) const { 211 if (!F || !hasProfileSummary()) 212 return false; 213 if (auto FunctionCount = F->getEntryCount()) { 214 if (isHot && 215 isHotCountNthPercentile(PercentileCutoff, FunctionCount.getCount())) 216 return true; 217 if (!isHot && 218 !isColdCountNthPercentile(PercentileCutoff, FunctionCount.getCount())) 219 return false; 220 } 221 if (hasSampleProfile()) { 222 uint64_t TotalCallCount = 0; 223 for (const auto &BB : *F) 224 for (const auto &I : BB) 225 if (isa<CallInst>(I) || isa<InvokeInst>(I)) 226 if (auto CallCount = getProfileCount(cast<CallBase>(I), nullptr)) 227 TotalCallCount += CallCount.getValue(); 228 if (isHot && isHotCountNthPercentile(PercentileCutoff, TotalCallCount)) 229 return true; 230 if (!isHot && !isColdCountNthPercentile(PercentileCutoff, TotalCallCount)) 231 return false; 232 } 233 for (const auto &BB : *F) { 234 if (isHot && isHotBlockNthPercentile(PercentileCutoff, &BB, &BFI)) 235 return true; 236 if (!isHot && !isColdBlockNthPercentile(PercentileCutoff, &BB, &BFI)) 237 return false; 238 } 239 return !isHot; 240 } 241 242 // Like isFunctionHotInCallGraph but for a given cutoff. 243 bool ProfileSummaryInfo::isFunctionHotInCallGraphNthPercentile( 244 int PercentileCutoff, const Function *F, BlockFrequencyInfo &BFI) const { 245 return isFunctionHotOrColdInCallGraphNthPercentile<true>( 246 PercentileCutoff, F, BFI); 247 } 248 249 bool ProfileSummaryInfo::isFunctionColdInCallGraphNthPercentile( 250 int PercentileCutoff, const Function *F, BlockFrequencyInfo &BFI) const { 251 return isFunctionHotOrColdInCallGraphNthPercentile<false>( 252 PercentileCutoff, F, BFI); 253 } 254 255 /// Returns true if the function's entry is a cold. If it returns false, it 256 /// either means it is not cold or it is unknown whether it is cold or not (for 257 /// example, no profile data is available). 258 bool ProfileSummaryInfo::isFunctionEntryCold(const Function *F) const { 259 if (!F) 260 return false; 261 if (F->hasFnAttribute(Attribute::Cold)) 262 return true; 263 if (!hasProfileSummary()) 264 return false; 265 auto FunctionCount = F->getEntryCount(); 266 // FIXME: The heuristic used below for determining coldness is based on 267 // preliminary SPEC tuning for inliner. This will eventually be a 268 // convenience method that calls isHotCount. 269 return FunctionCount && isColdCount(FunctionCount.getCount()); 270 } 271 272 /// Compute the hot and cold thresholds. 273 void ProfileSummaryInfo::computeThresholds() { 274 auto &DetailedSummary = Summary->getDetailedSummary(); 275 auto &HotEntry = ProfileSummaryBuilder::getEntryForPercentile( 276 DetailedSummary, ProfileSummaryCutoffHot); 277 HotCountThreshold = HotEntry.MinCount; 278 if (ProfileSummaryHotCount.getNumOccurrences() > 0) 279 HotCountThreshold = ProfileSummaryHotCount; 280 auto &ColdEntry = ProfileSummaryBuilder::getEntryForPercentile( 281 DetailedSummary, ProfileSummaryCutoffCold); 282 ColdCountThreshold = ColdEntry.MinCount; 283 if (ProfileSummaryColdCount.getNumOccurrences() > 0) 284 ColdCountThreshold = ProfileSummaryColdCount; 285 assert(ColdCountThreshold <= HotCountThreshold && 286 "Cold count threshold cannot exceed hot count threshold!"); 287 if (!hasPartialSampleProfile() || !ScalePartialSampleProfileWorkingSetSize) { 288 HasHugeWorkingSetSize = 289 HotEntry.NumCounts > ProfileSummaryHugeWorkingSetSizeThreshold; 290 HasLargeWorkingSetSize = 291 HotEntry.NumCounts > ProfileSummaryLargeWorkingSetSizeThreshold; 292 } else { 293 // Scale the working set size of the partial sample profile to reflect the 294 // size of the program being compiled. 295 double PartialProfileRatio = Summary->getPartialProfileRatio(); 296 uint64_t ScaledHotEntryNumCounts = 297 static_cast<uint64_t>(HotEntry.NumCounts * PartialProfileRatio * 298 PartialSampleProfileWorkingSetSizeScaleFactor); 299 HasHugeWorkingSetSize = 300 ScaledHotEntryNumCounts > ProfileSummaryHugeWorkingSetSizeThreshold; 301 HasLargeWorkingSetSize = 302 ScaledHotEntryNumCounts > ProfileSummaryLargeWorkingSetSizeThreshold; 303 } 304 } 305 306 Optional<uint64_t> 307 ProfileSummaryInfo::computeThreshold(int PercentileCutoff) const { 308 if (!hasProfileSummary()) 309 return None; 310 auto iter = ThresholdCache.find(PercentileCutoff); 311 if (iter != ThresholdCache.end()) { 312 return iter->second; 313 } 314 auto &DetailedSummary = Summary->getDetailedSummary(); 315 auto &Entry = ProfileSummaryBuilder::getEntryForPercentile(DetailedSummary, 316 PercentileCutoff); 317 uint64_t CountThreshold = Entry.MinCount; 318 ThresholdCache[PercentileCutoff] = CountThreshold; 319 return CountThreshold; 320 } 321 322 bool ProfileSummaryInfo::hasHugeWorkingSetSize() const { 323 return HasHugeWorkingSetSize && HasHugeWorkingSetSize.getValue(); 324 } 325 326 bool ProfileSummaryInfo::hasLargeWorkingSetSize() const { 327 return HasLargeWorkingSetSize && HasLargeWorkingSetSize.getValue(); 328 } 329 330 bool ProfileSummaryInfo::isHotCount(uint64_t C) const { 331 return HotCountThreshold && C >= HotCountThreshold.getValue(); 332 } 333 334 bool ProfileSummaryInfo::isColdCount(uint64_t C) const { 335 return ColdCountThreshold && C <= ColdCountThreshold.getValue(); 336 } 337 338 template <bool isHot> 339 bool ProfileSummaryInfo::isHotOrColdCountNthPercentile(int PercentileCutoff, 340 uint64_t C) const { 341 auto CountThreshold = computeThreshold(PercentileCutoff); 342 if (isHot) 343 return CountThreshold && C >= CountThreshold.getValue(); 344 else 345 return CountThreshold && C <= CountThreshold.getValue(); 346 } 347 348 bool ProfileSummaryInfo::isHotCountNthPercentile(int PercentileCutoff, 349 uint64_t C) const { 350 return isHotOrColdCountNthPercentile<true>(PercentileCutoff, C); 351 } 352 353 bool ProfileSummaryInfo::isColdCountNthPercentile(int PercentileCutoff, 354 uint64_t C) const { 355 return isHotOrColdCountNthPercentile<false>(PercentileCutoff, C); 356 } 357 358 uint64_t ProfileSummaryInfo::getOrCompHotCountThreshold() const { 359 return HotCountThreshold ? HotCountThreshold.getValue() : UINT64_MAX; 360 } 361 362 uint64_t ProfileSummaryInfo::getOrCompColdCountThreshold() const { 363 return ColdCountThreshold ? ColdCountThreshold.getValue() : 0; 364 } 365 366 bool ProfileSummaryInfo::isHotBlock(const BasicBlock *BB, 367 BlockFrequencyInfo *BFI) const { 368 auto Count = BFI->getBlockProfileCount(BB); 369 return Count && isHotCount(*Count); 370 } 371 372 bool ProfileSummaryInfo::isColdBlock(const BasicBlock *BB, 373 BlockFrequencyInfo *BFI) const { 374 auto Count = BFI->getBlockProfileCount(BB); 375 return Count && isColdCount(*Count); 376 } 377 378 template <bool isHot> 379 bool ProfileSummaryInfo::isHotOrColdBlockNthPercentile( 380 int PercentileCutoff, const BasicBlock *BB, BlockFrequencyInfo *BFI) const { 381 auto Count = BFI->getBlockProfileCount(BB); 382 if (isHot) 383 return Count && isHotCountNthPercentile(PercentileCutoff, *Count); 384 else 385 return Count && isColdCountNthPercentile(PercentileCutoff, *Count); 386 } 387 388 bool ProfileSummaryInfo::isHotBlockNthPercentile( 389 int PercentileCutoff, const BasicBlock *BB, BlockFrequencyInfo *BFI) const { 390 return isHotOrColdBlockNthPercentile<true>(PercentileCutoff, BB, BFI); 391 } 392 393 bool ProfileSummaryInfo::isColdBlockNthPercentile( 394 int PercentileCutoff, const BasicBlock *BB, BlockFrequencyInfo *BFI) const { 395 return isHotOrColdBlockNthPercentile<false>(PercentileCutoff, BB, BFI); 396 } 397 398 bool ProfileSummaryInfo::isHotCallSite(const CallBase &CB, 399 BlockFrequencyInfo *BFI) const { 400 auto C = getProfileCount(CB, BFI); 401 return C && isHotCount(*C); 402 } 403 404 bool ProfileSummaryInfo::isColdCallSite(const CallBase &CB, 405 BlockFrequencyInfo *BFI) const { 406 auto C = getProfileCount(CB, BFI); 407 if (C) 408 return isColdCount(*C); 409 410 // In SamplePGO, if the caller has been sampled, and there is no profile 411 // annotated on the callsite, we consider the callsite as cold. 412 return hasSampleProfile() && CB.getCaller()->hasProfileData(); 413 } 414 415 bool ProfileSummaryInfo::hasPartialSampleProfile() const { 416 return hasProfileSummary() && 417 Summary->getKind() == ProfileSummary::PSK_Sample && 418 (PartialProfile || Summary->isPartialProfile()); 419 } 420 421 INITIALIZE_PASS(ProfileSummaryInfoWrapperPass, "profile-summary-info", 422 "Profile summary info", false, true) 423 424 ProfileSummaryInfoWrapperPass::ProfileSummaryInfoWrapperPass() 425 : ImmutablePass(ID) { 426 initializeProfileSummaryInfoWrapperPassPass(*PassRegistry::getPassRegistry()); 427 } 428 429 bool ProfileSummaryInfoWrapperPass::doInitialization(Module &M) { 430 PSI.reset(new ProfileSummaryInfo(M)); 431 return false; 432 } 433 434 bool ProfileSummaryInfoWrapperPass::doFinalization(Module &M) { 435 PSI.reset(); 436 return false; 437 } 438 439 AnalysisKey ProfileSummaryAnalysis::Key; 440 ProfileSummaryInfo ProfileSummaryAnalysis::run(Module &M, 441 ModuleAnalysisManager &) { 442 return ProfileSummaryInfo(M); 443 } 444 445 PreservedAnalyses ProfileSummaryPrinterPass::run(Module &M, 446 ModuleAnalysisManager &AM) { 447 ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M); 448 449 OS << "Functions in " << M.getName() << " with hot/cold annotations: \n"; 450 for (auto &F : M) { 451 OS << F.getName(); 452 if (PSI.isFunctionEntryHot(&F)) 453 OS << " :hot entry "; 454 else if (PSI.isFunctionEntryCold(&F)) 455 OS << " :cold entry "; 456 OS << "\n"; 457 } 458 return PreservedAnalyses::all(); 459 } 460 461 char ProfileSummaryInfoWrapperPass::ID = 0; 462