llvm/lib/ProfileData/InstrProf.cpp

   1 //===- InstrProf.cpp - Instrumented profiling format support --------------===//
   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 support for clang's instrumentation based PGO and
  10 // coverage.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "llvm/ProfileData/InstrProf.h"
  15 #include "llvm/ADT/ArrayRef.h"
  16 #include "llvm/ADT/SetVector.h"
  17 #include "llvm/ADT/SmallString.h"
  18 #include "llvm/ADT/SmallVector.h"
  19 #include "llvm/ADT/StringExtras.h"
  20 #include "llvm/ADT/StringRef.h"
  21 #include "llvm/Config/config.h"
  22 #include "llvm/IR/Constant.h"
  23 #include "llvm/IR/Constants.h"
  24 #include "llvm/IR/Function.h"
  25 #include "llvm/IR/GlobalValue.h"
  26 #include "llvm/IR/GlobalVariable.h"
  27 #include "llvm/IR/Instruction.h"
  28 #include "llvm/IR/LLVMContext.h"
  29 #include "llvm/IR/MDBuilder.h"
  30 #include "llvm/IR/Metadata.h"
  31 #include "llvm/IR/Module.h"
  32 #include "llvm/IR/Type.h"
  33 #include "llvm/ProfileData/InstrProfReader.h"
  34 #include "llvm/Support/Casting.h"
  35 #include "llvm/Support/CommandLine.h"
  36 #include "llvm/Support/Compiler.h"
  37 #include "llvm/Support/Compression.h"
  38 #include "llvm/Support/Endian.h"
  39 #include "llvm/Support/Error.h"
  40 #include "llvm/Support/ErrorHandling.h"
  41 #include "llvm/Support/LEB128.h"
  42 #include "llvm/Support/MathExtras.h"
  43 #include "llvm/Support/Path.h"
  44 #include "llvm/Support/SwapByteOrder.h"
  45 #include "llvm/Support/VirtualFileSystem.h"
  46 #include "llvm/TargetParser/Triple.h"
  47 #include <algorithm>
  48 #include <cassert>
  49 #include <cstddef>
  50 #include <cstdint>
  51 #include <cstring>
  52 #include <memory>
  53 #include <string>
  54 #include <system_error>
  55 #include <type_traits>
  56 #include <utility>
  57 #include <vector>
  58
  59 using namespace llvm;
  60
  61 static cl::opt<bool> StaticFuncFullModulePrefix(
  62     "static-func-full-module-prefix", cl::init(true), cl::Hidden,
  63     cl::desc("Use full module build paths in the profile counter names for "
  64              "static functions."));
  65
  66 // This option is tailored to users that have different top-level directory in
  67 // profile-gen and profile-use compilation. Users need to specific the number
  68 // of levels to strip. A value larger than the number of directories in the
  69 // source file will strip all the directory names and only leave the basename.
  70 //
  71 // Note current ThinLTO module importing for the indirect-calls assumes
  72 // the source directory name not being stripped. A non-zero option value here
  73 // can potentially prevent some inter-module indirect-call-promotions.
  74 static cl::opt<unsigned> StaticFuncStripDirNamePrefix(
  75     "static-func-strip-dirname-prefix", cl::init(0), cl::Hidden,
  76     cl::desc("Strip specified level of directory name from source path in "
  77              "the profile counter name for static functions."));
  78
  79 static std::string getInstrProfErrString(instrprof_error Err,
  80                                          const std::string &ErrMsg = "") {
  81   std::string Msg;
  82   raw_string_ostream OS(Msg);
  83
  84   switch (Err) {
  85   case instrprof_error::success:
  86     OS << "success";
  87     break;
  88   case instrprof_error::eof:
  89     OS << "end of File";
  90     break;
  91   case instrprof_error::unrecognized_format:
  92     OS << "unrecognized instrumentation profile encoding format";
  93     break;
  94   case instrprof_error::bad_magic:
  95     OS << "invalid instrumentation profile data (bad magic)";
  96     break;
  97   case instrprof_error::bad_header:
  98     OS << "invalid instrumentation profile data (file header is corrupt)";
  99     break;
 100   case instrprof_error::unsupported_version:
 101     OS << "unsupported instrumentation profile format version";
 102     break;
 103   case instrprof_error::unsupported_hash_type:
 104     OS << "unsupported instrumentation profile hash type";
 105     break;
 106   case instrprof_error::too_large:
 107     OS << "too much profile data";
 108     break;
 109   case instrprof_error::truncated:
 110     OS << "truncated profile data";
 111     break;
 112   case instrprof_error::malformed:
 113     OS << "malformed instrumentation profile data";
 114     break;
 115   case instrprof_error::missing_debug_info_for_correlation:
 116     OS << "debug info for correlation is required";
 117     break;
 118   case instrprof_error::unexpected_debug_info_for_correlation:
 119     OS << "debug info for correlation is not necessary";
 120     break;
 121   case instrprof_error::unable_to_correlate_profile:
 122     OS << "unable to correlate profile";
 123     break;
 124   case instrprof_error::invalid_prof:
 125     OS << "invalid profile created. Please file a bug "
 126           "at: " BUG_REPORT_URL
 127           " and include the profraw files that caused this error.";
 128     break;
 129   case instrprof_error::unknown_function:
 130     OS << "no profile data available for function";
 131     break;
 132   case instrprof_error::hash_mismatch:
 133     OS << "function control flow change detected (hash mismatch)";
 134     break;
 135   case instrprof_error::count_mismatch:
 136     OS << "function basic block count change detected (counter mismatch)";
 137     break;
 138   case instrprof_error::counter_overflow:
 139     OS << "counter overflow";
 140     break;
 141   case instrprof_error::value_site_count_mismatch:
 142     OS << "function value site count change detected (counter mismatch)";
 143     break;
 144   case instrprof_error::compress_failed:
 145     OS << "failed to compress data (zlib)";
 146     break;
 147   case instrprof_error::uncompress_failed:
 148     OS << "failed to uncompress data (zlib)";
 149     break;
 150   case instrprof_error::empty_raw_profile:
 151     OS << "empty raw profile file";
 152     break;
 153   case instrprof_error::zlib_unavailable:
 154     OS << "profile uses zlib compression but the profile reader was built "
 155           "without zlib support";
 156     break;
 157   case instrprof_error::raw_profile_version_mismatch:
 158     OS << "raw profile version mismatch";
 159     break;
 160   }
 161
 162   // If optional error message is not empty, append it to the message.
 163   if (!ErrMsg.empty())
 164     OS << ": " << ErrMsg;
 165
 166   return OS.str();
 167 }
 168
 169 namespace {
 170
 171 // FIXME: This class is only here to support the transition to llvm::Error. It
 172 // will be removed once this transition is complete. Clients should prefer to
 173 // deal with the Error value directly, rather than converting to error_code.
 174 class InstrProfErrorCategoryType : public std::error_category {
 175   const char *name() const noexcept override { return "llvm.instrprof"; }
 176
 177   std::string message(int IE) const override {
 178     return getInstrProfErrString(static_cast<instrprof_error>(IE));
 179   }
 180 };
 181
 182 } // end anonymous namespace
 183
 184 const std::error_category &llvm::instrprof_category() {
 185   static InstrProfErrorCategoryType ErrorCategory;
 186   return ErrorCategory;
 187 }
 188
 189 namespace {
 190
 191 const char *InstrProfSectNameCommon[] = {
 192 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
 193   SectNameCommon,
 194 #include "llvm/ProfileData/InstrProfData.inc"
 195 };
 196
 197 const char *InstrProfSectNameCoff[] = {
 198 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
 199   SectNameCoff,
 200 #include "llvm/ProfileData/InstrProfData.inc"
 201 };
 202
 203 const char *InstrProfSectNamePrefix[] = {
 204 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
 205   Prefix,
 206 #include "llvm/ProfileData/InstrProfData.inc"
 207 };
 208
 209 } // namespace
 210
 211 namespace llvm {
 212
 213 cl::opt<bool> DoInstrProfNameCompression(
 214     "enable-name-compression",
 215     cl::desc("Enable name/filename string compression"), cl::init(true));
 216
 217 std::string getInstrProfSectionName(InstrProfSectKind IPSK,
 218                                     Triple::ObjectFormatType OF,
 219                                     bool AddSegmentInfo) {
 220   std::string SectName;
 221
 222   if (OF == Triple::MachO && AddSegmentInfo)
 223     SectName = InstrProfSectNamePrefix[IPSK];
 224
 225   if (OF == Triple::COFF)
 226     SectName += InstrProfSectNameCoff[IPSK];
 227   else
 228     SectName += InstrProfSectNameCommon[IPSK];
 229
 230   if (OF == Triple::MachO && IPSK == IPSK_data && AddSegmentInfo)
 231     SectName += ",regular,live_support";
 232
 233   return SectName;
 234 }
 235
 236 std::string InstrProfError::message() const {
 237   return getInstrProfErrString(Err, Msg);
 238 }
 239
 240 char InstrProfError::ID = 0;
 241
 242 std::string getPGOFuncName(StringRef RawFuncName,
 243                            GlobalValue::LinkageTypes Linkage,
 244                            StringRef FileName,
 245                            uint64_t Version LLVM_ATTRIBUTE_UNUSED) {
 246   return GlobalValue::getGlobalIdentifier(RawFuncName, Linkage, FileName);
 247 }
 248
 249 // Strip NumPrefix level of directory name from PathNameStr. If the number of
 250 // directory separators is less than NumPrefix, strip all the directories and
 251 // leave base file name only.
 252 static StringRef stripDirPrefix(StringRef PathNameStr, uint32_t NumPrefix) {
 253   uint32_t Count = NumPrefix;
 254   uint32_t Pos = 0, LastPos = 0;
 255   for (auto & CI : PathNameStr) {
 256     ++Pos;
 257     if (llvm::sys::path::is_separator(CI)) {
 258       LastPos = Pos;
 259       --Count;
 260     }
 261     if (Count == 0)
 262       break;
 263   }
 264   return PathNameStr.substr(LastPos);
 265 }
 266
 267 // Return the PGOFuncName. This function has some special handling when called
 268 // in LTO optimization. The following only applies when calling in LTO passes
 269 // (when \c InLTO is true): LTO's internalization privatizes many global linkage
 270 // symbols. This happens after value profile annotation, but those internal
 271 // linkage functions should not have a source prefix.
 272 // Additionally, for ThinLTO mode, exported internal functions are promoted
 273 // and renamed. We need to ensure that the original internal PGO name is
 274 // used when computing the GUID that is compared against the profiled GUIDs.
 275 // To differentiate compiler generated internal symbols from original ones,
 276 // PGOFuncName meta data are created and attached to the original internal
 277 // symbols in the value profile annotation step
 278 // (PGOUseFunc::annotateIndirectCallSites). If a symbol does not have the meta
 279 // data, its original linkage must be non-internal.
 280 std::string getPGOFuncName(const Function &F, bool InLTO, uint64_t Version) {
 281   if (!InLTO) {
 282     StringRef FileName(F.getParent()->getSourceFileName());
 283     uint32_t StripLevel = StaticFuncFullModulePrefix ? 0 : (uint32_t)-1;
 284     if (StripLevel < StaticFuncStripDirNamePrefix)
 285       StripLevel = StaticFuncStripDirNamePrefix;
 286     if (StripLevel)
 287       FileName = stripDirPrefix(FileName, StripLevel);
 288     return getPGOFuncName(F.getName(), F.getLinkage(), FileName, Version);
 289   }
 290
 291   // In LTO mode (when InLTO is true), first check if there is a meta data.
 292   if (MDNode *MD = getPGOFuncNameMetadata(F)) {
 293     StringRef S = cast<MDString>(MD->getOperand(0))->getString();
 294     return S.str();
 295   }
 296
 297   // If there is no meta data, the function must be a global before the value
 298   // profile annotation pass. Its current linkage may be internal if it is
 299   // internalized in LTO mode.
 300   return getPGOFuncName(F.getName(), GlobalValue::ExternalLinkage, "");
 301 }
 302
 303 StringRef getFuncNameWithoutPrefix(StringRef PGOFuncName, StringRef FileName) {
 304   if (FileName.empty())
 305     return PGOFuncName;
 306   // Drop the file name including ':'. See also getPGOFuncName.
 307   if (PGOFuncName.startswith(FileName))
 308     PGOFuncName = PGOFuncName.drop_front(FileName.size() + 1);
 309   return PGOFuncName;
 310 }
 311
 312 // \p FuncName is the string used as profile lookup key for the function. A
 313 // symbol is created to hold the name. Return the legalized symbol name.
 314 std::string getPGOFuncNameVarName(StringRef FuncName,
 315                                   GlobalValue::LinkageTypes Linkage) {
 316   std::string VarName = std::string(getInstrProfNameVarPrefix());
 317   VarName += FuncName;
 318
 319   if (!GlobalValue::isLocalLinkage(Linkage))
 320     return VarName;
 321
 322   // Now fix up illegal chars in local VarName that may upset the assembler.
 323   const char *InvalidChars = "-:<>/\"'";
 324   size_t found = VarName.find_first_of(InvalidChars);
 325   while (found != std::string::npos) {
 326     VarName[found] = '_';
 327     found = VarName.find_first_of(InvalidChars, found + 1);
 328   }
 329   return VarName;
 330 }
 331
 332 GlobalVariable *createPGOFuncNameVar(Module &M,
 333                                      GlobalValue::LinkageTypes Linkage,
 334                                      StringRef PGOFuncName) {
 335   // We generally want to match the function's linkage, but available_externally
 336   // and extern_weak both have the wrong semantics, and anything that doesn't
 337   // need to link across compilation units doesn't need to be visible at all.
 338   if (Linkage == GlobalValue::ExternalWeakLinkage)
 339     Linkage = GlobalValue::LinkOnceAnyLinkage;
 340   else if (Linkage == GlobalValue::AvailableExternallyLinkage)
 341     Linkage = GlobalValue::LinkOnceODRLinkage;
 342   else if (Linkage == GlobalValue::InternalLinkage ||
 343            Linkage == GlobalValue::ExternalLinkage)
 344     Linkage = GlobalValue::PrivateLinkage;
 345
 346   auto *Value =
 347       ConstantDataArray::getString(M.getContext(), PGOFuncName, false);
 348   auto FuncNameVar =
 349       new GlobalVariable(M, Value->getType(), true, Linkage, Value,
 350                          getPGOFuncNameVarName(PGOFuncName, Linkage));
 351
 352   // Hide the symbol so that we correctly get a copy for each executable.
 353   if (!GlobalValue::isLocalLinkage(FuncNameVar->getLinkage()))
 354     FuncNameVar->setVisibility(GlobalValue::HiddenVisibility);
 355
 356   return FuncNameVar;
 357 }
 358
 359 GlobalVariable *createPGOFuncNameVar(Function &F, StringRef PGOFuncName) {
 360   return createPGOFuncNameVar(*F.getParent(), F.getLinkage(), PGOFuncName);
 361 }
 362
 363 Error InstrProfSymtab::create(Module &M, bool InLTO) {
 364   for (Function &F : M) {
 365     // Function may not have a name: like using asm("") to overwrite the name.
 366     // Ignore in this case.
 367     if (!F.hasName())
 368       continue;
 369     const std::string &PGOFuncName = getPGOFuncName(F, InLTO);
 370     if (Error E = addFuncName(PGOFuncName))
 371       return E;
 372     MD5FuncMap.emplace_back(Function::getGUID(PGOFuncName), &F);
 373     // In ThinLTO, local function may have been promoted to global and have
 374     // suffix ".llvm." added to the function name. We need to add the
 375     // stripped function name to the symbol table so that we can find a match
 376     // from profile.
 377     //
 378     // We may have other suffixes similar as ".llvm." which are needed to
 379     // be stripped before the matching, but ".__uniq." suffix which is used
 380     // to differentiate internal linkage functions in different modules
 381     // should be kept. Now this is the only suffix with the pattern ".xxx"
 382     // which is kept before matching.
 383     const std::string UniqSuffix = ".__uniq.";
 384     auto pos = PGOFuncName.find(UniqSuffix);
 385     // Search '.' after ".__uniq." if ".__uniq." exists, otherwise
 386     // search '.' from the beginning.
 387     if (pos != std::string::npos)
 388       pos += UniqSuffix.length();
 389     else
 390       pos = 0;
 391     pos = PGOFuncName.find('.', pos);
 392     if (pos != std::string::npos && pos != 0) {
 393       const std::string &OtherFuncName = PGOFuncName.substr(0, pos);
 394       if (Error E = addFuncName(OtherFuncName))
 395         return E;
 396       MD5FuncMap.emplace_back(Function::getGUID(OtherFuncName), &F);
 397     }
 398   }
 399   Sorted = false;
 400   finalizeSymtab();
 401   return Error::success();
 402 }
 403
 404 uint64_t InstrProfSymtab::getFunctionHashFromAddress(uint64_t Address) {
 405   finalizeSymtab();
 406   auto It = partition_point(AddrToMD5Map, [=](std::pair<uint64_t, uint64_t> A) {
 407     return A.first < Address;
 408   });
 409   // Raw function pointer collected by value profiler may be from
 410   // external functions that are not instrumented. They won't have
 411   // mapping data to be used by the deserializer. Force the value to
 412   // be 0 in this case.
 413   if (It != AddrToMD5Map.end() && It->first == Address)
 414     return (uint64_t)It->second;
 415   return 0;
 416 }
 417
 418 Error collectPGOFuncNameStrings(ArrayRef<std::string> NameStrs,
 419                                 bool doCompression, std::string &Result) {
 420   assert(!NameStrs.empty() && "No name data to emit");
 421
 422   uint8_t Header[16], *P = Header;
 423   std::string UncompressedNameStrings =
 424       join(NameStrs.begin(), NameStrs.end(), getInstrProfNameSeparator());
 425
 426   assert(StringRef(UncompressedNameStrings)
 427                  .count(getInstrProfNameSeparator()) == (NameStrs.size() - 1) &&
 428          "PGO name is invalid (contains separator token)");
 429
 430   unsigned EncLen = encodeULEB128(UncompressedNameStrings.length(), P);
 431   P += EncLen;
 432
 433   auto WriteStringToResult = [&](size_t CompressedLen, StringRef InputStr) {
 434     EncLen = encodeULEB128(CompressedLen, P);
 435     P += EncLen;
 436     char *HeaderStr = reinterpret_cast<char *>(&Header[0]);
 437     unsigned HeaderLen = P - &Header[0];
 438     Result.append(HeaderStr, HeaderLen);
 439     Result += InputStr;
 440     return Error::success();
 441   };
 442
 443   if (!doCompression) {
 444     return WriteStringToResult(0, UncompressedNameStrings);
 445   }
 446
 447   SmallVector<uint8_t, 128> CompressedNameStrings;
 448   compression::zlib::compress(arrayRefFromStringRef(UncompressedNameStrings),
 449                               CompressedNameStrings,
 450                               compression::zlib::BestSizeCompression);
 451
 452   return WriteStringToResult(CompressedNameStrings.size(),
 453                              toStringRef(CompressedNameStrings));
 454 }
 455
 456 StringRef getPGOFuncNameVarInitializer(GlobalVariable *NameVar) {
 457   auto *Arr = cast<ConstantDataArray>(NameVar->getInitializer());
 458   StringRef NameStr =
 459       Arr->isCString() ? Arr->getAsCString() : Arr->getAsString();
 460   return NameStr;
 461 }
 462
 463 Error collectPGOFuncNameStrings(ArrayRef<GlobalVariable *> NameVars,
 464                                 std::string &Result, bool doCompression) {
 465   std::vector<std::string> NameStrs;
 466   for (auto *NameVar : NameVars) {
 467     NameStrs.push_back(std::string(getPGOFuncNameVarInitializer(NameVar)));
 468   }
 469   return collectPGOFuncNameStrings(
 470       NameStrs, compression::zlib::isAvailable() && doCompression, Result);
 471 }
 472
 473 Error readPGOFuncNameStrings(StringRef NameStrings, InstrProfSymtab &Symtab) {
 474   const uint8_t *P = NameStrings.bytes_begin();
 475   const uint8_t *EndP = NameStrings.bytes_end();
 476   while (P < EndP) {
 477     uint32_t N;
 478     uint64_t UncompressedSize = decodeULEB128(P, &N);
 479     P += N;
 480     uint64_t CompressedSize = decodeULEB128(P, &N);
 481     P += N;
 482     bool isCompressed = (CompressedSize != 0);
 483     SmallVector<uint8_t, 128> UncompressedNameStrings;
 484     StringRef NameStrings;
 485     if (isCompressed) {
 486       if (!llvm::compression::zlib::isAvailable())
 487         return make_error<InstrProfError>(instrprof_error::zlib_unavailable);
 488
 489       if (Error E = compression::zlib::decompress(ArrayRef(P, CompressedSize),
 490                                                   UncompressedNameStrings,
 491                                                   UncompressedSize)) {
 492         consumeError(std::move(E));
 493         return make_error<InstrProfError>(instrprof_error::uncompress_failed);
 494       }
 495       P += CompressedSize;
 496       NameStrings = toStringRef(UncompressedNameStrings);
 497     } else {
 498       NameStrings =
 499           StringRef(reinterpret_cast<const char *>(P), UncompressedSize);
 500       P += UncompressedSize;
 501     }
 502     // Now parse the name strings.
 503     SmallVector<StringRef, 0> Names;
 504     NameStrings.split(Names, getInstrProfNameSeparator());
 505     for (StringRef &Name : Names)
 506       if (Error E = Symtab.addFuncName(Name))
 507         return E;
 508
 509     while (P < EndP && *P == 0)
 510       P++;
 511   }
 512   return Error::success();
 513 }
 514
 515 void InstrProfRecord::accumulateCounts(CountSumOrPercent &Sum) const {
 516   uint64_t FuncSum = 0;
 517   Sum.NumEntries += Counts.size();
 518   for (uint64_t Count : Counts)
 519     FuncSum += Count;
 520   Sum.CountSum += FuncSum;
 521
 522   for (uint32_t VK = IPVK_First; VK <= IPVK_Last; ++VK) {
 523     uint64_t KindSum = 0;
 524     uint32_t NumValueSites = getNumValueSites(VK);
 525     for (size_t I = 0; I < NumValueSites; ++I) {
 526       uint32_t NV = getNumValueDataForSite(VK, I);
 527       std::unique_ptr<InstrProfValueData[]> VD = getValueForSite(VK, I);
 528       for (uint32_t V = 0; V < NV; V++)
 529         KindSum += VD[V].Count;
 530     }
 531     Sum.ValueCounts[VK] += KindSum;
 532   }
 533 }
 534
 535 void InstrProfValueSiteRecord::overlap(InstrProfValueSiteRecord &Input,
 536                                        uint32_t ValueKind,
 537                                        OverlapStats &Overlap,
 538                                        OverlapStats &FuncLevelOverlap) {
 539   this->sortByTargetValues();
 540   Input.sortByTargetValues();
 541   double Score = 0.0f, FuncLevelScore = 0.0f;
 542   auto I = ValueData.begin();
 543   auto IE = ValueData.end();
 544   auto J = Input.ValueData.begin();
 545   auto JE = Input.ValueData.end();
 546   while (I != IE && J != JE) {
 547     if (I->Value == J->Value) {
 548       Score += OverlapStats::score(I->Count, J->Count,
 549                                    Overlap.Base.ValueCounts[ValueKind],
 550                                    Overlap.Test.ValueCounts[ValueKind]);
 551       FuncLevelScore += OverlapStats::score(
 552           I->Count, J->Count, FuncLevelOverlap.Base.ValueCounts[ValueKind],
 553           FuncLevelOverlap.Test.ValueCounts[ValueKind]);
 554       ++I;
 555     } else if (I->Value < J->Value) {
 556       ++I;
 557       continue;
 558     }
 559     ++J;
 560   }
 561   Overlap.Overlap.ValueCounts[ValueKind] += Score;
 562   FuncLevelOverlap.Overlap.ValueCounts[ValueKind] += FuncLevelScore;
 563 }
 564
 565 // Return false on mismatch.
 566 void InstrProfRecord::overlapValueProfData(uint32_t ValueKind,
 567                                            InstrProfRecord &Other,
 568                                            OverlapStats &Overlap,
 569                                            OverlapStats &FuncLevelOverlap) {
 570   uint32_t ThisNumValueSites = getNumValueSites(ValueKind);
 571   assert(ThisNumValueSites == Other.getNumValueSites(ValueKind));
 572   if (!ThisNumValueSites)
 573     return;
 574
 575   std::vector<InstrProfValueSiteRecord> &ThisSiteRecords =
 576       getOrCreateValueSitesForKind(ValueKind);
 577   MutableArrayRef<InstrProfValueSiteRecord> OtherSiteRecords =
 578       Other.getValueSitesForKind(ValueKind);
 579   for (uint32_t I = 0; I < ThisNumValueSites; I++)
 580     ThisSiteRecords[I].overlap(OtherSiteRecords[I], ValueKind, Overlap,
 581                                FuncLevelOverlap);
 582 }
 583
 584 void InstrProfRecord::overlap(InstrProfRecord &Other, OverlapStats &Overlap,
 585                               OverlapStats &FuncLevelOverlap,
 586                               uint64_t ValueCutoff) {
 587   // FuncLevel CountSum for other should already computed and nonzero.
 588   assert(FuncLevelOverlap.Test.CountSum >= 1.0f);
 589   accumulateCounts(FuncLevelOverlap.Base);
 590   bool Mismatch = (Counts.size() != Other.Counts.size());
 591
 592   // Check if the value profiles mismatch.
 593   if (!Mismatch) {
 594     for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind) {
 595       uint32_t ThisNumValueSites = getNumValueSites(Kind);
 596       uint32_t OtherNumValueSites = Other.getNumValueSites(Kind);
 597       if (ThisNumValueSites != OtherNumValueSites) {
 598         Mismatch = true;
 599         break;
 600       }
 601     }
 602   }
 603   if (Mismatch) {
 604     Overlap.addOneMismatch(FuncLevelOverlap.Test);
 605     return;
 606   }
 607
 608   // Compute overlap for value counts.
 609   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
 610     overlapValueProfData(Kind, Other, Overlap, FuncLevelOverlap);
 611
 612   double Score = 0.0;
 613   uint64_t MaxCount = 0;
 614   // Compute overlap for edge counts.
 615   for (size_t I = 0, E = Other.Counts.size(); I < E; ++I) {
 616     Score += OverlapStats::score(Counts[I], Other.Counts[I],
 617                                  Overlap.Base.CountSum, Overlap.Test.CountSum);
 618     MaxCount = std::max(Other.Counts[I], MaxCount);
 619   }
 620   Overlap.Overlap.CountSum += Score;
 621   Overlap.Overlap.NumEntries += 1;
 622
 623   if (MaxCount >= ValueCutoff) {
 624     double FuncScore = 0.0;
 625     for (size_t I = 0, E = Other.Counts.size(); I < E; ++I)
 626       FuncScore += OverlapStats::score(Counts[I], Other.Counts[I],
 627                                        FuncLevelOverlap.Base.CountSum,
 628                                        FuncLevelOverlap.Test.CountSum);
 629     FuncLevelOverlap.Overlap.CountSum = FuncScore;
 630     FuncLevelOverlap.Overlap.NumEntries = Other.Counts.size();
 631     FuncLevelOverlap.Valid = true;
 632   }
 633 }
 634
 635 void InstrProfValueSiteRecord::merge(InstrProfValueSiteRecord &Input,
 636                                      uint64_t Weight,
 637                                      function_ref<void(instrprof_error)> Warn) {
 638   this->sortByTargetValues();
 639   Input.sortByTargetValues();
 640   auto I = ValueData.begin();
 641   auto IE = ValueData.end();
 642   for (const InstrProfValueData &J : Input.ValueData) {
 643     while (I != IE && I->Value < J.Value)
 644       ++I;
 645     if (I != IE && I->Value == J.Value) {
 646       bool Overflowed;
 647       I->Count = SaturatingMultiplyAdd(J.Count, Weight, I->Count, &Overflowed);
 648       if (Overflowed)
 649         Warn(instrprof_error::counter_overflow);
 650       ++I;
 651       continue;
 652     }
 653     ValueData.insert(I, J);
 654   }
 655 }
 656
 657 void InstrProfValueSiteRecord::scale(uint64_t N, uint64_t D,
 658                                      function_ref<void(instrprof_error)> Warn) {
 659   for (InstrProfValueData &I : ValueData) {
 660     bool Overflowed;
 661     I.Count = SaturatingMultiply(I.Count, N, &Overflowed) / D;
 662     if (Overflowed)
 663       Warn(instrprof_error::counter_overflow);
 664   }
 665 }
 666
 667 // Merge Value Profile data from Src record to this record for ValueKind.
 668 // Scale merged value counts by \p Weight.
 669 void InstrProfRecord::mergeValueProfData(
 670     uint32_t ValueKind, InstrProfRecord &Src, uint64_t Weight,
 671     function_ref<void(instrprof_error)> Warn) {
 672   uint32_t ThisNumValueSites = getNumValueSites(ValueKind);
 673   uint32_t OtherNumValueSites = Src.getNumValueSites(ValueKind);
 674   if (ThisNumValueSites != OtherNumValueSites) {
 675     Warn(instrprof_error::value_site_count_mismatch);
 676     return;
 677   }
 678   if (!ThisNumValueSites)
 679     return;
 680   std::vector<InstrProfValueSiteRecord> &ThisSiteRecords =
 681       getOrCreateValueSitesForKind(ValueKind);
 682   MutableArrayRef<InstrProfValueSiteRecord> OtherSiteRecords =
 683       Src.getValueSitesForKind(ValueKind);
 684   for (uint32_t I = 0; I < ThisNumValueSites; I++)
 685     ThisSiteRecords[I].merge(OtherSiteRecords[I], Weight, Warn);
 686 }
 687
 688 void InstrProfRecord::merge(InstrProfRecord &Other, uint64_t Weight,
 689                             function_ref<void(instrprof_error)> Warn) {
 690   // If the number of counters doesn't match we either have bad data
 691   // or a hash collision.
 692   if (Counts.size() != Other.Counts.size()) {
 693     Warn(instrprof_error::count_mismatch);
 694     return;
 695   }
 696
 697   // Special handling of the first count as the PseudoCount.
 698   CountPseudoKind OtherKind = Other.getCountPseudoKind();
 699   CountPseudoKind ThisKind = getCountPseudoKind();
 700   if (OtherKind != NotPseudo || ThisKind != NotPseudo) {
 701     // We don't allow the merge of a profile with pseudo counts and
 702     // a normal profile (i.e. without pesudo counts).
 703     // Profile supplimenation should be done after the profile merge.
 704     if (OtherKind == NotPseudo || ThisKind == NotPseudo) {
 705       Warn(instrprof_error::count_mismatch);
 706       return;
 707     }
 708     if (OtherKind == PseudoHot || ThisKind == PseudoHot)
 709       setPseudoCount(PseudoHot);
 710     else
 711       setPseudoCount(PseudoWarm);
 712     return;
 713   }
 714
 715   for (size_t I = 0, E = Other.Counts.size(); I < E; ++I) {
 716     bool Overflowed;
 717     uint64_t Value =
 718         SaturatingMultiplyAdd(Other.Counts[I], Weight, Counts[I], &Overflowed);
 719     if (Value > getInstrMaxCountValue()) {
 720       Value = getInstrMaxCountValue();
 721       Overflowed = true;
 722     }
 723     Counts[I] = Value;
 724     if (Overflowed)
 725       Warn(instrprof_error::counter_overflow);
 726   }
 727
 728   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
 729     mergeValueProfData(Kind, Other, Weight, Warn);
 730 }
 731
 732 void InstrProfRecord::scaleValueProfData(
 733     uint32_t ValueKind, uint64_t N, uint64_t D,
 734     function_ref<void(instrprof_error)> Warn) {
 735   for (auto &R : getValueSitesForKind(ValueKind))
 736     R.scale(N, D, Warn);
 737 }
 738
 739 void InstrProfRecord::scale(uint64_t N, uint64_t D,
 740                             function_ref<void(instrprof_error)> Warn) {
 741   assert(D != 0 && "D cannot be 0");
 742   for (auto &Count : this->Counts) {
 743     bool Overflowed;
 744     Count = SaturatingMultiply(Count, N, &Overflowed) / D;
 745     if (Count > getInstrMaxCountValue()) {
 746       Count = getInstrMaxCountValue();
 747       Overflowed = true;
 748     }
 749     if (Overflowed)
 750       Warn(instrprof_error::counter_overflow);
 751   }
 752   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
 753     scaleValueProfData(Kind, N, D, Warn);
 754 }
 755
 756 // Map indirect call target name hash to name string.
 757 uint64_t InstrProfRecord::remapValue(uint64_t Value, uint32_t ValueKind,
 758                                      InstrProfSymtab *SymTab) {
 759   if (!SymTab)
 760     return Value;
 761
 762   if (ValueKind == IPVK_IndirectCallTarget)
 763     return SymTab->getFunctionHashFromAddress(Value);
 764
 765   return Value;
 766 }
 767
 768 void InstrProfRecord::addValueData(uint32_t ValueKind, uint32_t Site,
 769                                    InstrProfValueData *VData, uint32_t N,
 770                                    InstrProfSymtab *ValueMap) {
 771   for (uint32_t I = 0; I < N; I++) {
 772     VData[I].Value = remapValue(VData[I].Value, ValueKind, ValueMap);
 773   }
 774   std::vector<InstrProfValueSiteRecord> &ValueSites =
 775       getOrCreateValueSitesForKind(ValueKind);
 776   if (N == 0)
 777     ValueSites.emplace_back();
 778   else
 779     ValueSites.emplace_back(VData, VData + N);
 780 }
 781
 782 std::vector<BPFunctionNode> TemporalProfTraceTy::createBPFunctionNodes(
 783     ArrayRef<TemporalProfTraceTy> Traces) {
 784   using IDT = BPFunctionNode::IDT;
 785   using UtilityNodeT = BPFunctionNode::UtilityNodeT;
 786   // Collect all function IDs ordered by their smallest timestamp. This will be
 787   // used as the initial FunctionNode order.
 788   SetVector<IDT> FunctionIds;
 789   size_t LargestTraceSize = 0;
 790   for (auto &Trace : Traces)
 791     LargestTraceSize =
 792         std::max(LargestTraceSize, Trace.FunctionNameRefs.size());
 793   for (size_t Timestamp = 0; Timestamp < LargestTraceSize; Timestamp++)
 794     for (auto &Trace : Traces)
 795       if (Timestamp < Trace.FunctionNameRefs.size())
 796         FunctionIds.insert(Trace.FunctionNameRefs[Timestamp]);
 797
 798   int N = std::ceil(std::log2(LargestTraceSize));
 799
 800   // TODO: We need to use the Trace.Weight field to give more weight to more
 801   // important utilities
 802   DenseMap<IDT, SmallVector<UtilityNodeT, 4>> FuncGroups;
 803   for (size_t TraceIdx = 0; TraceIdx < Traces.size(); TraceIdx++) {
 804     auto &Trace = Traces[TraceIdx].FunctionNameRefs;
 805     for (size_t Timestamp = 0; Timestamp < Trace.size(); Timestamp++) {
 806       for (int I = std::floor(std::log2(Timestamp + 1)); I < N; I++) {
 807         auto &FunctionId = Trace[Timestamp];
 808         UtilityNodeT GroupId = TraceIdx * N + I;
 809         FuncGroups[FunctionId].push_back(GroupId);
 810       }
 811     }
 812   }
 813
 814   std::vector<BPFunctionNode> Nodes;
 815   for (auto &Id : FunctionIds) {
 816     auto &UNs = FuncGroups[Id];
 817     llvm::sort(UNs);
 818     UNs.erase(std::unique(UNs.begin(), UNs.end()), UNs.end());
 819     Nodes.emplace_back(Id, UNs);
 820   }
 821   return Nodes;
 822 }
 823
 824 #define INSTR_PROF_COMMON_API_IMPL
 825 #include "llvm/ProfileData/InstrProfData.inc"
 826
 827 /*!
 828  * ValueProfRecordClosure Interface implementation for  InstrProfRecord
 829  *  class. These C wrappers are used as adaptors so that C++ code can be
 830  *  invoked as callbacks.
 831  */
 832 uint32_t getNumValueKindsInstrProf(const void *Record) {
 833   return reinterpret_cast<const InstrProfRecord *>(Record)->getNumValueKinds();
 834 }
 835
 836 uint32_t getNumValueSitesInstrProf(const void *Record, uint32_t VKind) {
 837   return reinterpret_cast<const InstrProfRecord *>(Record)
 838       ->getNumValueSites(VKind);
 839 }
 840
 841 uint32_t getNumValueDataInstrProf(const void *Record, uint32_t VKind) {
 842   return reinterpret_cast<const InstrProfRecord *>(Record)
 843       ->getNumValueData(VKind);
 844 }
 845
 846 uint32_t getNumValueDataForSiteInstrProf(const void *R, uint32_t VK,
 847                                          uint32_t S) {
 848   return reinterpret_cast<const InstrProfRecord *>(R)
 849       ->getNumValueDataForSite(VK, S);
 850 }
 851
 852 void getValueForSiteInstrProf(const void *R, InstrProfValueData *Dst,
 853                               uint32_t K, uint32_t S) {
 854   reinterpret_cast<const InstrProfRecord *>(R)->getValueForSite(Dst, K, S);
 855 }
 856
 857 ValueProfData *allocValueProfDataInstrProf(size_t TotalSizeInBytes) {
 858   ValueProfData *VD =
 859       (ValueProfData *)(new (::operator new(TotalSizeInBytes)) ValueProfData());
 860   memset(VD, 0, TotalSizeInBytes);
 861   return VD;
 862 }
 863
 864 static ValueProfRecordClosure InstrProfRecordClosure = {
 865     nullptr,
 866     getNumValueKindsInstrProf,
 867     getNumValueSitesInstrProf,
 868     getNumValueDataInstrProf,
 869     getNumValueDataForSiteInstrProf,
 870     nullptr,
 871     getValueForSiteInstrProf,
 872     allocValueProfDataInstrProf};
 873
 874 // Wrapper implementation using the closure mechanism.
 875 uint32_t ValueProfData::getSize(const InstrProfRecord &Record) {
 876   auto Closure = InstrProfRecordClosure;
 877   Closure.Record = &Record;
 878   return getValueProfDataSize(&Closure);
 879 }
 880
 881 // Wrapper implementation using the closure mechanism.
 882 std::unique_ptr<ValueProfData>
 883 ValueProfData::serializeFrom(const InstrProfRecord &Record) {
 884   InstrProfRecordClosure.Record = &Record;
 885
 886   std::unique_ptr<ValueProfData> VPD(
 887       serializeValueProfDataFrom(&InstrProfRecordClosure, nullptr));
 888   return VPD;
 889 }
 890
 891 void ValueProfRecord::deserializeTo(InstrProfRecord &Record,
 892                                     InstrProfSymtab *SymTab) {
 893   Record.reserveSites(Kind, NumValueSites);
 894
 895   InstrProfValueData *ValueData = getValueProfRecordValueData(this);
 896   for (uint64_t VSite = 0; VSite < NumValueSites; ++VSite) {
 897     uint8_t ValueDataCount = this->SiteCountArray[VSite];
 898     Record.addValueData(Kind, VSite, ValueData, ValueDataCount, SymTab);
 899     ValueData += ValueDataCount;
 900   }
 901 }
 902
 903 // For writing/serializing,  Old is the host endianness, and  New is
 904 // byte order intended on disk. For Reading/deserialization, Old
 905 // is the on-disk source endianness, and New is the host endianness.
 906 void ValueProfRecord::swapBytes(support::endianness Old,
 907                                 support::endianness New) {
 908   using namespace support;
 909
 910   if (Old == New)
 911     return;
 912
 913   if (getHostEndianness() != Old) {
 914     sys::swapByteOrder<uint32_t>(NumValueSites);
 915     sys::swapByteOrder<uint32_t>(Kind);
 916   }
 917   uint32_t ND = getValueProfRecordNumValueData(this);
 918   InstrProfValueData *VD = getValueProfRecordValueData(this);
 919
 920   // No need to swap byte array: SiteCountArrray.
 921   for (uint32_t I = 0; I < ND; I++) {
 922     sys::swapByteOrder<uint64_t>(VD[I].Value);
 923     sys::swapByteOrder<uint64_t>(VD[I].Count);
 924   }
 925   if (getHostEndianness() == Old) {
 926     sys::swapByteOrder<uint32_t>(NumValueSites);
 927     sys::swapByteOrder<uint32_t>(Kind);
 928   }
 929 }
 930
 931 void ValueProfData::deserializeTo(InstrProfRecord &Record,
 932                                   InstrProfSymtab *SymTab) {
 933   if (NumValueKinds == 0)
 934     return;
 935
 936   ValueProfRecord *VR = getFirstValueProfRecord(this);
 937   for (uint32_t K = 0; K < NumValueKinds; K++) {
 938     VR->deserializeTo(Record, SymTab);
 939     VR = getValueProfRecordNext(VR);
 940   }
 941 }
 942
 943 template <class T>
 944 static T swapToHostOrder(const unsigned char *&D, support::endianness Orig) {
 945   using namespace support;
 946
 947   if (Orig == little)
 948     return endian::readNext<T, little, unaligned>(D);
 949   else
 950     return endian::readNext<T, big, unaligned>(D);
 951 }
 952
 953 static std::unique_ptr<ValueProfData> allocValueProfData(uint32_t TotalSize) {
 954   return std::unique_ptr<ValueProfData>(new (::operator new(TotalSize))
 955                                             ValueProfData());
 956 }
 957
 958 Error ValueProfData::checkIntegrity() {
 959   if (NumValueKinds > IPVK_Last + 1)
 960     return make_error<InstrProfError>(
 961         instrprof_error::malformed, "number of value profile kinds is invalid");
 962   // Total size needs to be multiple of quadword size.
 963   if (TotalSize % sizeof(uint64_t))
 964     return make_error<InstrProfError>(
 965         instrprof_error::malformed, "total size is not multiples of quardword");
 966
 967   ValueProfRecord *VR = getFirstValueProfRecord(this);
 968   for (uint32_t K = 0; K < this->NumValueKinds; K++) {
 969     if (VR->Kind > IPVK_Last)
 970       return make_error<InstrProfError>(instrprof_error::malformed,
 971                                         "value kind is invalid");
 972     VR = getValueProfRecordNext(VR);
 973     if ((char *)VR - (char *)this > (ptrdiff_t)TotalSize)
 974       return make_error<InstrProfError>(
 975           instrprof_error::malformed,
 976           "value profile address is greater than total size");
 977   }
 978   return Error::success();
 979 }
 980
 981 Expected<std::unique_ptr<ValueProfData>>
 982 ValueProfData::getValueProfData(const unsigned char *D,
 983                                 const unsigned char *const BufferEnd,
 984                                 support::endianness Endianness) {
 985   using namespace support;
 986
 987   if (D + sizeof(ValueProfData) > BufferEnd)
 988     return make_error<InstrProfError>(instrprof_error::truncated);
 989
 990   const unsigned char *Header = D;
 991   uint32_t TotalSize = swapToHostOrder<uint32_t>(Header, Endianness);
 992   if (D + TotalSize > BufferEnd)
 993     return make_error<InstrProfError>(instrprof_error::too_large);
 994
 995   std::unique_ptr<ValueProfData> VPD = allocValueProfData(TotalSize);
 996   memcpy(VPD.get(), D, TotalSize);
 997   // Byte swap.
 998   VPD->swapBytesToHost(Endianness);
 999
1000   Error E = VPD->checkIntegrity();
1001   if (E)
1002     return std::move(E);
1003
1004   return std::move(VPD);
1005 }
1006
1007 void ValueProfData::swapBytesToHost(support::endianness Endianness) {
1008   using namespace support;
1009
1010   if (Endianness == getHostEndianness())
1011     return;
1012
1013   sys::swapByteOrder<uint32_t>(TotalSize);
1014   sys::swapByteOrder<uint32_t>(NumValueKinds);
1015
1016   ValueProfRecord *VR = getFirstValueProfRecord(this);
1017   for (uint32_t K = 0; K < NumValueKinds; K++) {
1018     VR->swapBytes(Endianness, getHostEndianness());
1019     VR = getValueProfRecordNext(VR);
1020   }
1021 }
1022
1023 void ValueProfData::swapBytesFromHost(support::endianness Endianness) {
1024   using namespace support;
1025
1026   if (Endianness == getHostEndianness())
1027     return;
1028
1029   ValueProfRecord *VR = getFirstValueProfRecord(this);
1030   for (uint32_t K = 0; K < NumValueKinds; K++) {
1031     ValueProfRecord *NVR = getValueProfRecordNext(VR);
1032     VR->swapBytes(getHostEndianness(), Endianness);
1033     VR = NVR;
1034   }
1035   sys::swapByteOrder<uint32_t>(TotalSize);
1036   sys::swapByteOrder<uint32_t>(NumValueKinds);
1037 }
1038
1039 void annotateValueSite(Module &M, Instruction &Inst,
1040                        const InstrProfRecord &InstrProfR,
1041                        InstrProfValueKind ValueKind, uint32_t SiteIdx,
1042                        uint32_t MaxMDCount) {
1043   uint32_t NV = InstrProfR.getNumValueDataForSite(ValueKind, SiteIdx);
1044   if (!NV)
1045     return;
1046
1047   uint64_t Sum = 0;
1048   std::unique_ptr<InstrProfValueData[]> VD =
1049       InstrProfR.getValueForSite(ValueKind, SiteIdx, &Sum);
1050
1051   ArrayRef<InstrProfValueData> VDs(VD.get(), NV);
1052   annotateValueSite(M, Inst, VDs, Sum, ValueKind, MaxMDCount);
1053 }
1054
1055 void annotateValueSite(Module &M, Instruction &Inst,
1056                        ArrayRef<InstrProfValueData> VDs,
1057                        uint64_t Sum, InstrProfValueKind ValueKind,
1058                        uint32_t MaxMDCount) {
1059   LLVMContext &Ctx = M.getContext();
1060   MDBuilder MDHelper(Ctx);
1061   SmallVector<Metadata *, 3> Vals;
1062   // Tag
1063   Vals.push_back(MDHelper.createString("VP"));
1064   // Value Kind
1065   Vals.push_back(MDHelper.createConstant(
1066       ConstantInt::get(Type::getInt32Ty(Ctx), ValueKind)));
1067   // Total Count
1068   Vals.push_back(
1069       MDHelper.createConstant(ConstantInt::get(Type::getInt64Ty(Ctx), Sum)));
1070
1071   // Value Profile Data
1072   uint32_t MDCount = MaxMDCount;
1073   for (auto &VD : VDs) {
1074     Vals.push_back(MDHelper.createConstant(
1075         ConstantInt::get(Type::getInt64Ty(Ctx), VD.Value)));
1076     Vals.push_back(MDHelper.createConstant(
1077         ConstantInt::get(Type::getInt64Ty(Ctx), VD.Count)));
1078     if (--MDCount == 0)
1079       break;
1080   }
1081   Inst.setMetadata(LLVMContext::MD_prof, MDNode::get(Ctx, Vals));
1082 }
1083
1084 bool getValueProfDataFromInst(const Instruction &Inst,
1085                               InstrProfValueKind ValueKind,
1086                               uint32_t MaxNumValueData,
1087                               InstrProfValueData ValueData[],
1088                               uint32_t &ActualNumValueData, uint64_t &TotalC,
1089                               bool GetNoICPValue) {
1090   MDNode *MD = Inst.getMetadata(LLVMContext::MD_prof);
1091   if (!MD)
1092     return false;
1093
1094   unsigned NOps = MD->getNumOperands();
1095
1096   if (NOps < 5)
1097     return false;
1098
1099   // Operand 0 is a string tag "VP":
1100   MDString *Tag = cast<MDString>(MD->getOperand(0));
1101   if (!Tag)
1102     return false;
1103
1104   if (!Tag->getString().equals("VP"))
1105     return false;
1106
1107   // Now check kind:
1108   ConstantInt *KindInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(1));
1109   if (!KindInt)
1110     return false;
1111   if (KindInt->getZExtValue() != ValueKind)
1112     return false;
1113
1114   // Get total count
1115   ConstantInt *TotalCInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(2));
1116   if (!TotalCInt)
1117     return false;
1118   TotalC = TotalCInt->getZExtValue();
1119
1120   ActualNumValueData = 0;
1121
1122   for (unsigned I = 3; I < NOps; I += 2) {
1123     if (ActualNumValueData >= MaxNumValueData)
1124       break;
1125     ConstantInt *Value = mdconst::dyn_extract<ConstantInt>(MD->getOperand(I));
1126     ConstantInt *Count =
1127         mdconst::dyn_extract<ConstantInt>(MD->getOperand(I + 1));
1128     if (!Value || !Count)
1129       return false;
1130     uint64_t CntValue = Count->getZExtValue();
1131     if (!GetNoICPValue && (CntValue == NOMORE_ICP_MAGICNUM))
1132       continue;
1133     ValueData[ActualNumValueData].Value = Value->getZExtValue();
1134     ValueData[ActualNumValueData].Count = CntValue;
1135     ActualNumValueData++;
1136   }
1137   return true;
1138 }
1139
1140 MDNode *getPGOFuncNameMetadata(const Function &F) {
1141   return F.getMetadata(getPGOFuncNameMetadataName());
1142 }
1143
1144 void createPGOFuncNameMetadata(Function &F, StringRef PGOFuncName) {
1145   // Only for internal linkage functions.
1146   if (PGOFuncName == F.getName())
1147       return;
1148   // Don't create duplicated meta-data.
1149   if (getPGOFuncNameMetadata(F))
1150     return;
1151   LLVMContext &C = F.getContext();
1152   MDNode *N = MDNode::get(C, MDString::get(C, PGOFuncName));
1153   F.setMetadata(getPGOFuncNameMetadataName(), N);
1154 }
1155
1156 bool needsComdatForCounter(const Function &F, const Module &M) {
1157   if (F.hasComdat())
1158     return true;
1159
1160   if (!Triple(M.getTargetTriple()).supportsCOMDAT())
1161     return false;
1162
1163   // See createPGOFuncNameVar for more details. To avoid link errors, profile
1164   // counters for function with available_externally linkage needs to be changed
1165   // to linkonce linkage. On ELF based systems, this leads to weak symbols to be
1166   // created. Without using comdat, duplicate entries won't be removed by the
1167   // linker leading to increased data segement size and raw profile size. Even
1168   // worse, since the referenced counter from profile per-function data object
1169   // will be resolved to the common strong definition, the profile counts for
1170   // available_externally functions will end up being duplicated in raw profile
1171   // data. This can result in distorted profile as the counts of those dups
1172   // will be accumulated by the profile merger.
1173   GlobalValue::LinkageTypes Linkage = F.getLinkage();
1174   if (Linkage != GlobalValue::ExternalWeakLinkage &&
1175       Linkage != GlobalValue::AvailableExternallyLinkage)
1176     return false;
1177
1178   return true;
1179 }
1180
1181 // Check if INSTR_PROF_RAW_VERSION_VAR is defined.
1182 bool isIRPGOFlagSet(const Module *M) {
1183   auto IRInstrVar =
1184       M->getNamedGlobal(INSTR_PROF_QUOTE(INSTR_PROF_RAW_VERSION_VAR));
1185   if (!IRInstrVar || IRInstrVar->hasLocalLinkage())
1186     return false;
1187
1188   // For CSPGO+LTO, this variable might be marked as non-prevailing and we only
1189   // have the decl.
1190   if (IRInstrVar->isDeclaration())
1191     return true;
1192
1193   // Check if the flag is set.
1194   if (!IRInstrVar->hasInitializer())
1195     return false;
1196
1197   auto *InitVal = dyn_cast_or_null<ConstantInt>(IRInstrVar->getInitializer());
1198   if (!InitVal)
1199     return false;
1200   return (InitVal->getZExtValue() & VARIANT_MASK_IR_PROF) != 0;
1201 }
1202
1203 // Check if we can safely rename this Comdat function.
1204 bool canRenameComdatFunc(const Function &F, bool CheckAddressTaken) {
1205   if (F.getName().empty())
1206     return false;
1207   if (!needsComdatForCounter(F, *(F.getParent())))
1208     return false;
1209   // Unsafe to rename the address-taken function (which can be used in
1210   // function comparison).
1211   if (CheckAddressTaken && F.hasAddressTaken())
1212     return false;
1213   // Only safe to do if this function may be discarded if it is not used
1214   // in the compilation unit.
1215   if (!GlobalValue::isDiscardableIfUnused(F.getLinkage()))
1216     return false;
1217
1218   // For AvailableExternallyLinkage functions.
1219   if (!F.hasComdat()) {
1220     assert(F.getLinkage() == GlobalValue::AvailableExternallyLinkage);
1221     return true;
1222   }
1223   return true;
1224 }
1225
1226 // Create the variable for the profile file name.
1227 void createProfileFileNameVar(Module &M, StringRef InstrProfileOutput) {
1228   if (InstrProfileOutput.empty())
1229     return;
1230   Constant *ProfileNameConst =
1231       ConstantDataArray::getString(M.getContext(), InstrProfileOutput, true);
1232   GlobalVariable *ProfileNameVar = new GlobalVariable(
1233       M, ProfileNameConst->getType(), true, GlobalValue::WeakAnyLinkage,
1234       ProfileNameConst, INSTR_PROF_QUOTE(INSTR_PROF_PROFILE_NAME_VAR));
1235   ProfileNameVar->setVisibility(GlobalValue::HiddenVisibility);
1236   Triple TT(M.getTargetTriple());
1237   if (TT.supportsCOMDAT()) {
1238     ProfileNameVar->setLinkage(GlobalValue::ExternalLinkage);
1239     ProfileNameVar->setComdat(M.getOrInsertComdat(
1240         StringRef(INSTR_PROF_QUOTE(INSTR_PROF_PROFILE_NAME_VAR))));
1241   }
1242 }
1243
1244 Error OverlapStats::accumulateCounts(const std::string &BaseFilename,
1245                                      const std::string &TestFilename,
1246                                      bool IsCS) {
1247   auto getProfileSum = [IsCS](const std::string &Filename,
1248                               CountSumOrPercent &Sum) -> Error {
1249     // This function is only used from llvm-profdata that doesn't use any kind
1250     // of VFS. Just create a default RealFileSystem to read profiles.
1251     auto FS = vfs::getRealFileSystem();
1252     auto ReaderOrErr = InstrProfReader::create(Filename, *FS);
1253     if (Error E = ReaderOrErr.takeError()) {
1254       return E;
1255     }
1256     auto Reader = std::move(ReaderOrErr.get());
1257     Reader->accumulateCounts(Sum, IsCS);
1258     return Error::success();
1259   };
1260   auto Ret = getProfileSum(BaseFilename, Base);
1261   if (Ret)
1262     return Ret;
1263   Ret = getProfileSum(TestFilename, Test);
1264   if (Ret)
1265     return Ret;
1266   this->BaseFilename = &BaseFilename;
1267   this->TestFilename = &TestFilename;
1268   Valid = true;
1269   return Error::success();
1270 }
1271
1272 void OverlapStats::addOneMismatch(const CountSumOrPercent &MismatchFunc) {
1273   Mismatch.NumEntries += 1;
1274   Mismatch.CountSum += MismatchFunc.CountSum / Test.CountSum;
1275   for (unsigned I = 0; I < IPVK_Last - IPVK_First + 1; I++) {
1276     if (Test.ValueCounts[I] >= 1.0f)
1277       Mismatch.ValueCounts[I] +=
1278           MismatchFunc.ValueCounts[I] / Test.ValueCounts[I];
1279   }
1280 }
1281
1282 void OverlapStats::addOneUnique(const CountSumOrPercent &UniqueFunc) {
1283   Unique.NumEntries += 1;
1284   Unique.CountSum += UniqueFunc.CountSum / Test.CountSum;
1285   for (unsigned I = 0; I < IPVK_Last - IPVK_First + 1; I++) {
1286     if (Test.ValueCounts[I] >= 1.0f)
1287       Unique.ValueCounts[I] += UniqueFunc.ValueCounts[I] / Test.ValueCounts[I];
1288   }
1289 }
1290
1291 void OverlapStats::dump(raw_fd_ostream &OS) const {
1292   if (!Valid)
1293     return;
1294
1295   const char *EntryName =
1296       (Level == ProgramLevel ? "functions" : "edge counters");
1297   if (Level == ProgramLevel) {
1298     OS << "Profile overlap infomation for base_profile: " << *BaseFilename
1299        << " and test_profile: " << *TestFilename << "\nProgram level:\n";
1300   } else {
1301     OS << "Function level:\n"
1302        << "  Function: " << FuncName << " (Hash=" << FuncHash << ")\n";
1303   }
1304
1305   OS << "  # of " << EntryName << " overlap: " << Overlap.NumEntries << "\n";
1306   if (Mismatch.NumEntries)
1307     OS << "  # of " << EntryName << " mismatch: " << Mismatch.NumEntries
1308        << "\n";
1309   if (Unique.NumEntries)
1310     OS << "  # of " << EntryName
1311        << " only in test_profile: " << Unique.NumEntries << "\n";
1312
1313   OS << "  Edge profile overlap: " << format("%.3f%%", Overlap.CountSum * 100)
1314      << "\n";
1315   if (Mismatch.NumEntries)
1316     OS << "  Mismatched count percentage (Edge): "
1317        << format("%.3f%%", Mismatch.CountSum * 100) << "\n";
1318   if (Unique.NumEntries)
1319     OS << "  Percentage of Edge profile only in test_profile: "
1320        << format("%.3f%%", Unique.CountSum * 100) << "\n";
1321   OS << "  Edge profile base count sum: " << format("%.0f", Base.CountSum)
1322      << "\n"
1323      << "  Edge profile test count sum: " << format("%.0f", Test.CountSum)
1324      << "\n";
1325
1326   for (unsigned I = 0; I < IPVK_Last - IPVK_First + 1; I++) {
1327     if (Base.ValueCounts[I] < 1.0f && Test.ValueCounts[I] < 1.0f)
1328       continue;
1329     char ProfileKindName[20];
1330     switch (I) {
1331     case IPVK_IndirectCallTarget:
1332       strncpy(ProfileKindName, "IndirectCall", 19);
1333       break;
1334     case IPVK_MemOPSize:
1335       strncpy(ProfileKindName, "MemOP", 19);
1336       break;
1337     default:
1338       snprintf(ProfileKindName, 19, "VP[%d]", I);
1339       break;
1340     }
1341     OS << "  " << ProfileKindName
1342        << " profile overlap: " << format("%.3f%%", Overlap.ValueCounts[I] * 100)
1343        << "\n";
1344     if (Mismatch.NumEntries)
1345       OS << "  Mismatched count percentage (" << ProfileKindName
1346          << "): " << format("%.3f%%", Mismatch.ValueCounts[I] * 100) << "\n";
1347     if (Unique.NumEntries)
1348       OS << "  Percentage of " << ProfileKindName
1349          << " profile only in test_profile: "
1350          << format("%.3f%%", Unique.ValueCounts[I] * 100) << "\n";
1351     OS << "  " << ProfileKindName
1352        << " profile base count sum: " << format("%.0f", Base.ValueCounts[I])
1353        << "\n"
1354        << "  " << ProfileKindName
1355        << " profile test count sum: " << format("%.0f", Test.ValueCounts[I])
1356        << "\n";
1357   }
1358 }
1359
1360 namespace IndexedInstrProf {
1361 // A C++14 compatible version of the offsetof macro.
1362 template <typename T1, typename T2>
1363 inline size_t constexpr offsetOf(T1 T2::*Member) {
1364   constexpr T2 Object{};
1365   return size_t(&(Object.*Member)) - size_t(&Object);
1366 }
1367
1368 static inline uint64_t read(const unsigned char *Buffer, size_t Offset) {
1369   return *reinterpret_cast<const uint64_t *>(Buffer + Offset);
1370 }
1371
1372 uint64_t Header::formatVersion() const {
1373   using namespace support;
1374   return endian::byte_swap<uint64_t, little>(Version);
1375 }
1376
1377 Expected<Header> Header::readFromBuffer(const unsigned char *Buffer) {
1378   using namespace support;
1379   static_assert(std::is_standard_layout_v<Header>,
1380                 "The header should be standard layout type since we use offset "
1381                 "of fields to read.");
1382   Header H;
1383
1384   H.Magic = read(Buffer, offsetOf(&Header::Magic));
1385   // Check the magic number.
1386   uint64_t Magic = endian::byte_swap<uint64_t, little>(H.Magic);
1387   if (Magic != IndexedInstrProf::Magic)
1388     return make_error<InstrProfError>(instrprof_error::bad_magic);
1389
1390   // Read the version.
1391   H.Version = read(Buffer, offsetOf(&Header::Version));
1392   if (GET_VERSION(H.formatVersion()) >
1393       IndexedInstrProf::ProfVersion::CurrentVersion)
1394     return make_error<InstrProfError>(instrprof_error::unsupported_version);
1395
1396   switch (GET_VERSION(H.formatVersion())) {
1397     // When a new field is added in the header add a case statement here to
1398     // populate it.
1399     static_assert(
1400         IndexedInstrProf::ProfVersion::CurrentVersion == Version10,
1401         "Please update the reading code below if a new field has been added, "
1402         "if not add a case statement to fall through to the latest version.");
1403   case 10ull:
1404     H.TemporalProfTracesOffset =
1405         read(Buffer, offsetOf(&Header::TemporalProfTracesOffset));
1406     [[fallthrough]];
1407   case 9ull:
1408     H.BinaryIdOffset = read(Buffer, offsetOf(&Header::BinaryIdOffset));
1409     [[fallthrough]];
1410   case 8ull:
1411     H.MemProfOffset = read(Buffer, offsetOf(&Header::MemProfOffset));
1412     [[fallthrough]];
1413   default: // Version7 (when the backwards compatible header was introduced).
1414     H.HashType = read(Buffer, offsetOf(&Header::HashType));
1415     H.HashOffset = read(Buffer, offsetOf(&Header::HashOffset));
1416   }
1417
1418   return H;
1419 }
1420
1421 size_t Header::size() const {
1422   switch (GET_VERSION(formatVersion())) {
1423     // When a new field is added to the header add a case statement here to
1424     // compute the size as offset of the new field + size of the new field. This
1425     // relies on the field being added to the end of the list.
1426     static_assert(IndexedInstrProf::ProfVersion::CurrentVersion == Version10,
1427                   "Please update the size computation below if a new field has "
1428                   "been added to the header, if not add a case statement to "
1429                   "fall through to the latest version.");
1430   case 10ull:
1431     return offsetOf(&Header::TemporalProfTracesOffset) +
1432            sizeof(Header::TemporalProfTracesOffset);
1433   case 9ull:
1434     return offsetOf(&Header::BinaryIdOffset) + sizeof(Header::BinaryIdOffset);
1435   case 8ull:
1436     return offsetOf(&Header::MemProfOffset) + sizeof(Header::MemProfOffset);
1437   default: // Version7 (when the backwards compatible header was introduced).
1438     return offsetOf(&Header::HashOffset) + sizeof(Header::HashOffset);
1439   }
1440 }
1441
1442 } // namespace IndexedInstrProf
1443
1444 } // end namespace llvm