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/SmallString.h"
  17 #include "llvm/ADT/SmallVector.h"
  18 #include "llvm/ADT/StringExtras.h"
  19 #include "llvm/ADT/StringRef.h"
  20 #include "llvm/ADT/Triple.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/ManagedStatic.h"
  43 #include "llvm/Support/MathExtras.h"
  44 #include "llvm/Support/Path.h"
  45 #include "llvm/Support/SwapByteOrder.h"
  46 #include <algorithm>
  47 #include <cassert>
  48 #include <cstddef>
  49 #include <cstdint>
  50 #include <cstring>
  51 #include <memory>
  52 #include <string>
  53 #include <system_error>
  54 #include <utility>
  55 #include <vector>
  56
  57 using namespace llvm;
  58
  59 static cl::opt<bool> StaticFuncFullModulePrefix(
  60     "static-func-full-module-prefix", cl::init(true), cl::Hidden,
  61     cl::desc("Use full module build paths in the profile counter names for "
  62              "static functions."));
  63
  64 // This option is tailored to users that have different top-level directory in
  65 // profile-gen and profile-use compilation. Users need to specific the number
  66 // of levels to strip. A value larger than the number of directories in the
  67 // source file will strip all the directory names and only leave the basename.
  68 //
  69 // Note current ThinLTO module importing for the indirect-calls assumes
  70 // the source directory name not being stripped. A non-zero option value here
  71 // can potentially prevent some inter-module indirect-call-promotions.
  72 static cl::opt<unsigned> StaticFuncStripDirNamePrefix(
  73     "static-func-strip-dirname-prefix", cl::init(0), cl::Hidden,
  74     cl::desc("Strip specified level of directory name from source path in "
  75              "the profile counter name for static functions."));
  76
  77 static std::string getInstrProfErrString(instrprof_error Err) {
  78   switch (Err) {
  79   case instrprof_error::success:
  80     return "success";
  81   case instrprof_error::eof:
  82     return "end of File";
  83   case instrprof_error::unrecognized_format:
  84     return "unrecognized instrumentation profile encoding format";
  85   case instrprof_error::bad_magic:
  86     return "invalid instrumentation profile data (bad magic)";
  87   case instrprof_error::bad_header:
  88     return "invalid instrumentation profile data (file header is corrupt)";
  89   case instrprof_error::unsupported_version:
  90     return "unsupported instrumentation profile format version";
  91   case instrprof_error::unsupported_hash_type:
  92     return "unsupported instrumentation profile hash type";
  93   case instrprof_error::too_large:
  94     return "too much profile data";
  95   case instrprof_error::truncated:
  96     return "truncated profile data";
  97   case instrprof_error::malformed:
  98     return "malformed instrumentation profile data";
  99   case instrprof_error::invalid_prof:
 100     return "invalid profile created. Please file a bug "
 101            "at: " BUG_REPORT_URL
 102            " and include the profraw files that caused this error.";
 103   case instrprof_error::unknown_function:
 104     return "no profile data available for function";
 105   case instrprof_error::hash_mismatch:
 106     return "function control flow change detected (hash mismatch)";
 107   case instrprof_error::count_mismatch:
 108     return "function basic block count change detected (counter mismatch)";
 109   case instrprof_error::counter_overflow:
 110     return "counter overflow";
 111   case instrprof_error::value_site_count_mismatch:
 112     return "function value site count change detected (counter mismatch)";
 113   case instrprof_error::compress_failed:
 114     return "failed to compress data (zlib)";
 115   case instrprof_error::uncompress_failed:
 116     return "failed to uncompress data (zlib)";
 117   case instrprof_error::empty_raw_profile:
 118     return "empty raw profile file";
 119   case instrprof_error::zlib_unavailable:
 120     return "profile uses zlib compression but the profile reader was built "
 121            "without zlib support";
 122   }
 123   llvm_unreachable("A value of instrprof_error has no message.");
 124 }
 125
 126 namespace {
 127
 128 // FIXME: This class is only here to support the transition to llvm::Error. It
 129 // will be removed once this transition is complete. Clients should prefer to
 130 // deal with the Error value directly, rather than converting to error_code.
 131 class InstrProfErrorCategoryType : public std::error_category {
 132   const char *name() const noexcept override { return "llvm.instrprof"; }
 133
 134   std::string message(int IE) const override {
 135     return getInstrProfErrString(static_cast<instrprof_error>(IE));
 136   }
 137 };
 138
 139 } // end anonymous namespace
 140
 141 static ManagedStatic<InstrProfErrorCategoryType> ErrorCategory;
 142
 143 const std::error_category &llvm::instrprof_category() {
 144   return *ErrorCategory;
 145 }
 146
 147 namespace {
 148
 149 const char *InstrProfSectNameCommon[] = {
 150 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
 151   SectNameCommon,
 152 #include "llvm/ProfileData/InstrProfData.inc"
 153 };
 154
 155 const char *InstrProfSectNameCoff[] = {
 156 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
 157   SectNameCoff,
 158 #include "llvm/ProfileData/InstrProfData.inc"
 159 };
 160
 161 const char *InstrProfSectNamePrefix[] = {
 162 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
 163   Prefix,
 164 #include "llvm/ProfileData/InstrProfData.inc"
 165 };
 166
 167 } // namespace
 168
 169 namespace llvm {
 170
 171 cl::opt<bool> DoInstrProfNameCompression(
 172     "enable-name-compression",
 173     cl::desc("Enable name/filename string compression"), cl::init(true));
 174
 175 std::string getInstrProfSectionName(InstrProfSectKind IPSK,
 176                                     Triple::ObjectFormatType OF,
 177                                     bool AddSegmentInfo) {
 178   std::string SectName;
 179
 180   if (OF == Triple::MachO && AddSegmentInfo)
 181     SectName = InstrProfSectNamePrefix[IPSK];
 182
 183   if (OF == Triple::COFF)
 184     SectName += InstrProfSectNameCoff[IPSK];
 185   else
 186     SectName += InstrProfSectNameCommon[IPSK];
 187
 188   if (OF == Triple::MachO && IPSK == IPSK_data && AddSegmentInfo)
 189     SectName += ",regular,live_support";
 190
 191   return SectName;
 192 }
 193
 194 void SoftInstrProfErrors::addError(instrprof_error IE) {
 195   if (IE == instrprof_error::success)
 196     return;
 197
 198   if (FirstError == instrprof_error::success)
 199     FirstError = IE;
 200
 201   switch (IE) {
 202   case instrprof_error::hash_mismatch:
 203     ++NumHashMismatches;
 204     break;
 205   case instrprof_error::count_mismatch:
 206     ++NumCountMismatches;
 207     break;
 208   case instrprof_error::counter_overflow:
 209     ++NumCounterOverflows;
 210     break;
 211   case instrprof_error::value_site_count_mismatch:
 212     ++NumValueSiteCountMismatches;
 213     break;
 214   default:
 215     llvm_unreachable("Not a soft error");
 216   }
 217 }
 218
 219 std::string InstrProfError::message() const {
 220   return getInstrProfErrString(Err);
 221 }
 222
 223 char InstrProfError::ID = 0;
 224
 225 std::string getPGOFuncName(StringRef RawFuncName,
 226                            GlobalValue::LinkageTypes Linkage,
 227                            StringRef FileName,
 228                            uint64_t Version LLVM_ATTRIBUTE_UNUSED) {
 229   return GlobalValue::getGlobalIdentifier(RawFuncName, Linkage, FileName);
 230 }
 231
 232 // Strip NumPrefix level of directory name from PathNameStr. If the number of
 233 // directory separators is less than NumPrefix, strip all the directories and
 234 // leave base file name only.
 235 static StringRef stripDirPrefix(StringRef PathNameStr, uint32_t NumPrefix) {
 236   uint32_t Count = NumPrefix;
 237   uint32_t Pos = 0, LastPos = 0;
 238   for (auto & CI : PathNameStr) {
 239     ++Pos;
 240     if (llvm::sys::path::is_separator(CI)) {
 241       LastPos = Pos;
 242       --Count;
 243     }
 244     if (Count == 0)
 245       break;
 246   }
 247   return PathNameStr.substr(LastPos);
 248 }
 249
 250 // Return the PGOFuncName. This function has some special handling when called
 251 // in LTO optimization. The following only applies when calling in LTO passes
 252 // (when \c InLTO is true): LTO's internalization privatizes many global linkage
 253 // symbols. This happens after value profile annotation, but those internal
 254 // linkage functions should not have a source prefix.
 255 // Additionally, for ThinLTO mode, exported internal functions are promoted
 256 // and renamed. We need to ensure that the original internal PGO name is
 257 // used when computing the GUID that is compared against the profiled GUIDs.
 258 // To differentiate compiler generated internal symbols from original ones,
 259 // PGOFuncName meta data are created and attached to the original internal
 260 // symbols in the value profile annotation step
 261 // (PGOUseFunc::annotateIndirectCallSites). If a symbol does not have the meta
 262 // data, its original linkage must be non-internal.
 263 std::string getPGOFuncName(const Function &F, bool InLTO, uint64_t Version) {
 264   if (!InLTO) {
 265     StringRef FileName(F.getParent()->getSourceFileName());
 266     uint32_t StripLevel = StaticFuncFullModulePrefix ? 0 : (uint32_t)-1;
 267     if (StripLevel < StaticFuncStripDirNamePrefix)
 268       StripLevel = StaticFuncStripDirNamePrefix;
 269     if (StripLevel)
 270       FileName = stripDirPrefix(FileName, StripLevel);
 271     return getPGOFuncName(F.getName(), F.getLinkage(), FileName, Version);
 272   }
 273
 274   // In LTO mode (when InLTO is true), first check if there is a meta data.
 275   if (MDNode *MD = getPGOFuncNameMetadata(F)) {
 276     StringRef S = cast<MDString>(MD->getOperand(0))->getString();
 277     return S.str();
 278   }
 279
 280   // If there is no meta data, the function must be a global before the value
 281   // profile annotation pass. Its current linkage may be internal if it is
 282   // internalized in LTO mode.
 283   return getPGOFuncName(F.getName(), GlobalValue::ExternalLinkage, "");
 284 }
 285
 286 StringRef getFuncNameWithoutPrefix(StringRef PGOFuncName, StringRef FileName) {
 287   if (FileName.empty())
 288     return PGOFuncName;
 289   // Drop the file name including ':'. See also getPGOFuncName.
 290   if (PGOFuncName.startswith(FileName))
 291     PGOFuncName = PGOFuncName.drop_front(FileName.size() + 1);
 292   return PGOFuncName;
 293 }
 294
 295 // \p FuncName is the string used as profile lookup key for the function. A
 296 // symbol is created to hold the name. Return the legalized symbol name.
 297 std::string getPGOFuncNameVarName(StringRef FuncName,
 298                                   GlobalValue::LinkageTypes Linkage) {
 299   std::string VarName = std::string(getInstrProfNameVarPrefix());
 300   VarName += FuncName;
 301
 302   if (!GlobalValue::isLocalLinkage(Linkage))
 303     return VarName;
 304
 305   // Now fix up illegal chars in local VarName that may upset the assembler.
 306   const char *InvalidChars = "-:<>/\"'";
 307   size_t found = VarName.find_first_of(InvalidChars);
 308   while (found != std::string::npos) {
 309     VarName[found] = '_';
 310     found = VarName.find_first_of(InvalidChars, found + 1);
 311   }
 312   return VarName;
 313 }
 314
 315 GlobalVariable *createPGOFuncNameVar(Module &M,
 316                                      GlobalValue::LinkageTypes Linkage,
 317                                      StringRef PGOFuncName) {
 318   // We generally want to match the function's linkage, but available_externally
 319   // and extern_weak both have the wrong semantics, and anything that doesn't
 320   // need to link across compilation units doesn't need to be visible at all.
 321   if (Linkage == GlobalValue::ExternalWeakLinkage)
 322     Linkage = GlobalValue::LinkOnceAnyLinkage;
 323   else if (Linkage == GlobalValue::AvailableExternallyLinkage)
 324     Linkage = GlobalValue::LinkOnceODRLinkage;
 325   else if (Linkage == GlobalValue::InternalLinkage ||
 326            Linkage == GlobalValue::ExternalLinkage)
 327     Linkage = GlobalValue::PrivateLinkage;
 328
 329   auto *Value =
 330       ConstantDataArray::getString(M.getContext(), PGOFuncName, false);
 331   auto FuncNameVar =
 332       new GlobalVariable(M, Value->getType(), true, Linkage, Value,
 333                          getPGOFuncNameVarName(PGOFuncName, Linkage));
 334
 335   // Hide the symbol so that we correctly get a copy for each executable.
 336   if (!GlobalValue::isLocalLinkage(FuncNameVar->getLinkage()))
 337     FuncNameVar->setVisibility(GlobalValue::HiddenVisibility);
 338
 339   return FuncNameVar;
 340 }
 341
 342 GlobalVariable *createPGOFuncNameVar(Function &F, StringRef PGOFuncName) {
 343   return createPGOFuncNameVar(*F.getParent(), F.getLinkage(), PGOFuncName);
 344 }
 345
 346 Error InstrProfSymtab::create(Module &M, bool InLTO) {
 347   for (Function &F : M) {
 348     // Function may not have a name: like using asm("") to overwrite the name.
 349     // Ignore in this case.
 350     if (!F.hasName())
 351       continue;
 352     const std::string &PGOFuncName = getPGOFuncName(F, InLTO);
 353     if (Error E = addFuncName(PGOFuncName))
 354       return E;
 355     MD5FuncMap.emplace_back(Function::getGUID(PGOFuncName), &F);
 356     // In ThinLTO, local function may have been promoted to global and have
 357     // suffix ".llvm." added to the function name. We need to add the
 358     // stripped function name to the symbol table so that we can find a match
 359     // from profile.
 360     //
 361     // We may have other suffixes similar as ".llvm." which are needed to
 362     // be stripped before the matching, but ".__uniq." suffix which is used
 363     // to differentiate internal linkage functions in different modules
 364     // should be kept. Now this is the only suffix with the pattern ".xxx"
 365     // which is kept before matching.
 366     const std::string UniqSuffix = ".__uniq.";
 367     auto pos = PGOFuncName.find(UniqSuffix);
 368     // Search '.' after ".__uniq." if ".__uniq." exists, otherwise
 369     // search '.' from the beginning.
 370     if (pos != std::string::npos)
 371       pos += UniqSuffix.length();
 372     else
 373       pos = 0;
 374     pos = PGOFuncName.find('.', pos);
 375     if (pos != std::string::npos && pos != 0) {
 376       const std::string &OtherFuncName = PGOFuncName.substr(0, pos);
 377       if (Error E = addFuncName(OtherFuncName))
 378         return E;
 379       MD5FuncMap.emplace_back(Function::getGUID(OtherFuncName), &F);
 380     }
 381   }
 382   Sorted = false;
 383   finalizeSymtab();
 384   return Error::success();
 385 }
 386
 387 uint64_t InstrProfSymtab::getFunctionHashFromAddress(uint64_t Address) {
 388   finalizeSymtab();
 389   auto It = partition_point(AddrToMD5Map, [=](std::pair<uint64_t, uint64_t> A) {
 390     return A.first < Address;
 391   });
 392   // Raw function pointer collected by value profiler may be from
 393   // external functions that are not instrumented. They won't have
 394   // mapping data to be used by the deserializer. Force the value to
 395   // be 0 in this case.
 396   if (It != AddrToMD5Map.end() && It->first == Address)
 397     return (uint64_t)It->second;
 398   return 0;
 399 }
 400
 401 Error collectPGOFuncNameStrings(ArrayRef<std::string> NameStrs,
 402                                 bool doCompression, std::string &Result) {
 403   assert(!NameStrs.empty() && "No name data to emit");
 404
 405   uint8_t Header[16], *P = Header;
 406   std::string UncompressedNameStrings =
 407       join(NameStrs.begin(), NameStrs.end(), getInstrProfNameSeparator());
 408
 409   assert(StringRef(UncompressedNameStrings)
 410                  .count(getInstrProfNameSeparator()) == (NameStrs.size() - 1) &&
 411          "PGO name is invalid (contains separator token)");
 412
 413   unsigned EncLen = encodeULEB128(UncompressedNameStrings.length(), P);
 414   P += EncLen;
 415
 416   auto WriteStringToResult = [&](size_t CompressedLen, StringRef InputStr) {
 417     EncLen = encodeULEB128(CompressedLen, P);
 418     P += EncLen;
 419     char *HeaderStr = reinterpret_cast<char *>(&Header[0]);
 420     unsigned HeaderLen = P - &Header[0];
 421     Result.append(HeaderStr, HeaderLen);
 422     Result += InputStr;
 423     return Error::success();
 424   };
 425
 426   if (!doCompression) {
 427     return WriteStringToResult(0, UncompressedNameStrings);
 428   }
 429
 430   SmallString<128> CompressedNameStrings;
 431   Error E = zlib::compress(StringRef(UncompressedNameStrings),
 432                            CompressedNameStrings, zlib::BestSizeCompression);
 433   if (E) {
 434     consumeError(std::move(E));
 435     return make_error<InstrProfError>(instrprof_error::compress_failed);
 436   }
 437
 438   return WriteStringToResult(CompressedNameStrings.size(),
 439                              CompressedNameStrings);
 440 }
 441
 442 StringRef getPGOFuncNameVarInitializer(GlobalVariable *NameVar) {
 443   auto *Arr = cast<ConstantDataArray>(NameVar->getInitializer());
 444   StringRef NameStr =
 445       Arr->isCString() ? Arr->getAsCString() : Arr->getAsString();
 446   return NameStr;
 447 }
 448
 449 Error collectPGOFuncNameStrings(ArrayRef<GlobalVariable *> NameVars,
 450                                 std::string &Result, bool doCompression) {
 451   std::vector<std::string> NameStrs;
 452   for (auto *NameVar : NameVars) {
 453     NameStrs.push_back(std::string(getPGOFuncNameVarInitializer(NameVar)));
 454   }
 455   return collectPGOFuncNameStrings(
 456       NameStrs, zlib::isAvailable() && doCompression, Result);
 457 }
 458
 459 Error readPGOFuncNameStrings(StringRef NameStrings, InstrProfSymtab &Symtab) {
 460   const uint8_t *P = NameStrings.bytes_begin();
 461   const uint8_t *EndP = NameStrings.bytes_end();
 462   while (P < EndP) {
 463     uint32_t N;
 464     uint64_t UncompressedSize = decodeULEB128(P, &N);
 465     P += N;
 466     uint64_t CompressedSize = decodeULEB128(P, &N);
 467     P += N;
 468     bool isCompressed = (CompressedSize != 0);
 469     SmallString<128> UncompressedNameStrings;
 470     StringRef NameStrings;
 471     if (isCompressed) {
 472       if (!llvm::zlib::isAvailable())
 473         return make_error<InstrProfError>(instrprof_error::zlib_unavailable);
 474
 475       StringRef CompressedNameStrings(reinterpret_cast<const char *>(P),
 476                                       CompressedSize);
 477       if (Error E =
 478               zlib::uncompress(CompressedNameStrings, UncompressedNameStrings,
 479                                UncompressedSize)) {
 480         consumeError(std::move(E));
 481         return make_error<InstrProfError>(instrprof_error::uncompress_failed);
 482       }
 483       P += CompressedSize;
 484       NameStrings = StringRef(UncompressedNameStrings.data(),
 485                               UncompressedNameStrings.size());
 486     } else {
 487       NameStrings =
 488           StringRef(reinterpret_cast<const char *>(P), UncompressedSize);
 489       P += UncompressedSize;
 490     }
 491     // Now parse the name strings.
 492     SmallVector<StringRef, 0> Names;
 493     NameStrings.split(Names, getInstrProfNameSeparator());
 494     for (StringRef &Name : Names)
 495       if (Error E = Symtab.addFuncName(Name))
 496         return E;
 497
 498     while (P < EndP && *P == 0)
 499       P++;
 500   }
 501   return Error::success();
 502 }
 503
 504 void InstrProfRecord::accumulateCounts(CountSumOrPercent &Sum) const {
 505   uint64_t FuncSum = 0;
 506   Sum.NumEntries += Counts.size();
 507   for (size_t F = 0, E = Counts.size(); F < E; ++F)
 508     FuncSum += Counts[F];
 509   Sum.CountSum += FuncSum;
 510
 511   for (uint32_t VK = IPVK_First; VK <= IPVK_Last; ++VK) {
 512     uint64_t KindSum = 0;
 513     uint32_t NumValueSites = getNumValueSites(VK);
 514     for (size_t I = 0; I < NumValueSites; ++I) {
 515       uint32_t NV = getNumValueDataForSite(VK, I);
 516       std::unique_ptr<InstrProfValueData[]> VD = getValueForSite(VK, I);
 517       for (uint32_t V = 0; V < NV; V++)
 518         KindSum += VD[V].Count;
 519     }
 520     Sum.ValueCounts[VK] += KindSum;
 521   }
 522 }
 523
 524 void InstrProfValueSiteRecord::overlap(InstrProfValueSiteRecord &Input,
 525                                        uint32_t ValueKind,
 526                                        OverlapStats &Overlap,
 527                                        OverlapStats &FuncLevelOverlap) {
 528   this->sortByTargetValues();
 529   Input.sortByTargetValues();
 530   double Score = 0.0f, FuncLevelScore = 0.0f;
 531   auto I = ValueData.begin();
 532   auto IE = ValueData.end();
 533   auto J = Input.ValueData.begin();
 534   auto JE = Input.ValueData.end();
 535   while (I != IE && J != JE) {
 536     if (I->Value == J->Value) {
 537       Score += OverlapStats::score(I->Count, J->Count,
 538                                    Overlap.Base.ValueCounts[ValueKind],
 539                                    Overlap.Test.ValueCounts[ValueKind]);
 540       FuncLevelScore += OverlapStats::score(
 541           I->Count, J->Count, FuncLevelOverlap.Base.ValueCounts[ValueKind],
 542           FuncLevelOverlap.Test.ValueCounts[ValueKind]);
 543       ++I;
 544     } else if (I->Value < J->Value) {
 545       ++I;
 546       continue;
 547     }
 548     ++J;
 549   }
 550   Overlap.Overlap.ValueCounts[ValueKind] += Score;
 551   FuncLevelOverlap.Overlap.ValueCounts[ValueKind] += FuncLevelScore;
 552 }
 553
 554 // Return false on mismatch.
 555 void InstrProfRecord::overlapValueProfData(uint32_t ValueKind,
 556                                            InstrProfRecord &Other,
 557                                            OverlapStats &Overlap,
 558                                            OverlapStats &FuncLevelOverlap) {
 559   uint32_t ThisNumValueSites = getNumValueSites(ValueKind);
 560   assert(ThisNumValueSites == Other.getNumValueSites(ValueKind));
 561   if (!ThisNumValueSites)
 562     return;
 563
 564   std::vector<InstrProfValueSiteRecord> &ThisSiteRecords =
 565       getOrCreateValueSitesForKind(ValueKind);
 566   MutableArrayRef<InstrProfValueSiteRecord> OtherSiteRecords =
 567       Other.getValueSitesForKind(ValueKind);
 568   for (uint32_t I = 0; I < ThisNumValueSites; I++)
 569     ThisSiteRecords[I].overlap(OtherSiteRecords[I], ValueKind, Overlap,
 570                                FuncLevelOverlap);
 571 }
 572
 573 void InstrProfRecord::overlap(InstrProfRecord &Other, OverlapStats &Overlap,
 574                               OverlapStats &FuncLevelOverlap,
 575                               uint64_t ValueCutoff) {
 576   // FuncLevel CountSum for other should already computed and nonzero.
 577   assert(FuncLevelOverlap.Test.CountSum >= 1.0f);
 578   accumulateCounts(FuncLevelOverlap.Base);
 579   bool Mismatch = (Counts.size() != Other.Counts.size());
 580
 581   // Check if the value profiles mismatch.
 582   if (!Mismatch) {
 583     for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind) {
 584       uint32_t ThisNumValueSites = getNumValueSites(Kind);
 585       uint32_t OtherNumValueSites = Other.getNumValueSites(Kind);
 586       if (ThisNumValueSites != OtherNumValueSites) {
 587         Mismatch = true;
 588         break;
 589       }
 590     }
 591   }
 592   if (Mismatch) {
 593     Overlap.addOneMismatch(FuncLevelOverlap.Test);
 594     return;
 595   }
 596
 597   // Compute overlap for value counts.
 598   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
 599     overlapValueProfData(Kind, Other, Overlap, FuncLevelOverlap);
 600
 601   double Score = 0.0;
 602   uint64_t MaxCount = 0;
 603   // Compute overlap for edge counts.
 604   for (size_t I = 0, E = Other.Counts.size(); I < E; ++I) {
 605     Score += OverlapStats::score(Counts[I], Other.Counts[I],
 606                                  Overlap.Base.CountSum, Overlap.Test.CountSum);
 607     MaxCount = std::max(Other.Counts[I], MaxCount);
 608   }
 609   Overlap.Overlap.CountSum += Score;
 610   Overlap.Overlap.NumEntries += 1;
 611
 612   if (MaxCount >= ValueCutoff) {
 613     double FuncScore = 0.0;
 614     for (size_t I = 0, E = Other.Counts.size(); I < E; ++I)
 615       FuncScore += OverlapStats::score(Counts[I], Other.Counts[I],
 616                                        FuncLevelOverlap.Base.CountSum,
 617                                        FuncLevelOverlap.Test.CountSum);
 618     FuncLevelOverlap.Overlap.CountSum = FuncScore;
 619     FuncLevelOverlap.Overlap.NumEntries = Other.Counts.size();
 620     FuncLevelOverlap.Valid = true;
 621   }
 622 }
 623
 624 void InstrProfValueSiteRecord::merge(InstrProfValueSiteRecord &Input,
 625                                      uint64_t Weight,
 626                                      function_ref<void(instrprof_error)> Warn) {
 627   this->sortByTargetValues();
 628   Input.sortByTargetValues();
 629   auto I = ValueData.begin();
 630   auto IE = ValueData.end();
 631   for (auto J = Input.ValueData.begin(), JE = Input.ValueData.end(); J != JE;
 632        ++J) {
 633     while (I != IE && I->Value < J->Value)
 634       ++I;
 635     if (I != IE && I->Value == J->Value) {
 636       bool Overflowed;
 637       I->Count = SaturatingMultiplyAdd(J->Count, Weight, I->Count, &Overflowed);
 638       if (Overflowed)
 639         Warn(instrprof_error::counter_overflow);
 640       ++I;
 641       continue;
 642     }
 643     ValueData.insert(I, *J);
 644   }
 645 }
 646
 647 void InstrProfValueSiteRecord::scale(uint64_t N, uint64_t D,
 648                                      function_ref<void(instrprof_error)> Warn) {
 649   for (auto I = ValueData.begin(), IE = ValueData.end(); I != IE; ++I) {
 650     bool Overflowed;
 651     I->Count = SaturatingMultiply(I->Count, N, &Overflowed) / D;
 652     if (Overflowed)
 653       Warn(instrprof_error::counter_overflow);
 654   }
 655 }
 656
 657 // Merge Value Profile data from Src record to this record for ValueKind.
 658 // Scale merged value counts by \p Weight.
 659 void InstrProfRecord::mergeValueProfData(
 660     uint32_t ValueKind, InstrProfRecord &Src, uint64_t Weight,
 661     function_ref<void(instrprof_error)> Warn) {
 662   uint32_t ThisNumValueSites = getNumValueSites(ValueKind);
 663   uint32_t OtherNumValueSites = Src.getNumValueSites(ValueKind);
 664   if (ThisNumValueSites != OtherNumValueSites) {
 665     Warn(instrprof_error::value_site_count_mismatch);
 666     return;
 667   }
 668   if (!ThisNumValueSites)
 669     return;
 670   std::vector<InstrProfValueSiteRecord> &ThisSiteRecords =
 671       getOrCreateValueSitesForKind(ValueKind);
 672   MutableArrayRef<InstrProfValueSiteRecord> OtherSiteRecords =
 673       Src.getValueSitesForKind(ValueKind);
 674   for (uint32_t I = 0; I < ThisNumValueSites; I++)
 675     ThisSiteRecords[I].merge(OtherSiteRecords[I], Weight, Warn);
 676 }
 677
 678 void InstrProfRecord::merge(InstrProfRecord &Other, uint64_t Weight,
 679                             function_ref<void(instrprof_error)> Warn) {
 680   // If the number of counters doesn't match we either have bad data
 681   // or a hash collision.
 682   if (Counts.size() != Other.Counts.size()) {
 683     Warn(instrprof_error::count_mismatch);
 684     return;
 685   }
 686
 687   for (size_t I = 0, E = Other.Counts.size(); I < E; ++I) {
 688     bool Overflowed;
 689     Counts[I] =
 690         SaturatingMultiplyAdd(Other.Counts[I], Weight, Counts[I], &Overflowed);
 691     if (Overflowed)
 692       Warn(instrprof_error::counter_overflow);
 693   }
 694
 695   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
 696     mergeValueProfData(Kind, Other, Weight, Warn);
 697 }
 698
 699 void InstrProfRecord::scaleValueProfData(
 700     uint32_t ValueKind, uint64_t N, uint64_t D,
 701     function_ref<void(instrprof_error)> Warn) {
 702   for (auto &R : getValueSitesForKind(ValueKind))
 703     R.scale(N, D, Warn);
 704 }
 705
 706 void InstrProfRecord::scale(uint64_t N, uint64_t D,
 707                             function_ref<void(instrprof_error)> Warn) {
 708   assert(D != 0 && "D cannot be 0");
 709   for (auto &Count : this->Counts) {
 710     bool Overflowed;
 711     Count = SaturatingMultiply(Count, N, &Overflowed) / D;
 712     if (Overflowed)
 713       Warn(instrprof_error::counter_overflow);
 714   }
 715   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
 716     scaleValueProfData(Kind, N, D, Warn);
 717 }
 718
 719 // Map indirect call target name hash to name string.
 720 uint64_t InstrProfRecord::remapValue(uint64_t Value, uint32_t ValueKind,
 721                                      InstrProfSymtab *SymTab) {
 722   if (!SymTab)
 723     return Value;
 724
 725   if (ValueKind == IPVK_IndirectCallTarget)
 726     return SymTab->getFunctionHashFromAddress(Value);
 727
 728   return Value;
 729 }
 730
 731 void InstrProfRecord::addValueData(uint32_t ValueKind, uint32_t Site,
 732                                    InstrProfValueData *VData, uint32_t N,
 733                                    InstrProfSymtab *ValueMap) {
 734   for (uint32_t I = 0; I < N; I++) {
 735     VData[I].Value = remapValue(VData[I].Value, ValueKind, ValueMap);
 736   }
 737   std::vector<InstrProfValueSiteRecord> &ValueSites =
 738       getOrCreateValueSitesForKind(ValueKind);
 739   if (N == 0)
 740     ValueSites.emplace_back();
 741   else
 742     ValueSites.emplace_back(VData, VData + N);
 743 }
 744
 745 #define INSTR_PROF_COMMON_API_IMPL
 746 #include "llvm/ProfileData/InstrProfData.inc"
 747
 748 /*!
 749  * ValueProfRecordClosure Interface implementation for  InstrProfRecord
 750  *  class. These C wrappers are used as adaptors so that C++ code can be
 751  *  invoked as callbacks.
 752  */
 753 uint32_t getNumValueKindsInstrProf(const void *Record) {
 754   return reinterpret_cast<const InstrProfRecord *>(Record)->getNumValueKinds();
 755 }
 756
 757 uint32_t getNumValueSitesInstrProf(const void *Record, uint32_t VKind) {
 758   return reinterpret_cast<const InstrProfRecord *>(Record)
 759       ->getNumValueSites(VKind);
 760 }
 761
 762 uint32_t getNumValueDataInstrProf(const void *Record, uint32_t VKind) {
 763   return reinterpret_cast<const InstrProfRecord *>(Record)
 764       ->getNumValueData(VKind);
 765 }
 766
 767 uint32_t getNumValueDataForSiteInstrProf(const void *R, uint32_t VK,
 768                                          uint32_t S) {
 769   return reinterpret_cast<const InstrProfRecord *>(R)
 770       ->getNumValueDataForSite(VK, S);
 771 }
 772
 773 void getValueForSiteInstrProf(const void *R, InstrProfValueData *Dst,
 774                               uint32_t K, uint32_t S) {
 775   reinterpret_cast<const InstrProfRecord *>(R)->getValueForSite(Dst, K, S);
 776 }
 777
 778 ValueProfData *allocValueProfDataInstrProf(size_t TotalSizeInBytes) {
 779   ValueProfData *VD =
 780       (ValueProfData *)(new (::operator new(TotalSizeInBytes)) ValueProfData());
 781   memset(VD, 0, TotalSizeInBytes);
 782   return VD;
 783 }
 784
 785 static ValueProfRecordClosure InstrProfRecordClosure = {
 786     nullptr,
 787     getNumValueKindsInstrProf,
 788     getNumValueSitesInstrProf,
 789     getNumValueDataInstrProf,
 790     getNumValueDataForSiteInstrProf,
 791     nullptr,
 792     getValueForSiteInstrProf,
 793     allocValueProfDataInstrProf};
 794
 795 // Wrapper implementation using the closure mechanism.
 796 uint32_t ValueProfData::getSize(const InstrProfRecord &Record) {
 797   auto Closure = InstrProfRecordClosure;
 798   Closure.Record = &Record;
 799   return getValueProfDataSize(&Closure);
 800 }
 801
 802 // Wrapper implementation using the closure mechanism.
 803 std::unique_ptr<ValueProfData>
 804 ValueProfData::serializeFrom(const InstrProfRecord &Record) {
 805   InstrProfRecordClosure.Record = &Record;
 806
 807   std::unique_ptr<ValueProfData> VPD(
 808       serializeValueProfDataFrom(&InstrProfRecordClosure, nullptr));
 809   return VPD;
 810 }
 811
 812 void ValueProfRecord::deserializeTo(InstrProfRecord &Record,
 813                                     InstrProfSymtab *SymTab) {
 814   Record.reserveSites(Kind, NumValueSites);
 815
 816   InstrProfValueData *ValueData = getValueProfRecordValueData(this);
 817   for (uint64_t VSite = 0; VSite < NumValueSites; ++VSite) {
 818     uint8_t ValueDataCount = this->SiteCountArray[VSite];
 819     Record.addValueData(Kind, VSite, ValueData, ValueDataCount, SymTab);
 820     ValueData += ValueDataCount;
 821   }
 822 }
 823
 824 // For writing/serializing,  Old is the host endianness, and  New is
 825 // byte order intended on disk. For Reading/deserialization, Old
 826 // is the on-disk source endianness, and New is the host endianness.
 827 void ValueProfRecord::swapBytes(support::endianness Old,
 828                                 support::endianness New) {
 829   using namespace support;
 830
 831   if (Old == New)
 832     return;
 833
 834   if (getHostEndianness() != Old) {
 835     sys::swapByteOrder<uint32_t>(NumValueSites);
 836     sys::swapByteOrder<uint32_t>(Kind);
 837   }
 838   uint32_t ND = getValueProfRecordNumValueData(this);
 839   InstrProfValueData *VD = getValueProfRecordValueData(this);
 840
 841   // No need to swap byte array: SiteCountArrray.
 842   for (uint32_t I = 0; I < ND; I++) {
 843     sys::swapByteOrder<uint64_t>(VD[I].Value);
 844     sys::swapByteOrder<uint64_t>(VD[I].Count);
 845   }
 846   if (getHostEndianness() == Old) {
 847     sys::swapByteOrder<uint32_t>(NumValueSites);
 848     sys::swapByteOrder<uint32_t>(Kind);
 849   }
 850 }
 851
 852 void ValueProfData::deserializeTo(InstrProfRecord &Record,
 853                                   InstrProfSymtab *SymTab) {
 854   if (NumValueKinds == 0)
 855     return;
 856
 857   ValueProfRecord *VR = getFirstValueProfRecord(this);
 858   for (uint32_t K = 0; K < NumValueKinds; K++) {
 859     VR->deserializeTo(Record, SymTab);
 860     VR = getValueProfRecordNext(VR);
 861   }
 862 }
 863
 864 template <class T>
 865 static T swapToHostOrder(const unsigned char *&D, support::endianness Orig) {
 866   using namespace support;
 867
 868   if (Orig == little)
 869     return endian::readNext<T, little, unaligned>(D);
 870   else
 871     return endian::readNext<T, big, unaligned>(D);
 872 }
 873
 874 static std::unique_ptr<ValueProfData> allocValueProfData(uint32_t TotalSize) {
 875   return std::unique_ptr<ValueProfData>(new (::operator new(TotalSize))
 876                                             ValueProfData());
 877 }
 878
 879 Error ValueProfData::checkIntegrity() {
 880   if (NumValueKinds > IPVK_Last + 1)
 881     return make_error<InstrProfError>(instrprof_error::malformed);
 882   // Total size needs to be mulltiple of quadword size.
 883   if (TotalSize % sizeof(uint64_t))
 884     return make_error<InstrProfError>(instrprof_error::malformed);
 885
 886   ValueProfRecord *VR = getFirstValueProfRecord(this);
 887   for (uint32_t K = 0; K < this->NumValueKinds; K++) {
 888     if (VR->Kind > IPVK_Last)
 889       return make_error<InstrProfError>(instrprof_error::malformed);
 890     VR = getValueProfRecordNext(VR);
 891     if ((char *)VR - (char *)this > (ptrdiff_t)TotalSize)
 892       return make_error<InstrProfError>(instrprof_error::malformed);
 893   }
 894   return Error::success();
 895 }
 896
 897 Expected<std::unique_ptr<ValueProfData>>
 898 ValueProfData::getValueProfData(const unsigned char *D,
 899                                 const unsigned char *const BufferEnd,
 900                                 support::endianness Endianness) {
 901   using namespace support;
 902
 903   if (D + sizeof(ValueProfData) > BufferEnd)
 904     return make_error<InstrProfError>(instrprof_error::truncated);
 905
 906   const unsigned char *Header = D;
 907   uint32_t TotalSize = swapToHostOrder<uint32_t>(Header, Endianness);
 908   if (D + TotalSize > BufferEnd)
 909     return make_error<InstrProfError>(instrprof_error::too_large);
 910
 911   std::unique_ptr<ValueProfData> VPD = allocValueProfData(TotalSize);
 912   memcpy(VPD.get(), D, TotalSize);
 913   // Byte swap.
 914   VPD->swapBytesToHost(Endianness);
 915
 916   Error E = VPD->checkIntegrity();
 917   if (E)
 918     return std::move(E);
 919
 920   return std::move(VPD);
 921 }
 922
 923 void ValueProfData::swapBytesToHost(support::endianness Endianness) {
 924   using namespace support;
 925
 926   if (Endianness == getHostEndianness())
 927     return;
 928
 929   sys::swapByteOrder<uint32_t>(TotalSize);
 930   sys::swapByteOrder<uint32_t>(NumValueKinds);
 931
 932   ValueProfRecord *VR = getFirstValueProfRecord(this);
 933   for (uint32_t K = 0; K < NumValueKinds; K++) {
 934     VR->swapBytes(Endianness, getHostEndianness());
 935     VR = getValueProfRecordNext(VR);
 936   }
 937 }
 938
 939 void ValueProfData::swapBytesFromHost(support::endianness Endianness) {
 940   using namespace support;
 941
 942   if (Endianness == getHostEndianness())
 943     return;
 944
 945   ValueProfRecord *VR = getFirstValueProfRecord(this);
 946   for (uint32_t K = 0; K < NumValueKinds; K++) {
 947     ValueProfRecord *NVR = getValueProfRecordNext(VR);
 948     VR->swapBytes(getHostEndianness(), Endianness);
 949     VR = NVR;
 950   }
 951   sys::swapByteOrder<uint32_t>(TotalSize);
 952   sys::swapByteOrder<uint32_t>(NumValueKinds);
 953 }
 954
 955 void annotateValueSite(Module &M, Instruction &Inst,
 956                        const InstrProfRecord &InstrProfR,
 957                        InstrProfValueKind ValueKind, uint32_t SiteIdx,
 958                        uint32_t MaxMDCount) {
 959   uint32_t NV = InstrProfR.getNumValueDataForSite(ValueKind, SiteIdx);
 960   if (!NV)
 961     return;
 962
 963   uint64_t Sum = 0;
 964   std::unique_ptr<InstrProfValueData[]> VD =
 965       InstrProfR.getValueForSite(ValueKind, SiteIdx, &Sum);
 966
 967   ArrayRef<InstrProfValueData> VDs(VD.get(), NV);
 968   annotateValueSite(M, Inst, VDs, Sum, ValueKind, MaxMDCount);
 969 }
 970
 971 void annotateValueSite(Module &M, Instruction &Inst,
 972                        ArrayRef<InstrProfValueData> VDs,
 973                        uint64_t Sum, InstrProfValueKind ValueKind,
 974                        uint32_t MaxMDCount) {
 975   LLVMContext &Ctx = M.getContext();
 976   MDBuilder MDHelper(Ctx);
 977   SmallVector<Metadata *, 3> Vals;
 978   // Tag
 979   Vals.push_back(MDHelper.createString("VP"));
 980   // Value Kind
 981   Vals.push_back(MDHelper.createConstant(
 982       ConstantInt::get(Type::getInt32Ty(Ctx), ValueKind)));
 983   // Total Count
 984   Vals.push_back(
 985       MDHelper.createConstant(ConstantInt::get(Type::getInt64Ty(Ctx), Sum)));
 986
 987   // Value Profile Data
 988   uint32_t MDCount = MaxMDCount;
 989   for (auto &VD : VDs) {
 990     Vals.push_back(MDHelper.createConstant(
 991         ConstantInt::get(Type::getInt64Ty(Ctx), VD.Value)));
 992     Vals.push_back(MDHelper.createConstant(
 993         ConstantInt::get(Type::getInt64Ty(Ctx), VD.Count)));
 994     if (--MDCount == 0)
 995       break;
 996   }
 997   Inst.setMetadata(LLVMContext::MD_prof, MDNode::get(Ctx, Vals));
 998 }
 999
1000 bool getValueProfDataFromInst(const Instruction &Inst,
1001                               InstrProfValueKind ValueKind,
1002                               uint32_t MaxNumValueData,
1003                               InstrProfValueData ValueData[],
1004                               uint32_t &ActualNumValueData, uint64_t &TotalC,
1005                               bool GetNoICPValue) {
1006   MDNode *MD = Inst.getMetadata(LLVMContext::MD_prof);
1007   if (!MD)
1008     return false;
1009
1010   unsigned NOps = MD->getNumOperands();
1011
1012   if (NOps < 5)
1013     return false;
1014
1015   // Operand 0 is a string tag "VP":
1016   MDString *Tag = cast<MDString>(MD->getOperand(0));
1017   if (!Tag)
1018     return false;
1019
1020   if (!Tag->getString().equals("VP"))
1021     return false;
1022
1023   // Now check kind:
1024   ConstantInt *KindInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(1));
1025   if (!KindInt)
1026     return false;
1027   if (KindInt->getZExtValue() != ValueKind)
1028     return false;
1029
1030   // Get total count
1031   ConstantInt *TotalCInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(2));
1032   if (!TotalCInt)
1033     return false;
1034   TotalC = TotalCInt->getZExtValue();
1035
1036   ActualNumValueData = 0;
1037
1038   for (unsigned I = 3; I < NOps; I += 2) {
1039     if (ActualNumValueData >= MaxNumValueData)
1040       break;
1041     ConstantInt *Value = mdconst::dyn_extract<ConstantInt>(MD->getOperand(I));
1042     ConstantInt *Count =
1043         mdconst::dyn_extract<ConstantInt>(MD->getOperand(I + 1));
1044     if (!Value || !Count)
1045       return false;
1046     uint64_t CntValue = Count->getZExtValue();
1047     if (!GetNoICPValue && (CntValue == NOMORE_ICP_MAGICNUM))
1048       continue;
1049     ValueData[ActualNumValueData].Value = Value->getZExtValue();
1050     ValueData[ActualNumValueData].Count = CntValue;
1051     ActualNumValueData++;
1052   }
1053   return true;
1054 }
1055
1056 MDNode *getPGOFuncNameMetadata(const Function &F) {
1057   return F.getMetadata(getPGOFuncNameMetadataName());
1058 }
1059
1060 void createPGOFuncNameMetadata(Function &F, StringRef PGOFuncName) {
1061   // Only for internal linkage functions.
1062   if (PGOFuncName == F.getName())
1063       return;
1064   // Don't create duplicated meta-data.
1065   if (getPGOFuncNameMetadata(F))
1066     return;
1067   LLVMContext &C = F.getContext();
1068   MDNode *N = MDNode::get(C, MDString::get(C, PGOFuncName));
1069   F.setMetadata(getPGOFuncNameMetadataName(), N);
1070 }
1071
1072 bool needsComdatForCounter(const Function &F, const Module &M) {
1073   if (F.hasComdat())
1074     return true;
1075
1076   if (!Triple(M.getTargetTriple()).supportsCOMDAT())
1077     return false;
1078
1079   // See createPGOFuncNameVar for more details. To avoid link errors, profile
1080   // counters for function with available_externally linkage needs to be changed
1081   // to linkonce linkage. On ELF based systems, this leads to weak symbols to be
1082   // created. Without using comdat, duplicate entries won't be removed by the
1083   // linker leading to increased data segement size and raw profile size. Even
1084   // worse, since the referenced counter from profile per-function data object
1085   // will be resolved to the common strong definition, the profile counts for
1086   // available_externally functions will end up being duplicated in raw profile
1087   // data. This can result in distorted profile as the counts of those dups
1088   // will be accumulated by the profile merger.
1089   GlobalValue::LinkageTypes Linkage = F.getLinkage();
1090   if (Linkage != GlobalValue::ExternalWeakLinkage &&
1091       Linkage != GlobalValue::AvailableExternallyLinkage)
1092     return false;
1093
1094   return true;
1095 }
1096
1097 // Check if INSTR_PROF_RAW_VERSION_VAR is defined.
1098 bool isIRPGOFlagSet(const Module *M) {
1099   auto IRInstrVar =
1100       M->getNamedGlobal(INSTR_PROF_QUOTE(INSTR_PROF_RAW_VERSION_VAR));
1101   if (!IRInstrVar || IRInstrVar->isDeclaration() ||
1102       IRInstrVar->hasLocalLinkage())
1103     return false;
1104
1105   // Check if the flag is set.
1106   if (!IRInstrVar->hasInitializer())
1107     return false;
1108
1109   auto *InitVal = dyn_cast_or_null<ConstantInt>(IRInstrVar->getInitializer());
1110   if (!InitVal)
1111     return false;
1112   return (InitVal->getZExtValue() & VARIANT_MASK_IR_PROF) != 0;
1113 }
1114
1115 // Check if we can safely rename this Comdat function.
1116 bool canRenameComdatFunc(const Function &F, bool CheckAddressTaken) {
1117   if (F.getName().empty())
1118     return false;
1119   if (!needsComdatForCounter(F, *(F.getParent())))
1120     return false;
1121   // Unsafe to rename the address-taken function (which can be used in
1122   // function comparison).
1123   if (CheckAddressTaken && F.hasAddressTaken())
1124     return false;
1125   // Only safe to do if this function may be discarded if it is not used
1126   // in the compilation unit.
1127   if (!GlobalValue::isDiscardableIfUnused(F.getLinkage()))
1128     return false;
1129
1130   // For AvailableExternallyLinkage functions.
1131   if (!F.hasComdat()) {
1132     assert(F.getLinkage() == GlobalValue::AvailableExternallyLinkage);
1133     return true;
1134   }
1135   return true;
1136 }
1137
1138 // Create a COMDAT variable INSTR_PROF_RAW_VERSION_VAR to make the runtime
1139 // aware this is an ir_level profile so it can set the version flag.
1140 void createIRLevelProfileFlagVar(Module &M, bool IsCS,
1141                                  bool InstrEntryBBEnabled) {
1142   const StringRef VarName(INSTR_PROF_QUOTE(INSTR_PROF_RAW_VERSION_VAR));
1143   Type *IntTy64 = Type::getInt64Ty(M.getContext());
1144   uint64_t ProfileVersion = (INSTR_PROF_RAW_VERSION | VARIANT_MASK_IR_PROF);
1145   if (IsCS)
1146     ProfileVersion |= VARIANT_MASK_CSIR_PROF;
1147   if (InstrEntryBBEnabled)
1148     ProfileVersion |= VARIANT_MASK_INSTR_ENTRY;
1149   auto IRLevelVersionVariable = new GlobalVariable(
1150       M, IntTy64, true, GlobalValue::WeakAnyLinkage,
1151       Constant::getIntegerValue(IntTy64, APInt(64, ProfileVersion)), VarName);
1152   IRLevelVersionVariable->setVisibility(GlobalValue::DefaultVisibility);
1153   Triple TT(M.getTargetTriple());
1154   if (TT.supportsCOMDAT()) {
1155     IRLevelVersionVariable->setLinkage(GlobalValue::ExternalLinkage);
1156     IRLevelVersionVariable->setComdat(M.getOrInsertComdat(VarName));
1157   }
1158 }
1159
1160 // Create the variable for the profile file name.
1161 void createProfileFileNameVar(Module &M, StringRef InstrProfileOutput) {
1162   if (InstrProfileOutput.empty())
1163     return;
1164   Constant *ProfileNameConst =
1165       ConstantDataArray::getString(M.getContext(), InstrProfileOutput, true);
1166   GlobalVariable *ProfileNameVar = new GlobalVariable(
1167       M, ProfileNameConst->getType(), true, GlobalValue::WeakAnyLinkage,
1168       ProfileNameConst, INSTR_PROF_QUOTE(INSTR_PROF_PROFILE_NAME_VAR));
1169   Triple TT(M.getTargetTriple());
1170   if (TT.supportsCOMDAT()) {
1171     ProfileNameVar->setLinkage(GlobalValue::ExternalLinkage);
1172     ProfileNameVar->setComdat(M.getOrInsertComdat(
1173         StringRef(INSTR_PROF_QUOTE(INSTR_PROF_PROFILE_NAME_VAR))));
1174   }
1175 }
1176
1177 Error OverlapStats::accumulateCounts(const std::string &BaseFilename,
1178                                      const std::string &TestFilename,
1179                                      bool IsCS) {
1180   auto getProfileSum = [IsCS](const std::string &Filename,
1181                               CountSumOrPercent &Sum) -> Error {
1182     auto ReaderOrErr = InstrProfReader::create(Filename);
1183     if (Error E = ReaderOrErr.takeError()) {
1184       return E;
1185     }
1186     auto Reader = std::move(ReaderOrErr.get());
1187     Reader->accumulateCounts(Sum, IsCS);
1188     return Error::success();
1189   };
1190   auto Ret = getProfileSum(BaseFilename, Base);
1191   if (Ret)
1192     return Ret;
1193   Ret = getProfileSum(TestFilename, Test);
1194   if (Ret)
1195     return Ret;
1196   this->BaseFilename = &BaseFilename;
1197   this->TestFilename = &TestFilename;
1198   Valid = true;
1199   return Error::success();
1200 }
1201
1202 void OverlapStats::addOneMismatch(const CountSumOrPercent &MismatchFunc) {
1203   Mismatch.NumEntries += 1;
1204   Mismatch.CountSum += MismatchFunc.CountSum / Test.CountSum;
1205   for (unsigned I = 0; I < IPVK_Last - IPVK_First + 1; I++) {
1206     if (Test.ValueCounts[I] >= 1.0f)
1207       Mismatch.ValueCounts[I] +=
1208           MismatchFunc.ValueCounts[I] / Test.ValueCounts[I];
1209   }
1210 }
1211
1212 void OverlapStats::addOneUnique(const CountSumOrPercent &UniqueFunc) {
1213   Unique.NumEntries += 1;
1214   Unique.CountSum += UniqueFunc.CountSum / Test.CountSum;
1215   for (unsigned I = 0; I < IPVK_Last - IPVK_First + 1; I++) {
1216     if (Test.ValueCounts[I] >= 1.0f)
1217       Unique.ValueCounts[I] += UniqueFunc.ValueCounts[I] / Test.ValueCounts[I];
1218   }
1219 }
1220
1221 void OverlapStats::dump(raw_fd_ostream &OS) const {
1222   if (!Valid)
1223     return;
1224
1225   const char *EntryName =
1226       (Level == ProgramLevel ? "functions" : "edge counters");
1227   if (Level == ProgramLevel) {
1228     OS << "Profile overlap infomation for base_profile: " << *BaseFilename
1229        << " and test_profile: " << *TestFilename << "\nProgram level:\n";
1230   } else {
1231     OS << "Function level:\n"
1232        << "  Function: " << FuncName << " (Hash=" << FuncHash << ")\n";
1233   }
1234
1235   OS << "  # of " << EntryName << " overlap: " << Overlap.NumEntries << "\n";
1236   if (Mismatch.NumEntries)
1237     OS << "  # of " << EntryName << " mismatch: " << Mismatch.NumEntries
1238        << "\n";
1239   if (Unique.NumEntries)
1240     OS << "  # of " << EntryName
1241        << " only in test_profile: " << Unique.NumEntries << "\n";
1242
1243   OS << "  Edge profile overlap: " << format("%.3f%%", Overlap.CountSum * 100)
1244      << "\n";
1245   if (Mismatch.NumEntries)
1246     OS << "  Mismatched count percentage (Edge): "
1247        << format("%.3f%%", Mismatch.CountSum * 100) << "\n";
1248   if (Unique.NumEntries)
1249     OS << "  Percentage of Edge profile only in test_profile: "
1250        << format("%.3f%%", Unique.CountSum * 100) << "\n";
1251   OS << "  Edge profile base count sum: " << format("%.0f", Base.CountSum)
1252      << "\n"
1253      << "  Edge profile test count sum: " << format("%.0f", Test.CountSum)
1254      << "\n";
1255
1256   for (unsigned I = 0; I < IPVK_Last - IPVK_First + 1; I++) {
1257     if (Base.ValueCounts[I] < 1.0f && Test.ValueCounts[I] < 1.0f)
1258       continue;
1259     char ProfileKindName[20];
1260     switch (I) {
1261     case IPVK_IndirectCallTarget:
1262       strncpy(ProfileKindName, "IndirectCall", 19);
1263       break;
1264     case IPVK_MemOPSize:
1265       strncpy(ProfileKindName, "MemOP", 19);
1266       break;
1267     default:
1268       snprintf(ProfileKindName, 19, "VP[%d]", I);
1269       break;
1270     }
1271     OS << "  " << ProfileKindName
1272        << " profile overlap: " << format("%.3f%%", Overlap.ValueCounts[I] * 100)
1273        << "\n";
1274     if (Mismatch.NumEntries)
1275       OS << "  Mismatched count percentage (" << ProfileKindName
1276          << "): " << format("%.3f%%", Mismatch.ValueCounts[I] * 100) << "\n";
1277     if (Unique.NumEntries)
1278       OS << "  Percentage of " << ProfileKindName
1279          << " profile only in test_profile: "
1280          << format("%.3f%%", Unique.ValueCounts[I] * 100) << "\n";
1281     OS << "  " << ProfileKindName
1282        << " profile base count sum: " << format("%.0f", Base.ValueCounts[I])
1283        << "\n"
1284        << "  " << ProfileKindName
1285        << " profile test count sum: " << format("%.0f", Test.ValueCounts[I])
1286        << "\n";
1287   }
1288 }
1289
1290 } // end namespace llvm