1 //=-- InstrProf.cpp - Instrumented profiling format support -----------------=//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file contains support for clang's instrumentation based PGO and
11 // coverage.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "llvm/ProfileData/InstrProf.h"
16 #include "llvm/ADT/StringExtras.h"
17 #include "llvm/ADT/Triple.h"
18 #include "llvm/IR/Constants.h"
19 #include "llvm/IR/Function.h"
20 #include "llvm/IR/GlobalVariable.h"
21 #include "llvm/IR/MDBuilder.h"
22 #include "llvm/IR/Module.h"
23 #include "llvm/Support/Compression.h"
24 #include "llvm/Support/ErrorHandling.h"
25 #include "llvm/Support/LEB128.h"
26 #include "llvm/Support/ManagedStatic.h"
27 #include "llvm/Support/Path.h"
28 
29 using namespace llvm;
30 
31 static cl::opt<bool> StaticFuncFullModulePrefix(
32     "static-func-full-module-prefix", cl::init(false),
33     cl::desc("Use full module build paths in the profile counter names for "
34              "static functions."));
35 
36 namespace {
37 std::string getInstrProfErrString(instrprof_error Err) {
38   switch (Err) {
39   case instrprof_error::success:
40     return "Success";
41   case instrprof_error::eof:
42     return "End of File";
43   case instrprof_error::unrecognized_format:
44     return "Unrecognized instrumentation profile encoding format";
45   case instrprof_error::bad_magic:
46     return "Invalid instrumentation profile data (bad magic)";
47   case instrprof_error::bad_header:
48     return "Invalid instrumentation profile data (file header is corrupt)";
49   case instrprof_error::unsupported_version:
50     return "Unsupported instrumentation profile format version";
51   case instrprof_error::unsupported_hash_type:
52     return "Unsupported instrumentation profile hash type";
53   case instrprof_error::too_large:
54     return "Too much profile data";
55   case instrprof_error::truncated:
56     return "Truncated profile data";
57   case instrprof_error::malformed:
58     return "Malformed instrumentation profile data";
59   case instrprof_error::unknown_function:
60     return "No profile data available for function";
61   case instrprof_error::hash_mismatch:
62     return "Function control flow change detected (hash mismatch)";
63   case instrprof_error::count_mismatch:
64     return "Function basic block count change detected (counter mismatch)";
65   case instrprof_error::counter_overflow:
66     return "Counter overflow";
67   case instrprof_error::value_site_count_mismatch:
68     return "Function value site count change detected (counter mismatch)";
69   case instrprof_error::compress_failed:
70     return "Failed to compress data (zlib)";
71   case instrprof_error::uncompress_failed:
72     return "Failed to uncompress data (zlib)";
73   case instrprof_error::empty_raw_profile:
74     return "Empty raw profile file";
75   }
76   llvm_unreachable("A value of instrprof_error has no message.");
77 }
78 
79 // FIXME: This class is only here to support the transition to llvm::Error. It
80 // will be removed once this transition is complete. Clients should prefer to
81 // deal with the Error value directly, rather than converting to error_code.
82 class InstrProfErrorCategoryType : public std::error_category {
83   const char *name() const noexcept override { return "llvm.instrprof"; }
84   std::string message(int IE) const override {
85     return getInstrProfErrString(static_cast<instrprof_error>(IE));
86   }
87 };
88 } // end anonymous namespace
89 
90 static ManagedStatic<InstrProfErrorCategoryType> ErrorCategory;
91 
92 const std::error_category &llvm::instrprof_category() {
93   return *ErrorCategory;
94 }
95 
96 namespace llvm {
97 
98 void SoftInstrProfErrors::addError(instrprof_error IE) {
99   if (IE == instrprof_error::success)
100     return;
101 
102   if (FirstError == instrprof_error::success)
103     FirstError = IE;
104 
105   switch (IE) {
106   case instrprof_error::hash_mismatch:
107     ++NumHashMismatches;
108     break;
109   case instrprof_error::count_mismatch:
110     ++NumCountMismatches;
111     break;
112   case instrprof_error::counter_overflow:
113     ++NumCounterOverflows;
114     break;
115   case instrprof_error::value_site_count_mismatch:
116     ++NumValueSiteCountMismatches;
117     break;
118   default:
119     llvm_unreachable("Not a soft error");
120   }
121 }
122 
123 std::string InstrProfError::message() const {
124   return getInstrProfErrString(Err);
125 }
126 
127 char InstrProfError::ID = 0;
128 
129 std::string getPGOFuncName(StringRef RawFuncName,
130                            GlobalValue::LinkageTypes Linkage,
131                            StringRef FileName,
132                            uint64_t Version LLVM_ATTRIBUTE_UNUSED) {
133   return GlobalValue::getGlobalIdentifier(RawFuncName, Linkage, FileName);
134 }
135 
136 // Return the PGOFuncName. This function has some special handling when called
137 // in LTO optimization. The following only applies when calling in LTO passes
138 // (when \c InLTO is true): LTO's internalization privatizes many global linkage
139 // symbols. This happens after value profile annotation, but those internal
140 // linkage functions should not have a source prefix.
141 // Additionally, for ThinLTO mode, exported internal functions are promoted
142 // and renamed. We need to ensure that the original internal PGO name is
143 // used when computing the GUID that is compared against the profiled GUIDs.
144 // To differentiate compiler generated internal symbols from original ones,
145 // PGOFuncName meta data are created and attached to the original internal
146 // symbols in the value profile annotation step
147 // (PGOUseFunc::annotateIndirectCallSites). If a symbol does not have the meta
148 // data, its original linkage must be non-internal.
149 std::string getPGOFuncName(const Function &F, bool InLTO, uint64_t Version) {
150   if (!InLTO) {
151     StringRef FileName = (StaticFuncFullModulePrefix
152                               ? F.getParent()->getName()
153                               : sys::path::filename(F.getParent()->getName()));
154     return getPGOFuncName(F.getName(), F.getLinkage(), FileName, Version);
155   }
156 
157   // In LTO mode (when InLTO is true), first check if there is a meta data.
158   if (MDNode *MD = getPGOFuncNameMetadata(F)) {
159     StringRef S = cast<MDString>(MD->getOperand(0))->getString();
160     return S.str();
161   }
162 
163   // If there is no meta data, the function must be a global before the value
164   // profile annotation pass. Its current linkage may be internal if it is
165   // internalized in LTO mode.
166   return getPGOFuncName(F.getName(), GlobalValue::ExternalLinkage, "");
167 }
168 
169 StringRef getFuncNameWithoutPrefix(StringRef PGOFuncName, StringRef FileName) {
170   if (FileName.empty())
171     return PGOFuncName;
172   // Drop the file name including ':'. See also getPGOFuncName.
173   if (PGOFuncName.startswith(FileName))
174     PGOFuncName = PGOFuncName.drop_front(FileName.size() + 1);
175   return PGOFuncName;
176 }
177 
178 // \p FuncName is the string used as profile lookup key for the function. A
179 // symbol is created to hold the name. Return the legalized symbol name.
180 std::string getPGOFuncNameVarName(StringRef FuncName,
181                                   GlobalValue::LinkageTypes Linkage) {
182   std::string VarName = getInstrProfNameVarPrefix();
183   VarName += FuncName;
184 
185   if (!GlobalValue::isLocalLinkage(Linkage))
186     return VarName;
187 
188   // Now fix up illegal chars in local VarName that may upset the assembler.
189   const char *InvalidChars = "-:<>/\"'";
190   size_t found = VarName.find_first_of(InvalidChars);
191   while (found != std::string::npos) {
192     VarName[found] = '_';
193     found = VarName.find_first_of(InvalidChars, found + 1);
194   }
195   return VarName;
196 }
197 
198 GlobalVariable *createPGOFuncNameVar(Module &M,
199                                      GlobalValue::LinkageTypes Linkage,
200                                      StringRef PGOFuncName) {
201 
202   // We generally want to match the function's linkage, but available_externally
203   // and extern_weak both have the wrong semantics, and anything that doesn't
204   // need to link across compilation units doesn't need to be visible at all.
205   if (Linkage == GlobalValue::ExternalWeakLinkage)
206     Linkage = GlobalValue::LinkOnceAnyLinkage;
207   else if (Linkage == GlobalValue::AvailableExternallyLinkage)
208     Linkage = GlobalValue::LinkOnceODRLinkage;
209   else if (Linkage == GlobalValue::InternalLinkage ||
210            Linkage == GlobalValue::ExternalLinkage)
211     Linkage = GlobalValue::PrivateLinkage;
212 
213   auto *Value =
214       ConstantDataArray::getString(M.getContext(), PGOFuncName, false);
215   auto FuncNameVar =
216       new GlobalVariable(M, Value->getType(), true, Linkage, Value,
217                          getPGOFuncNameVarName(PGOFuncName, Linkage));
218 
219   // Hide the symbol so that we correctly get a copy for each executable.
220   if (!GlobalValue::isLocalLinkage(FuncNameVar->getLinkage()))
221     FuncNameVar->setVisibility(GlobalValue::HiddenVisibility);
222 
223   return FuncNameVar;
224 }
225 
226 GlobalVariable *createPGOFuncNameVar(Function &F, StringRef PGOFuncName) {
227   return createPGOFuncNameVar(*F.getParent(), F.getLinkage(), PGOFuncName);
228 }
229 
230 void InstrProfSymtab::create(Module &M, bool InLTO) {
231   for (Function &F : M) {
232     // Function may not have a name: like using asm("") to overwrite the name.
233     // Ignore in this case.
234     if (!F.hasName())
235       continue;
236     const std::string &PGOFuncName = getPGOFuncName(F, InLTO);
237     addFuncName(PGOFuncName);
238     MD5FuncMap.emplace_back(Function::getGUID(PGOFuncName), &F);
239   }
240 
241   finalizeSymtab();
242 }
243 
244 Error collectPGOFuncNameStrings(const std::vector<std::string> &NameStrs,
245                                 bool doCompression, std::string &Result) {
246   assert(NameStrs.size() && "No name data to emit");
247 
248   uint8_t Header[16], *P = Header;
249   std::string UncompressedNameStrings =
250       join(NameStrs.begin(), NameStrs.end(), getInstrProfNameSeparator());
251 
252   assert(StringRef(UncompressedNameStrings)
253                  .count(getInstrProfNameSeparator()) == (NameStrs.size() - 1) &&
254          "PGO name is invalid (contains separator token)");
255 
256   unsigned EncLen = encodeULEB128(UncompressedNameStrings.length(), P);
257   P += EncLen;
258 
259   auto WriteStringToResult = [&](size_t CompressedLen, StringRef InputStr) {
260     EncLen = encodeULEB128(CompressedLen, P);
261     P += EncLen;
262     char *HeaderStr = reinterpret_cast<char *>(&Header[0]);
263     unsigned HeaderLen = P - &Header[0];
264     Result.append(HeaderStr, HeaderLen);
265     Result += InputStr;
266     return Error::success();
267   };
268 
269   if (!doCompression) {
270     return WriteStringToResult(0, UncompressedNameStrings);
271   }
272 
273   SmallString<128> CompressedNameStrings;
274   zlib::Status Success =
275       zlib::compress(StringRef(UncompressedNameStrings), CompressedNameStrings,
276                      zlib::BestSizeCompression);
277 
278   if (Success != zlib::StatusOK)
279     return make_error<InstrProfError>(instrprof_error::compress_failed);
280 
281   return WriteStringToResult(CompressedNameStrings.size(),
282                              CompressedNameStrings);
283 }
284 
285 StringRef getPGOFuncNameVarInitializer(GlobalVariable *NameVar) {
286   auto *Arr = cast<ConstantDataArray>(NameVar->getInitializer());
287   StringRef NameStr =
288       Arr->isCString() ? Arr->getAsCString() : Arr->getAsString();
289   return NameStr;
290 }
291 
292 Error collectPGOFuncNameStrings(const std::vector<GlobalVariable *> &NameVars,
293                                 std::string &Result, bool doCompression) {
294   std::vector<std::string> NameStrs;
295   for (auto *NameVar : NameVars) {
296     NameStrs.push_back(getPGOFuncNameVarInitializer(NameVar));
297   }
298   return collectPGOFuncNameStrings(
299       NameStrs, zlib::isAvailable() && doCompression, Result);
300 }
301 
302 Error readPGOFuncNameStrings(StringRef NameStrings, InstrProfSymtab &Symtab) {
303   const uint8_t *P = reinterpret_cast<const uint8_t *>(NameStrings.data());
304   const uint8_t *EndP = reinterpret_cast<const uint8_t *>(NameStrings.data() +
305                                                           NameStrings.size());
306   while (P < EndP) {
307     uint32_t N;
308     uint64_t UncompressedSize = decodeULEB128(P, &N);
309     P += N;
310     uint64_t CompressedSize = decodeULEB128(P, &N);
311     P += N;
312     bool isCompressed = (CompressedSize != 0);
313     SmallString<128> UncompressedNameStrings;
314     StringRef NameStrings;
315     if (isCompressed) {
316       StringRef CompressedNameStrings(reinterpret_cast<const char *>(P),
317                                       CompressedSize);
318       if (zlib::uncompress(CompressedNameStrings, UncompressedNameStrings,
319                            UncompressedSize) != zlib::StatusOK)
320         return make_error<InstrProfError>(instrprof_error::uncompress_failed);
321       P += CompressedSize;
322       NameStrings = StringRef(UncompressedNameStrings.data(),
323                               UncompressedNameStrings.size());
324     } else {
325       NameStrings =
326           StringRef(reinterpret_cast<const char *>(P), UncompressedSize);
327       P += UncompressedSize;
328     }
329     // Now parse the name strings.
330     SmallVector<StringRef, 0> Names;
331     NameStrings.split(Names, getInstrProfNameSeparator());
332     for (StringRef &Name : Names)
333       Symtab.addFuncName(Name);
334 
335     while (P < EndP && *P == 0)
336       P++;
337   }
338   Symtab.finalizeSymtab();
339   return Error::success();
340 }
341 
342 void InstrProfValueSiteRecord::merge(SoftInstrProfErrors &SIPE,
343                                      InstrProfValueSiteRecord &Input,
344                                      uint64_t Weight) {
345   this->sortByTargetValues();
346   Input.sortByTargetValues();
347   auto I = ValueData.begin();
348   auto IE = ValueData.end();
349   for (auto J = Input.ValueData.begin(), JE = Input.ValueData.end(); J != JE;
350        ++J) {
351     while (I != IE && I->Value < J->Value)
352       ++I;
353     if (I != IE && I->Value == J->Value) {
354       bool Overflowed;
355       I->Count = SaturatingMultiplyAdd(J->Count, Weight, I->Count, &Overflowed);
356       if (Overflowed)
357         SIPE.addError(instrprof_error::counter_overflow);
358       ++I;
359       continue;
360     }
361     ValueData.insert(I, *J);
362   }
363 }
364 
365 void InstrProfValueSiteRecord::scale(SoftInstrProfErrors &SIPE,
366                                      uint64_t Weight) {
367   for (auto I = ValueData.begin(), IE = ValueData.end(); I != IE; ++I) {
368     bool Overflowed;
369     I->Count = SaturatingMultiply(I->Count, Weight, &Overflowed);
370     if (Overflowed)
371       SIPE.addError(instrprof_error::counter_overflow);
372   }
373 }
374 
375 // Merge Value Profile data from Src record to this record for ValueKind.
376 // Scale merged value counts by \p Weight.
377 void InstrProfRecord::mergeValueProfData(uint32_t ValueKind,
378                                          InstrProfRecord &Src,
379                                          uint64_t Weight) {
380   uint32_t ThisNumValueSites = getNumValueSites(ValueKind);
381   uint32_t OtherNumValueSites = Src.getNumValueSites(ValueKind);
382   if (ThisNumValueSites != OtherNumValueSites) {
383     SIPE.addError(instrprof_error::value_site_count_mismatch);
384     return;
385   }
386   std::vector<InstrProfValueSiteRecord> &ThisSiteRecords =
387       getValueSitesForKind(ValueKind);
388   std::vector<InstrProfValueSiteRecord> &OtherSiteRecords =
389       Src.getValueSitesForKind(ValueKind);
390   for (uint32_t I = 0; I < ThisNumValueSites; I++)
391     ThisSiteRecords[I].merge(SIPE, OtherSiteRecords[I], Weight);
392 }
393 
394 void InstrProfRecord::merge(InstrProfRecord &Other, uint64_t Weight) {
395   // If the number of counters doesn't match we either have bad data
396   // or a hash collision.
397   if (Counts.size() != Other.Counts.size()) {
398     SIPE.addError(instrprof_error::count_mismatch);
399     return;
400   }
401 
402   for (size_t I = 0, E = Other.Counts.size(); I < E; ++I) {
403     bool Overflowed;
404     Counts[I] =
405         SaturatingMultiplyAdd(Other.Counts[I], Weight, Counts[I], &Overflowed);
406     if (Overflowed)
407       SIPE.addError(instrprof_error::counter_overflow);
408   }
409 
410   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
411     mergeValueProfData(Kind, Other, Weight);
412 }
413 
414 void InstrProfRecord::scaleValueProfData(uint32_t ValueKind, uint64_t Weight) {
415   uint32_t ThisNumValueSites = getNumValueSites(ValueKind);
416   std::vector<InstrProfValueSiteRecord> &ThisSiteRecords =
417       getValueSitesForKind(ValueKind);
418   for (uint32_t I = 0; I < ThisNumValueSites; I++)
419     ThisSiteRecords[I].scale(SIPE, Weight);
420 }
421 
422 void InstrProfRecord::scale(uint64_t Weight) {
423   for (auto &Count : this->Counts) {
424     bool Overflowed;
425     Count = SaturatingMultiply(Count, Weight, &Overflowed);
426     if (Overflowed)
427       SIPE.addError(instrprof_error::counter_overflow);
428   }
429   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
430     scaleValueProfData(Kind, Weight);
431 }
432 
433 // Map indirect call target name hash to name string.
434 uint64_t InstrProfRecord::remapValue(uint64_t Value, uint32_t ValueKind,
435                                      ValueMapType *ValueMap) {
436   if (!ValueMap)
437     return Value;
438   switch (ValueKind) {
439   case IPVK_IndirectCallTarget: {
440     auto Result =
441         std::lower_bound(ValueMap->begin(), ValueMap->end(), Value,
442                          [](const std::pair<uint64_t, uint64_t> &LHS,
443                             uint64_t RHS) { return LHS.first < RHS; });
444    // Raw function pointer collected by value profiler may be from
445    // external functions that are not instrumented. They won't have
446    // mapping data to be used by the deserializer. Force the value to
447    // be 0 in this case.
448     if (Result != ValueMap->end() && Result->first == Value)
449       Value = (uint64_t)Result->second;
450     else
451       Value = 0;
452     break;
453   }
454   }
455   return Value;
456 }
457 
458 void InstrProfRecord::addValueData(uint32_t ValueKind, uint32_t Site,
459                                    InstrProfValueData *VData, uint32_t N,
460                                    ValueMapType *ValueMap) {
461   for (uint32_t I = 0; I < N; I++) {
462     VData[I].Value = remapValue(VData[I].Value, ValueKind, ValueMap);
463   }
464   std::vector<InstrProfValueSiteRecord> &ValueSites =
465       getValueSitesForKind(ValueKind);
466   if (N == 0)
467     ValueSites.emplace_back();
468   else
469     ValueSites.emplace_back(VData, VData + N);
470 }
471 
472 #define INSTR_PROF_COMMON_API_IMPL
473 #include "llvm/ProfileData/InstrProfData.inc"
474 
475 /*!
476  * \brief ValueProfRecordClosure Interface implementation for  InstrProfRecord
477  *  class. These C wrappers are used as adaptors so that C++ code can be
478  *  invoked as callbacks.
479  */
480 uint32_t getNumValueKindsInstrProf(const void *Record) {
481   return reinterpret_cast<const InstrProfRecord *>(Record)->getNumValueKinds();
482 }
483 
484 uint32_t getNumValueSitesInstrProf(const void *Record, uint32_t VKind) {
485   return reinterpret_cast<const InstrProfRecord *>(Record)
486       ->getNumValueSites(VKind);
487 }
488 
489 uint32_t getNumValueDataInstrProf(const void *Record, uint32_t VKind) {
490   return reinterpret_cast<const InstrProfRecord *>(Record)
491       ->getNumValueData(VKind);
492 }
493 
494 uint32_t getNumValueDataForSiteInstrProf(const void *R, uint32_t VK,
495                                          uint32_t S) {
496   return reinterpret_cast<const InstrProfRecord *>(R)
497       ->getNumValueDataForSite(VK, S);
498 }
499 
500 void getValueForSiteInstrProf(const void *R, InstrProfValueData *Dst,
501                               uint32_t K, uint32_t S) {
502   reinterpret_cast<const InstrProfRecord *>(R)->getValueForSite(Dst, K, S);
503 }
504 
505 ValueProfData *allocValueProfDataInstrProf(size_t TotalSizeInBytes) {
506   ValueProfData *VD =
507       (ValueProfData *)(new (::operator new(TotalSizeInBytes)) ValueProfData());
508   memset(VD, 0, TotalSizeInBytes);
509   return VD;
510 }
511 
512 static ValueProfRecordClosure InstrProfRecordClosure = {
513     nullptr,
514     getNumValueKindsInstrProf,
515     getNumValueSitesInstrProf,
516     getNumValueDataInstrProf,
517     getNumValueDataForSiteInstrProf,
518     nullptr,
519     getValueForSiteInstrProf,
520     allocValueProfDataInstrProf};
521 
522 // Wrapper implementation using the closure mechanism.
523 uint32_t ValueProfData::getSize(const InstrProfRecord &Record) {
524   InstrProfRecordClosure.Record = &Record;
525   return getValueProfDataSize(&InstrProfRecordClosure);
526 }
527 
528 // Wrapper implementation using the closure mechanism.
529 std::unique_ptr<ValueProfData>
530 ValueProfData::serializeFrom(const InstrProfRecord &Record) {
531   InstrProfRecordClosure.Record = &Record;
532 
533   std::unique_ptr<ValueProfData> VPD(
534       serializeValueProfDataFrom(&InstrProfRecordClosure, nullptr));
535   return VPD;
536 }
537 
538 void ValueProfRecord::deserializeTo(InstrProfRecord &Record,
539                                     InstrProfRecord::ValueMapType *VMap) {
540   Record.reserveSites(Kind, NumValueSites);
541 
542   InstrProfValueData *ValueData = getValueProfRecordValueData(this);
543   for (uint64_t VSite = 0; VSite < NumValueSites; ++VSite) {
544     uint8_t ValueDataCount = this->SiteCountArray[VSite];
545     Record.addValueData(Kind, VSite, ValueData, ValueDataCount, VMap);
546     ValueData += ValueDataCount;
547   }
548 }
549 
550 // For writing/serializing,  Old is the host endianness, and  New is
551 // byte order intended on disk. For Reading/deserialization, Old
552 // is the on-disk source endianness, and New is the host endianness.
553 void ValueProfRecord::swapBytes(support::endianness Old,
554                                 support::endianness New) {
555   using namespace support;
556   if (Old == New)
557     return;
558 
559   if (getHostEndianness() != Old) {
560     sys::swapByteOrder<uint32_t>(NumValueSites);
561     sys::swapByteOrder<uint32_t>(Kind);
562   }
563   uint32_t ND = getValueProfRecordNumValueData(this);
564   InstrProfValueData *VD = getValueProfRecordValueData(this);
565 
566   // No need to swap byte array: SiteCountArrray.
567   for (uint32_t I = 0; I < ND; I++) {
568     sys::swapByteOrder<uint64_t>(VD[I].Value);
569     sys::swapByteOrder<uint64_t>(VD[I].Count);
570   }
571   if (getHostEndianness() == Old) {
572     sys::swapByteOrder<uint32_t>(NumValueSites);
573     sys::swapByteOrder<uint32_t>(Kind);
574   }
575 }
576 
577 void ValueProfData::deserializeTo(InstrProfRecord &Record,
578                                   InstrProfRecord::ValueMapType *VMap) {
579   if (NumValueKinds == 0)
580     return;
581 
582   ValueProfRecord *VR = getFirstValueProfRecord(this);
583   for (uint32_t K = 0; K < NumValueKinds; K++) {
584     VR->deserializeTo(Record, VMap);
585     VR = getValueProfRecordNext(VR);
586   }
587 }
588 
589 template <class T>
590 static T swapToHostOrder(const unsigned char *&D, support::endianness Orig) {
591   using namespace support;
592   if (Orig == little)
593     return endian::readNext<T, little, unaligned>(D);
594   else
595     return endian::readNext<T, big, unaligned>(D);
596 }
597 
598 static std::unique_ptr<ValueProfData> allocValueProfData(uint32_t TotalSize) {
599   return std::unique_ptr<ValueProfData>(new (::operator new(TotalSize))
600                                             ValueProfData());
601 }
602 
603 Error ValueProfData::checkIntegrity() {
604   if (NumValueKinds > IPVK_Last + 1)
605     return make_error<InstrProfError>(instrprof_error::malformed);
606   // Total size needs to be mulltiple of quadword size.
607   if (TotalSize % sizeof(uint64_t))
608     return make_error<InstrProfError>(instrprof_error::malformed);
609 
610   ValueProfRecord *VR = getFirstValueProfRecord(this);
611   for (uint32_t K = 0; K < this->NumValueKinds; K++) {
612     if (VR->Kind > IPVK_Last)
613       return make_error<InstrProfError>(instrprof_error::malformed);
614     VR = getValueProfRecordNext(VR);
615     if ((char *)VR - (char *)this > (ptrdiff_t)TotalSize)
616       return make_error<InstrProfError>(instrprof_error::malformed);
617   }
618   return Error::success();
619 }
620 
621 Expected<std::unique_ptr<ValueProfData>>
622 ValueProfData::getValueProfData(const unsigned char *D,
623                                 const unsigned char *const BufferEnd,
624                                 support::endianness Endianness) {
625   using namespace support;
626   if (D + sizeof(ValueProfData) > BufferEnd)
627     return make_error<InstrProfError>(instrprof_error::truncated);
628 
629   const unsigned char *Header = D;
630   uint32_t TotalSize = swapToHostOrder<uint32_t>(Header, Endianness);
631   if (D + TotalSize > BufferEnd)
632     return make_error<InstrProfError>(instrprof_error::too_large);
633 
634   std::unique_ptr<ValueProfData> VPD = allocValueProfData(TotalSize);
635   memcpy(VPD.get(), D, TotalSize);
636   // Byte swap.
637   VPD->swapBytesToHost(Endianness);
638 
639   Error E = VPD->checkIntegrity();
640   if (E)
641     return std::move(E);
642 
643   return std::move(VPD);
644 }
645 
646 void ValueProfData::swapBytesToHost(support::endianness Endianness) {
647   using namespace support;
648   if (Endianness == getHostEndianness())
649     return;
650 
651   sys::swapByteOrder<uint32_t>(TotalSize);
652   sys::swapByteOrder<uint32_t>(NumValueKinds);
653 
654   ValueProfRecord *VR = getFirstValueProfRecord(this);
655   for (uint32_t K = 0; K < NumValueKinds; K++) {
656     VR->swapBytes(Endianness, getHostEndianness());
657     VR = getValueProfRecordNext(VR);
658   }
659 }
660 
661 void ValueProfData::swapBytesFromHost(support::endianness Endianness) {
662   using namespace support;
663   if (Endianness == getHostEndianness())
664     return;
665 
666   ValueProfRecord *VR = getFirstValueProfRecord(this);
667   for (uint32_t K = 0; K < NumValueKinds; K++) {
668     ValueProfRecord *NVR = getValueProfRecordNext(VR);
669     VR->swapBytes(getHostEndianness(), Endianness);
670     VR = NVR;
671   }
672   sys::swapByteOrder<uint32_t>(TotalSize);
673   sys::swapByteOrder<uint32_t>(NumValueKinds);
674 }
675 
676 void annotateValueSite(Module &M, Instruction &Inst,
677                        const InstrProfRecord &InstrProfR,
678                        InstrProfValueKind ValueKind, uint32_t SiteIdx,
679                        uint32_t MaxMDCount) {
680   uint32_t NV = InstrProfR.getNumValueDataForSite(ValueKind, SiteIdx);
681   if (!NV)
682     return;
683 
684   uint64_t Sum = 0;
685   std::unique_ptr<InstrProfValueData[]> VD =
686       InstrProfR.getValueForSite(ValueKind, SiteIdx, &Sum);
687 
688   ArrayRef<InstrProfValueData> VDs(VD.get(), NV);
689   annotateValueSite(M, Inst, VDs, Sum, ValueKind, MaxMDCount);
690 }
691 
692 void annotateValueSite(Module &M, Instruction &Inst,
693                        ArrayRef<InstrProfValueData> VDs,
694                        uint64_t Sum, InstrProfValueKind ValueKind,
695                        uint32_t MaxMDCount) {
696   LLVMContext &Ctx = M.getContext();
697   MDBuilder MDHelper(Ctx);
698   SmallVector<Metadata *, 3> Vals;
699   // Tag
700   Vals.push_back(MDHelper.createString("VP"));
701   // Value Kind
702   Vals.push_back(MDHelper.createConstant(
703       ConstantInt::get(Type::getInt32Ty(Ctx), ValueKind)));
704   // Total Count
705   Vals.push_back(
706       MDHelper.createConstant(ConstantInt::get(Type::getInt64Ty(Ctx), Sum)));
707 
708   // Value Profile Data
709   uint32_t MDCount = MaxMDCount;
710   for (auto &VD : VDs) {
711     Vals.push_back(MDHelper.createConstant(
712         ConstantInt::get(Type::getInt64Ty(Ctx), VD.Value)));
713     Vals.push_back(MDHelper.createConstant(
714         ConstantInt::get(Type::getInt64Ty(Ctx), VD.Count)));
715     if (--MDCount == 0)
716       break;
717   }
718   Inst.setMetadata(LLVMContext::MD_prof, MDNode::get(Ctx, Vals));
719 }
720 
721 bool getValueProfDataFromInst(const Instruction &Inst,
722                               InstrProfValueKind ValueKind,
723                               uint32_t MaxNumValueData,
724                               InstrProfValueData ValueData[],
725                               uint32_t &ActualNumValueData, uint64_t &TotalC) {
726   MDNode *MD = Inst.getMetadata(LLVMContext::MD_prof);
727   if (!MD)
728     return false;
729 
730   unsigned NOps = MD->getNumOperands();
731 
732   if (NOps < 5)
733     return false;
734 
735   // Operand 0 is a string tag "VP":
736   MDString *Tag = cast<MDString>(MD->getOperand(0));
737   if (!Tag)
738     return false;
739 
740   if (!Tag->getString().equals("VP"))
741     return false;
742 
743   // Now check kind:
744   ConstantInt *KindInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(1));
745   if (!KindInt)
746     return false;
747   if (KindInt->getZExtValue() != ValueKind)
748     return false;
749 
750   // Get total count
751   ConstantInt *TotalCInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(2));
752   if (!TotalCInt)
753     return false;
754   TotalC = TotalCInt->getZExtValue();
755 
756   ActualNumValueData = 0;
757 
758   for (unsigned I = 3; I < NOps; I += 2) {
759     if (ActualNumValueData >= MaxNumValueData)
760       break;
761     ConstantInt *Value = mdconst::dyn_extract<ConstantInt>(MD->getOperand(I));
762     ConstantInt *Count =
763         mdconst::dyn_extract<ConstantInt>(MD->getOperand(I + 1));
764     if (!Value || !Count)
765       return false;
766     ValueData[ActualNumValueData].Value = Value->getZExtValue();
767     ValueData[ActualNumValueData].Count = Count->getZExtValue();
768     ActualNumValueData++;
769   }
770   return true;
771 }
772 
773 MDNode *getPGOFuncNameMetadata(const Function &F) {
774   return F.getMetadata(getPGOFuncNameMetadataName());
775 }
776 
777 void createPGOFuncNameMetadata(Function &F, StringRef PGOFuncName) {
778   // Only for internal linkage functions.
779   if (PGOFuncName == F.getName())
780       return;
781   // Don't create duplicated meta-data.
782   if (getPGOFuncNameMetadata(F))
783     return;
784   LLVMContext &C = F.getContext();
785   MDNode *N = MDNode::get(C, MDString::get(C, PGOFuncName));
786   F.setMetadata(getPGOFuncNameMetadataName(), N);
787 }
788 
789 bool needsComdatForCounter(const Function &F, const Module &M) {
790   if (F.hasComdat())
791     return true;
792 
793   Triple TT(M.getTargetTriple());
794   if (!TT.isOSBinFormatELF())
795     return false;
796 
797   // See createPGOFuncNameVar for more details. To avoid link errors, profile
798   // counters for function with available_externally linkage needs to be changed
799   // to linkonce linkage. On ELF based systems, this leads to weak symbols to be
800   // created. Without using comdat, duplicate entries won't be removed by the
801   // linker leading to increased data segement size and raw profile size. Even
802   // worse, since the referenced counter from profile per-function data object
803   // will be resolved to the common strong definition, the profile counts for
804   // available_externally functions will end up being duplicated in raw profile
805   // data. This can result in distorted profile as the counts of those dups
806   // will be accumulated by the profile merger.
807   GlobalValue::LinkageTypes Linkage = F.getLinkage();
808   if (Linkage != GlobalValue::ExternalWeakLinkage &&
809       Linkage != GlobalValue::AvailableExternallyLinkage)
810     return false;
811 
812   return true;
813 }
814 } // end namespace llvm
815