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/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/IR/Constant.h"
22 #include "llvm/IR/Constants.h"
23 #include "llvm/IR/Function.h"
24 #include "llvm/IR/GlobalValue.h"
25 #include "llvm/IR/GlobalVariable.h"
26 #include "llvm/IR/Instruction.h"
27 #include "llvm/IR/LLVMContext.h"
28 #include "llvm/IR/MDBuilder.h"
29 #include "llvm/IR/Metadata.h"
30 #include "llvm/IR/Module.h"
31 #include "llvm/IR/Type.h"
32 #include "llvm/ProfileData/InstrProf.h"
33 #include "llvm/Support/Casting.h"
34 #include "llvm/Support/CommandLine.h"
35 #include "llvm/Support/Compiler.h"
36 #include "llvm/Support/Compression.h"
37 #include "llvm/Support/Endian.h"
38 #include "llvm/Support/Error.h"
39 #include "llvm/Support/ErrorHandling.h"
40 #include "llvm/Support/LEB128.h"
41 #include "llvm/Support/ManagedStatic.h"
42 #include "llvm/Support/MathExtras.h"
43 #include "llvm/Support/Path.h"
44 #include "llvm/Support/SwapByteOrder.h"
45 #include <algorithm>
46 #include <cassert>
47 #include <cstddef>
48 #include <cstring>
49 #include <cstdint>
50 #include <memory>
51 #include <string>
52 #include <system_error>
53 #include <utility>
54 #include <vector>
55 
56 using namespace llvm;
57 
58 static cl::opt<bool> StaticFuncFullModulePrefix(
59     "static-func-full-module-prefix", cl::init(true),
60     cl::desc("Use full module build paths in the profile counter names for "
61              "static functions."));
62 
63 // This option is tailored to users that have different top-level directory in
64 // profile-gen and profile-use compilation. Users need to specific the number
65 // of levels to strip. A value larger than the number of directories in the
66 // source file will strip all the directory names and only leave the basename.
67 //
68 // Note current ThinLTO module importing for the indirect-calls assumes
69 // the source directory name not being stripped. A non-zero option value here
70 // can potentially prevent some inter-module indirect-call-promotions.
71 static cl::opt<unsigned> StaticFuncStripDirNamePrefix(
72     "static-func-strip-dirname-prefix", cl::init(0),
73     cl::desc("Strip specified level of directory name from source path in "
74              "the profile counter name for static functions."));
75 
76 static std::string getInstrProfErrString(instrprof_error Err) {
77   switch (Err) {
78   case instrprof_error::success:
79     return "Success";
80   case instrprof_error::eof:
81     return "End of File";
82   case instrprof_error::unrecognized_format:
83     return "Unrecognized instrumentation profile encoding format";
84   case instrprof_error::bad_magic:
85     return "Invalid instrumentation profile data (bad magic)";
86   case instrprof_error::bad_header:
87     return "Invalid instrumentation profile data (file header is corrupt)";
88   case instrprof_error::unsupported_version:
89     return "Unsupported instrumentation profile format version";
90   case instrprof_error::unsupported_hash_type:
91     return "Unsupported instrumentation profile hash type";
92   case instrprof_error::too_large:
93     return "Too much profile data";
94   case instrprof_error::truncated:
95     return "Truncated profile data";
96   case instrprof_error::malformed:
97     return "Malformed instrumentation profile data";
98   case instrprof_error::unknown_function:
99     return "No profile data available for function";
100   case instrprof_error::hash_mismatch:
101     return "Function control flow change detected (hash mismatch)";
102   case instrprof_error::count_mismatch:
103     return "Function basic block count change detected (counter mismatch)";
104   case instrprof_error::counter_overflow:
105     return "Counter overflow";
106   case instrprof_error::value_site_count_mismatch:
107     return "Function value site count change detected (counter mismatch)";
108   case instrprof_error::compress_failed:
109     return "Failed to compress data (zlib)";
110   case instrprof_error::uncompress_failed:
111     return "Failed to uncompress data (zlib)";
112   case instrprof_error::empty_raw_profile:
113     return "Empty raw profile file";
114   }
115   llvm_unreachable("A value of instrprof_error has no message.");
116 }
117 
118 namespace {
119 
120 // FIXME: This class is only here to support the transition to llvm::Error. It
121 // will be removed once this transition is complete. Clients should prefer to
122 // deal with the Error value directly, rather than converting to error_code.
123 class InstrProfErrorCategoryType : public std::error_category {
124   const char *name() const noexcept override { return "llvm.instrprof"; }
125 
126   std::string message(int IE) const override {
127     return getInstrProfErrString(static_cast<instrprof_error>(IE));
128   }
129 };
130 
131 } // end anonymous namespace
132 
133 static ManagedStatic<InstrProfErrorCategoryType> ErrorCategory;
134 
135 const std::error_category &llvm::instrprof_category() {
136   return *ErrorCategory;
137 }
138 
139 namespace {
140 
141 const char *InstrProfSectNameCommon[] = {
142 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
143   SectNameCommon,
144 #include "llvm/ProfileData/InstrProfData.inc"
145 };
146 
147 const char *InstrProfSectNameCoff[] = {
148 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
149   SectNameCoff,
150 #include "llvm/ProfileData/InstrProfData.inc"
151 };
152 
153 const char *InstrProfSectNamePrefix[] = {
154 #define INSTR_PROF_SECT_ENTRY(Kind, SectNameCommon, SectNameCoff, Prefix)      \
155   Prefix,
156 #include "llvm/ProfileData/InstrProfData.inc"
157 };
158 
159 } // namespace
160 
161 namespace llvm {
162 
163 std::string getInstrProfSectionName(InstrProfSectKind IPSK,
164                                     Triple::ObjectFormatType OF,
165                                     bool AddSegmentInfo) {
166   std::string SectName;
167 
168   if (OF == Triple::MachO && AddSegmentInfo)
169     SectName = InstrProfSectNamePrefix[IPSK];
170 
171   if (OF == Triple::COFF)
172     SectName += InstrProfSectNameCoff[IPSK];
173   else
174     SectName += InstrProfSectNameCommon[IPSK];
175 
176   if (OF == Triple::MachO && IPSK == IPSK_data && AddSegmentInfo)
177     SectName += ",regular,live_support";
178 
179   return SectName;
180 }
181 
182 void SoftInstrProfErrors::addError(instrprof_error IE) {
183   if (IE == instrprof_error::success)
184     return;
185 
186   if (FirstError == instrprof_error::success)
187     FirstError = IE;
188 
189   switch (IE) {
190   case instrprof_error::hash_mismatch:
191     ++NumHashMismatches;
192     break;
193   case instrprof_error::count_mismatch:
194     ++NumCountMismatches;
195     break;
196   case instrprof_error::counter_overflow:
197     ++NumCounterOverflows;
198     break;
199   case instrprof_error::value_site_count_mismatch:
200     ++NumValueSiteCountMismatches;
201     break;
202   default:
203     llvm_unreachable("Not a soft error");
204   }
205 }
206 
207 std::string InstrProfError::message() const {
208   return getInstrProfErrString(Err);
209 }
210 
211 char InstrProfError::ID = 0;
212 
213 std::string getPGOFuncName(StringRef RawFuncName,
214                            GlobalValue::LinkageTypes Linkage,
215                            StringRef FileName,
216                            uint64_t Version LLVM_ATTRIBUTE_UNUSED) {
217   return GlobalValue::getGlobalIdentifier(RawFuncName, Linkage, FileName);
218 }
219 
220 // Strip NumPrefix level of directory name from PathNameStr. If the number of
221 // directory separators is less than NumPrefix, strip all the directories and
222 // leave base file name only.
223 static StringRef stripDirPrefix(StringRef PathNameStr, uint32_t NumPrefix) {
224   uint32_t Count = NumPrefix;
225   uint32_t Pos = 0, LastPos = 0;
226   for (auto & CI : PathNameStr) {
227     ++Pos;
228     if (llvm::sys::path::is_separator(CI)) {
229       LastPos = Pos;
230       --Count;
231     }
232     if (Count == 0)
233       break;
234   }
235   return PathNameStr.substr(LastPos);
236 }
237 
238 // Return the PGOFuncName. This function has some special handling when called
239 // in LTO optimization. The following only applies when calling in LTO passes
240 // (when \c InLTO is true): LTO's internalization privatizes many global linkage
241 // symbols. This happens after value profile annotation, but those internal
242 // linkage functions should not have a source prefix.
243 // Additionally, for ThinLTO mode, exported internal functions are promoted
244 // and renamed. We need to ensure that the original internal PGO name is
245 // used when computing the GUID that is compared against the profiled GUIDs.
246 // To differentiate compiler generated internal symbols from original ones,
247 // PGOFuncName meta data are created and attached to the original internal
248 // symbols in the value profile annotation step
249 // (PGOUseFunc::annotateIndirectCallSites). If a symbol does not have the meta
250 // data, its original linkage must be non-internal.
251 std::string getPGOFuncName(const Function &F, bool InLTO, uint64_t Version) {
252   if (!InLTO) {
253     StringRef FileName = (StaticFuncFullModulePrefix
254                               ? F.getParent()->getName()
255                               : sys::path::filename(F.getParent()->getName()));
256     if (StaticFuncFullModulePrefix && StaticFuncStripDirNamePrefix != 0)
257       FileName = stripDirPrefix(FileName, StaticFuncStripDirNamePrefix);
258     return getPGOFuncName(F.getName(), F.getLinkage(), FileName, Version);
259   }
260 
261   // In LTO mode (when InLTO is true), first check if there is a meta data.
262   if (MDNode *MD = getPGOFuncNameMetadata(F)) {
263     StringRef S = cast<MDString>(MD->getOperand(0))->getString();
264     return S.str();
265   }
266 
267   // If there is no meta data, the function must be a global before the value
268   // profile annotation pass. Its current linkage may be internal if it is
269   // internalized in LTO mode.
270   return getPGOFuncName(F.getName(), GlobalValue::ExternalLinkage, "");
271 }
272 
273 StringRef getFuncNameWithoutPrefix(StringRef PGOFuncName, StringRef FileName) {
274   if (FileName.empty())
275     return PGOFuncName;
276   // Drop the file name including ':'. See also getPGOFuncName.
277   if (PGOFuncName.startswith(FileName))
278     PGOFuncName = PGOFuncName.drop_front(FileName.size() + 1);
279   return PGOFuncName;
280 }
281 
282 // \p FuncName is the string used as profile lookup key for the function. A
283 // symbol is created to hold the name. Return the legalized symbol name.
284 std::string getPGOFuncNameVarName(StringRef FuncName,
285                                   GlobalValue::LinkageTypes Linkage) {
286   std::string VarName = getInstrProfNameVarPrefix();
287   VarName += FuncName;
288 
289   if (!GlobalValue::isLocalLinkage(Linkage))
290     return VarName;
291 
292   // Now fix up illegal chars in local VarName that may upset the assembler.
293   const char *InvalidChars = "-:<>/\"'";
294   size_t found = VarName.find_first_of(InvalidChars);
295   while (found != std::string::npos) {
296     VarName[found] = '_';
297     found = VarName.find_first_of(InvalidChars, found + 1);
298   }
299   return VarName;
300 }
301 
302 GlobalVariable *createPGOFuncNameVar(Module &M,
303                                      GlobalValue::LinkageTypes Linkage,
304                                      StringRef PGOFuncName) {
305   // We generally want to match the function's linkage, but available_externally
306   // and extern_weak both have the wrong semantics, and anything that doesn't
307   // need to link across compilation units doesn't need to be visible at all.
308   if (Linkage == GlobalValue::ExternalWeakLinkage)
309     Linkage = GlobalValue::LinkOnceAnyLinkage;
310   else if (Linkage == GlobalValue::AvailableExternallyLinkage)
311     Linkage = GlobalValue::LinkOnceODRLinkage;
312   else if (Linkage == GlobalValue::InternalLinkage ||
313            Linkage == GlobalValue::ExternalLinkage)
314     Linkage = GlobalValue::PrivateLinkage;
315 
316   auto *Value =
317       ConstantDataArray::getString(M.getContext(), PGOFuncName, false);
318   auto FuncNameVar =
319       new GlobalVariable(M, Value->getType(), true, Linkage, Value,
320                          getPGOFuncNameVarName(PGOFuncName, Linkage));
321 
322   // Hide the symbol so that we correctly get a copy for each executable.
323   if (!GlobalValue::isLocalLinkage(FuncNameVar->getLinkage()))
324     FuncNameVar->setVisibility(GlobalValue::HiddenVisibility);
325 
326   return FuncNameVar;
327 }
328 
329 GlobalVariable *createPGOFuncNameVar(Function &F, StringRef PGOFuncName) {
330   return createPGOFuncNameVar(*F.getParent(), F.getLinkage(), PGOFuncName);
331 }
332 
333 void InstrProfSymtab::create(Module &M, bool InLTO) {
334   for (Function &F : M) {
335     // Function may not have a name: like using asm("") to overwrite the name.
336     // Ignore in this case.
337     if (!F.hasName())
338       continue;
339     const std::string &PGOFuncName = getPGOFuncName(F, InLTO);
340     addFuncName(PGOFuncName);
341     MD5FuncMap.emplace_back(Function::getGUID(PGOFuncName), &F);
342     // In ThinLTO, local function may have been promoted to global and have
343     // suffix added to the function name. We need to add the stripped function
344     // name to the symbol table so that we can find a match from profile.
345     if (InLTO) {
346       auto pos = PGOFuncName.find('.');
347       if (pos != std::string::npos) {
348         const std::string &OtherFuncName = PGOFuncName.substr(0, pos);
349         addFuncName(OtherFuncName);
350         MD5FuncMap.emplace_back(Function::getGUID(OtherFuncName), &F);
351       }
352     }
353   }
354 
355   finalizeSymtab();
356 }
357 
358 Error collectPGOFuncNameStrings(const std::vector<std::string> &NameStrs,
359                                 bool doCompression, std::string &Result) {
360   assert(!NameStrs.empty() && "No name data to emit");
361 
362   uint8_t Header[16], *P = Header;
363   std::string UncompressedNameStrings =
364       join(NameStrs.begin(), NameStrs.end(), getInstrProfNameSeparator());
365 
366   assert(StringRef(UncompressedNameStrings)
367                  .count(getInstrProfNameSeparator()) == (NameStrs.size() - 1) &&
368          "PGO name is invalid (contains separator token)");
369 
370   unsigned EncLen = encodeULEB128(UncompressedNameStrings.length(), P);
371   P += EncLen;
372 
373   auto WriteStringToResult = [&](size_t CompressedLen, StringRef InputStr) {
374     EncLen = encodeULEB128(CompressedLen, P);
375     P += EncLen;
376     char *HeaderStr = reinterpret_cast<char *>(&Header[0]);
377     unsigned HeaderLen = P - &Header[0];
378     Result.append(HeaderStr, HeaderLen);
379     Result += InputStr;
380     return Error::success();
381   };
382 
383   if (!doCompression) {
384     return WriteStringToResult(0, UncompressedNameStrings);
385   }
386 
387   SmallString<128> CompressedNameStrings;
388   Error E = zlib::compress(StringRef(UncompressedNameStrings),
389                            CompressedNameStrings, zlib::BestSizeCompression);
390   if (E) {
391     consumeError(std::move(E));
392     return make_error<InstrProfError>(instrprof_error::compress_failed);
393   }
394 
395   return WriteStringToResult(CompressedNameStrings.size(),
396                              CompressedNameStrings);
397 }
398 
399 StringRef getPGOFuncNameVarInitializer(GlobalVariable *NameVar) {
400   auto *Arr = cast<ConstantDataArray>(NameVar->getInitializer());
401   StringRef NameStr =
402       Arr->isCString() ? Arr->getAsCString() : Arr->getAsString();
403   return NameStr;
404 }
405 
406 Error collectPGOFuncNameStrings(const std::vector<GlobalVariable *> &NameVars,
407                                 std::string &Result, bool doCompression) {
408   std::vector<std::string> NameStrs;
409   for (auto *NameVar : NameVars) {
410     NameStrs.push_back(getPGOFuncNameVarInitializer(NameVar));
411   }
412   return collectPGOFuncNameStrings(
413       NameStrs, zlib::isAvailable() && doCompression, Result);
414 }
415 
416 Error readPGOFuncNameStrings(StringRef NameStrings, InstrProfSymtab &Symtab) {
417   const uint8_t *P = reinterpret_cast<const uint8_t *>(NameStrings.data());
418   const uint8_t *EndP = reinterpret_cast<const uint8_t *>(NameStrings.data() +
419                                                           NameStrings.size());
420   while (P < EndP) {
421     uint32_t N;
422     uint64_t UncompressedSize = decodeULEB128(P, &N);
423     P += N;
424     uint64_t CompressedSize = decodeULEB128(P, &N);
425     P += N;
426     bool isCompressed = (CompressedSize != 0);
427     SmallString<128> UncompressedNameStrings;
428     StringRef NameStrings;
429     if (isCompressed) {
430       StringRef CompressedNameStrings(reinterpret_cast<const char *>(P),
431                                       CompressedSize);
432       if (Error E =
433               zlib::uncompress(CompressedNameStrings, UncompressedNameStrings,
434                                UncompressedSize)) {
435         consumeError(std::move(E));
436         return make_error<InstrProfError>(instrprof_error::uncompress_failed);
437       }
438       P += CompressedSize;
439       NameStrings = StringRef(UncompressedNameStrings.data(),
440                               UncompressedNameStrings.size());
441     } else {
442       NameStrings =
443           StringRef(reinterpret_cast<const char *>(P), UncompressedSize);
444       P += UncompressedSize;
445     }
446     // Now parse the name strings.
447     SmallVector<StringRef, 0> Names;
448     NameStrings.split(Names, getInstrProfNameSeparator());
449     for (StringRef &Name : Names)
450       Symtab.addFuncName(Name);
451 
452     while (P < EndP && *P == 0)
453       P++;
454   }
455   Symtab.finalizeSymtab();
456   return Error::success();
457 }
458 
459 void InstrProfValueSiteRecord::merge(SoftInstrProfErrors &SIPE,
460                                      InstrProfValueSiteRecord &Input,
461                                      uint64_t Weight) {
462   this->sortByTargetValues();
463   Input.sortByTargetValues();
464   auto I = ValueData.begin();
465   auto IE = ValueData.end();
466   for (auto J = Input.ValueData.begin(), JE = Input.ValueData.end(); J != JE;
467        ++J) {
468     while (I != IE && I->Value < J->Value)
469       ++I;
470     if (I != IE && I->Value == J->Value) {
471       bool Overflowed;
472       I->Count = SaturatingMultiplyAdd(J->Count, Weight, I->Count, &Overflowed);
473       if (Overflowed)
474         SIPE.addError(instrprof_error::counter_overflow);
475       ++I;
476       continue;
477     }
478     ValueData.insert(I, *J);
479   }
480 }
481 
482 void InstrProfValueSiteRecord::scale(SoftInstrProfErrors &SIPE,
483                                      uint64_t Weight) {
484   for (auto I = ValueData.begin(), IE = ValueData.end(); I != IE; ++I) {
485     bool Overflowed;
486     I->Count = SaturatingMultiply(I->Count, Weight, &Overflowed);
487     if (Overflowed)
488       SIPE.addError(instrprof_error::counter_overflow);
489   }
490 }
491 
492 // Merge Value Profile data from Src record to this record for ValueKind.
493 // Scale merged value counts by \p Weight.
494 void InstrProfRecord::mergeValueProfData(uint32_t ValueKind,
495                                          InstrProfRecord &Src,
496                                          uint64_t Weight) {
497   uint32_t ThisNumValueSites = getNumValueSites(ValueKind);
498   uint32_t OtherNumValueSites = Src.getNumValueSites(ValueKind);
499   if (ThisNumValueSites != OtherNumValueSites) {
500     SIPE.addError(instrprof_error::value_site_count_mismatch);
501     return;
502   }
503   std::vector<InstrProfValueSiteRecord> &ThisSiteRecords =
504       getValueSitesForKind(ValueKind);
505   std::vector<InstrProfValueSiteRecord> &OtherSiteRecords =
506       Src.getValueSitesForKind(ValueKind);
507   for (uint32_t I = 0; I < ThisNumValueSites; I++)
508     ThisSiteRecords[I].merge(SIPE, OtherSiteRecords[I], Weight);
509 }
510 
511 void InstrProfRecord::merge(InstrProfRecord &Other, uint64_t Weight) {
512   // If the number of counters doesn't match we either have bad data
513   // or a hash collision.
514   if (Counts.size() != Other.Counts.size()) {
515     SIPE.addError(instrprof_error::count_mismatch);
516     return;
517   }
518 
519   for (size_t I = 0, E = Other.Counts.size(); I < E; ++I) {
520     bool Overflowed;
521     Counts[I] =
522         SaturatingMultiplyAdd(Other.Counts[I], Weight, Counts[I], &Overflowed);
523     if (Overflowed)
524       SIPE.addError(instrprof_error::counter_overflow);
525   }
526 
527   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
528     mergeValueProfData(Kind, Other, Weight);
529 }
530 
531 void InstrProfRecord::scaleValueProfData(uint32_t ValueKind, uint64_t Weight) {
532   uint32_t ThisNumValueSites = getNumValueSites(ValueKind);
533   std::vector<InstrProfValueSiteRecord> &ThisSiteRecords =
534       getValueSitesForKind(ValueKind);
535   for (uint32_t I = 0; I < ThisNumValueSites; I++)
536     ThisSiteRecords[I].scale(SIPE, Weight);
537 }
538 
539 void InstrProfRecord::scale(uint64_t Weight) {
540   for (auto &Count : this->Counts) {
541     bool Overflowed;
542     Count = SaturatingMultiply(Count, Weight, &Overflowed);
543     if (Overflowed)
544       SIPE.addError(instrprof_error::counter_overflow);
545   }
546   for (uint32_t Kind = IPVK_First; Kind <= IPVK_Last; ++Kind)
547     scaleValueProfData(Kind, Weight);
548 }
549 
550 // Map indirect call target name hash to name string.
551 uint64_t InstrProfRecord::remapValue(uint64_t Value, uint32_t ValueKind,
552                                      ValueMapType *ValueMap) {
553   if (!ValueMap)
554     return Value;
555   switch (ValueKind) {
556   case IPVK_IndirectCallTarget: {
557     auto Result =
558         std::lower_bound(ValueMap->begin(), ValueMap->end(), Value,
559                          [](const std::pair<uint64_t, uint64_t> &LHS,
560                             uint64_t RHS) { return LHS.first < RHS; });
561    // Raw function pointer collected by value profiler may be from
562    // external functions that are not instrumented. They won't have
563    // mapping data to be used by the deserializer. Force the value to
564    // be 0 in this case.
565     if (Result != ValueMap->end() && Result->first == Value)
566       Value = (uint64_t)Result->second;
567     else
568       Value = 0;
569     break;
570   }
571   }
572   return Value;
573 }
574 
575 void InstrProfRecord::addValueData(uint32_t ValueKind, uint32_t Site,
576                                    InstrProfValueData *VData, uint32_t N,
577                                    ValueMapType *ValueMap) {
578   for (uint32_t I = 0; I < N; I++) {
579     VData[I].Value = remapValue(VData[I].Value, ValueKind, ValueMap);
580   }
581   std::vector<InstrProfValueSiteRecord> &ValueSites =
582       getValueSitesForKind(ValueKind);
583   if (N == 0)
584     ValueSites.emplace_back();
585   else
586     ValueSites.emplace_back(VData, VData + N);
587 }
588 
589 #define INSTR_PROF_COMMON_API_IMPL
590 #include "llvm/ProfileData/InstrProfData.inc"
591 
592 /*!
593  * \brief ValueProfRecordClosure Interface implementation for  InstrProfRecord
594  *  class. These C wrappers are used as adaptors so that C++ code can be
595  *  invoked as callbacks.
596  */
597 uint32_t getNumValueKindsInstrProf(const void *Record) {
598   return reinterpret_cast<const InstrProfRecord *>(Record)->getNumValueKinds();
599 }
600 
601 uint32_t getNumValueSitesInstrProf(const void *Record, uint32_t VKind) {
602   return reinterpret_cast<const InstrProfRecord *>(Record)
603       ->getNumValueSites(VKind);
604 }
605 
606 uint32_t getNumValueDataInstrProf(const void *Record, uint32_t VKind) {
607   return reinterpret_cast<const InstrProfRecord *>(Record)
608       ->getNumValueData(VKind);
609 }
610 
611 uint32_t getNumValueDataForSiteInstrProf(const void *R, uint32_t VK,
612                                          uint32_t S) {
613   return reinterpret_cast<const InstrProfRecord *>(R)
614       ->getNumValueDataForSite(VK, S);
615 }
616 
617 void getValueForSiteInstrProf(const void *R, InstrProfValueData *Dst,
618                               uint32_t K, uint32_t S) {
619   reinterpret_cast<const InstrProfRecord *>(R)->getValueForSite(Dst, K, S);
620 }
621 
622 ValueProfData *allocValueProfDataInstrProf(size_t TotalSizeInBytes) {
623   ValueProfData *VD =
624       (ValueProfData *)(new (::operator new(TotalSizeInBytes)) ValueProfData());
625   memset(VD, 0, TotalSizeInBytes);
626   return VD;
627 }
628 
629 static ValueProfRecordClosure InstrProfRecordClosure = {
630     nullptr,
631     getNumValueKindsInstrProf,
632     getNumValueSitesInstrProf,
633     getNumValueDataInstrProf,
634     getNumValueDataForSiteInstrProf,
635     nullptr,
636     getValueForSiteInstrProf,
637     allocValueProfDataInstrProf};
638 
639 // Wrapper implementation using the closure mechanism.
640 uint32_t ValueProfData::getSize(const InstrProfRecord &Record) {
641   InstrProfRecordClosure.Record = &Record;
642   return getValueProfDataSize(&InstrProfRecordClosure);
643 }
644 
645 // Wrapper implementation using the closure mechanism.
646 std::unique_ptr<ValueProfData>
647 ValueProfData::serializeFrom(const InstrProfRecord &Record) {
648   InstrProfRecordClosure.Record = &Record;
649 
650   std::unique_ptr<ValueProfData> VPD(
651       serializeValueProfDataFrom(&InstrProfRecordClosure, nullptr));
652   return VPD;
653 }
654 
655 void ValueProfRecord::deserializeTo(InstrProfRecord &Record,
656                                     InstrProfRecord::ValueMapType *VMap) {
657   Record.reserveSites(Kind, NumValueSites);
658 
659   InstrProfValueData *ValueData = getValueProfRecordValueData(this);
660   for (uint64_t VSite = 0; VSite < NumValueSites; ++VSite) {
661     uint8_t ValueDataCount = this->SiteCountArray[VSite];
662     Record.addValueData(Kind, VSite, ValueData, ValueDataCount, VMap);
663     ValueData += ValueDataCount;
664   }
665 }
666 
667 // For writing/serializing,  Old is the host endianness, and  New is
668 // byte order intended on disk. For Reading/deserialization, Old
669 // is the on-disk source endianness, and New is the host endianness.
670 void ValueProfRecord::swapBytes(support::endianness Old,
671                                 support::endianness New) {
672   using namespace support;
673 
674   if (Old == New)
675     return;
676 
677   if (getHostEndianness() != Old) {
678     sys::swapByteOrder<uint32_t>(NumValueSites);
679     sys::swapByteOrder<uint32_t>(Kind);
680   }
681   uint32_t ND = getValueProfRecordNumValueData(this);
682   InstrProfValueData *VD = getValueProfRecordValueData(this);
683 
684   // No need to swap byte array: SiteCountArrray.
685   for (uint32_t I = 0; I < ND; I++) {
686     sys::swapByteOrder<uint64_t>(VD[I].Value);
687     sys::swapByteOrder<uint64_t>(VD[I].Count);
688   }
689   if (getHostEndianness() == Old) {
690     sys::swapByteOrder<uint32_t>(NumValueSites);
691     sys::swapByteOrder<uint32_t>(Kind);
692   }
693 }
694 
695 void ValueProfData::deserializeTo(InstrProfRecord &Record,
696                                   InstrProfRecord::ValueMapType *VMap) {
697   if (NumValueKinds == 0)
698     return;
699 
700   ValueProfRecord *VR = getFirstValueProfRecord(this);
701   for (uint32_t K = 0; K < NumValueKinds; K++) {
702     VR->deserializeTo(Record, VMap);
703     VR = getValueProfRecordNext(VR);
704   }
705 }
706 
707 template <class T>
708 static T swapToHostOrder(const unsigned char *&D, support::endianness Orig) {
709   using namespace support;
710 
711   if (Orig == little)
712     return endian::readNext<T, little, unaligned>(D);
713   else
714     return endian::readNext<T, big, unaligned>(D);
715 }
716 
717 static std::unique_ptr<ValueProfData> allocValueProfData(uint32_t TotalSize) {
718   return std::unique_ptr<ValueProfData>(new (::operator new(TotalSize))
719                                             ValueProfData());
720 }
721 
722 Error ValueProfData::checkIntegrity() {
723   if (NumValueKinds > IPVK_Last + 1)
724     return make_error<InstrProfError>(instrprof_error::malformed);
725   // Total size needs to be mulltiple of quadword size.
726   if (TotalSize % sizeof(uint64_t))
727     return make_error<InstrProfError>(instrprof_error::malformed);
728 
729   ValueProfRecord *VR = getFirstValueProfRecord(this);
730   for (uint32_t K = 0; K < this->NumValueKinds; K++) {
731     if (VR->Kind > IPVK_Last)
732       return make_error<InstrProfError>(instrprof_error::malformed);
733     VR = getValueProfRecordNext(VR);
734     if ((char *)VR - (char *)this > (ptrdiff_t)TotalSize)
735       return make_error<InstrProfError>(instrprof_error::malformed);
736   }
737   return Error::success();
738 }
739 
740 Expected<std::unique_ptr<ValueProfData>>
741 ValueProfData::getValueProfData(const unsigned char *D,
742                                 const unsigned char *const BufferEnd,
743                                 support::endianness Endianness) {
744   using namespace support;
745 
746   if (D + sizeof(ValueProfData) > BufferEnd)
747     return make_error<InstrProfError>(instrprof_error::truncated);
748 
749   const unsigned char *Header = D;
750   uint32_t TotalSize = swapToHostOrder<uint32_t>(Header, Endianness);
751   if (D + TotalSize > BufferEnd)
752     return make_error<InstrProfError>(instrprof_error::too_large);
753 
754   std::unique_ptr<ValueProfData> VPD = allocValueProfData(TotalSize);
755   memcpy(VPD.get(), D, TotalSize);
756   // Byte swap.
757   VPD->swapBytesToHost(Endianness);
758 
759   Error E = VPD->checkIntegrity();
760   if (E)
761     return std::move(E);
762 
763   return std::move(VPD);
764 }
765 
766 void ValueProfData::swapBytesToHost(support::endianness Endianness) {
767   using namespace support;
768 
769   if (Endianness == getHostEndianness())
770     return;
771 
772   sys::swapByteOrder<uint32_t>(TotalSize);
773   sys::swapByteOrder<uint32_t>(NumValueKinds);
774 
775   ValueProfRecord *VR = getFirstValueProfRecord(this);
776   for (uint32_t K = 0; K < NumValueKinds; K++) {
777     VR->swapBytes(Endianness, getHostEndianness());
778     VR = getValueProfRecordNext(VR);
779   }
780 }
781 
782 void ValueProfData::swapBytesFromHost(support::endianness Endianness) {
783   using namespace support;
784 
785   if (Endianness == getHostEndianness())
786     return;
787 
788   ValueProfRecord *VR = getFirstValueProfRecord(this);
789   for (uint32_t K = 0; K < NumValueKinds; K++) {
790     ValueProfRecord *NVR = getValueProfRecordNext(VR);
791     VR->swapBytes(getHostEndianness(), Endianness);
792     VR = NVR;
793   }
794   sys::swapByteOrder<uint32_t>(TotalSize);
795   sys::swapByteOrder<uint32_t>(NumValueKinds);
796 }
797 
798 void annotateValueSite(Module &M, Instruction &Inst,
799                        const InstrProfRecord &InstrProfR,
800                        InstrProfValueKind ValueKind, uint32_t SiteIdx,
801                        uint32_t MaxMDCount) {
802   uint32_t NV = InstrProfR.getNumValueDataForSite(ValueKind, SiteIdx);
803   if (!NV)
804     return;
805 
806   uint64_t Sum = 0;
807   std::unique_ptr<InstrProfValueData[]> VD =
808       InstrProfR.getValueForSite(ValueKind, SiteIdx, &Sum);
809 
810   ArrayRef<InstrProfValueData> VDs(VD.get(), NV);
811   annotateValueSite(M, Inst, VDs, Sum, ValueKind, MaxMDCount);
812 }
813 
814 void annotateValueSite(Module &M, Instruction &Inst,
815                        ArrayRef<InstrProfValueData> VDs,
816                        uint64_t Sum, InstrProfValueKind ValueKind,
817                        uint32_t MaxMDCount) {
818   LLVMContext &Ctx = M.getContext();
819   MDBuilder MDHelper(Ctx);
820   SmallVector<Metadata *, 3> Vals;
821   // Tag
822   Vals.push_back(MDHelper.createString("VP"));
823   // Value Kind
824   Vals.push_back(MDHelper.createConstant(
825       ConstantInt::get(Type::getInt32Ty(Ctx), ValueKind)));
826   // Total Count
827   Vals.push_back(
828       MDHelper.createConstant(ConstantInt::get(Type::getInt64Ty(Ctx), Sum)));
829 
830   // Value Profile Data
831   uint32_t MDCount = MaxMDCount;
832   for (auto &VD : VDs) {
833     Vals.push_back(MDHelper.createConstant(
834         ConstantInt::get(Type::getInt64Ty(Ctx), VD.Value)));
835     Vals.push_back(MDHelper.createConstant(
836         ConstantInt::get(Type::getInt64Ty(Ctx), VD.Count)));
837     if (--MDCount == 0)
838       break;
839   }
840   Inst.setMetadata(LLVMContext::MD_prof, MDNode::get(Ctx, Vals));
841 }
842 
843 bool getValueProfDataFromInst(const Instruction &Inst,
844                               InstrProfValueKind ValueKind,
845                               uint32_t MaxNumValueData,
846                               InstrProfValueData ValueData[],
847                               uint32_t &ActualNumValueData, uint64_t &TotalC) {
848   MDNode *MD = Inst.getMetadata(LLVMContext::MD_prof);
849   if (!MD)
850     return false;
851 
852   unsigned NOps = MD->getNumOperands();
853 
854   if (NOps < 5)
855     return false;
856 
857   // Operand 0 is a string tag "VP":
858   MDString *Tag = cast<MDString>(MD->getOperand(0));
859   if (!Tag)
860     return false;
861 
862   if (!Tag->getString().equals("VP"))
863     return false;
864 
865   // Now check kind:
866   ConstantInt *KindInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(1));
867   if (!KindInt)
868     return false;
869   if (KindInt->getZExtValue() != ValueKind)
870     return false;
871 
872   // Get total count
873   ConstantInt *TotalCInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(2));
874   if (!TotalCInt)
875     return false;
876   TotalC = TotalCInt->getZExtValue();
877 
878   ActualNumValueData = 0;
879 
880   for (unsigned I = 3; I < NOps; I += 2) {
881     if (ActualNumValueData >= MaxNumValueData)
882       break;
883     ConstantInt *Value = mdconst::dyn_extract<ConstantInt>(MD->getOperand(I));
884     ConstantInt *Count =
885         mdconst::dyn_extract<ConstantInt>(MD->getOperand(I + 1));
886     if (!Value || !Count)
887       return false;
888     ValueData[ActualNumValueData].Value = Value->getZExtValue();
889     ValueData[ActualNumValueData].Count = Count->getZExtValue();
890     ActualNumValueData++;
891   }
892   return true;
893 }
894 
895 MDNode *getPGOFuncNameMetadata(const Function &F) {
896   return F.getMetadata(getPGOFuncNameMetadataName());
897 }
898 
899 void createPGOFuncNameMetadata(Function &F, StringRef PGOFuncName) {
900   // Only for internal linkage functions.
901   if (PGOFuncName == F.getName())
902       return;
903   // Don't create duplicated meta-data.
904   if (getPGOFuncNameMetadata(F))
905     return;
906   LLVMContext &C = F.getContext();
907   MDNode *N = MDNode::get(C, MDString::get(C, PGOFuncName));
908   F.setMetadata(getPGOFuncNameMetadataName(), N);
909 }
910 
911 bool needsComdatForCounter(const Function &F, const Module &M) {
912   if (F.hasComdat())
913     return true;
914 
915   Triple TT(M.getTargetTriple());
916   if (!TT.isOSBinFormatELF() && !TT.isOSBinFormatWasm())
917     return false;
918 
919   // See createPGOFuncNameVar for more details. To avoid link errors, profile
920   // counters for function with available_externally linkage needs to be changed
921   // to linkonce linkage. On ELF based systems, this leads to weak symbols to be
922   // created. Without using comdat, duplicate entries won't be removed by the
923   // linker leading to increased data segement size and raw profile size. Even
924   // worse, since the referenced counter from profile per-function data object
925   // will be resolved to the common strong definition, the profile counts for
926   // available_externally functions will end up being duplicated in raw profile
927   // data. This can result in distorted profile as the counts of those dups
928   // will be accumulated by the profile merger.
929   GlobalValue::LinkageTypes Linkage = F.getLinkage();
930   if (Linkage != GlobalValue::ExternalWeakLinkage &&
931       Linkage != GlobalValue::AvailableExternallyLinkage)
932     return false;
933 
934   return true;
935 }
936 
937 // Check if INSTR_PROF_RAW_VERSION_VAR is defined.
938 bool isIRPGOFlagSet(const Module *M) {
939   auto IRInstrVar =
940       M->getNamedGlobal(INSTR_PROF_QUOTE(INSTR_PROF_RAW_VERSION_VAR));
941   if (!IRInstrVar || IRInstrVar->isDeclaration() ||
942       IRInstrVar->hasLocalLinkage())
943     return false;
944 
945   // Check if the flag is set.
946   if (!IRInstrVar->hasInitializer())
947     return false;
948 
949   const Constant *InitVal = IRInstrVar->getInitializer();
950   if (!InitVal)
951     return false;
952 
953   return (dyn_cast<ConstantInt>(InitVal)->getZExtValue() &
954           VARIANT_MASK_IR_PROF) != 0;
955 }
956 
957 // Check if we can safely rename this Comdat function.
958 bool canRenameComdatFunc(const Function &F, bool CheckAddressTaken) {
959   if (F.getName().empty())
960     return false;
961   if (!needsComdatForCounter(F, *(F.getParent())))
962     return false;
963   // Unsafe to rename the address-taken function (which can be used in
964   // function comparison).
965   if (CheckAddressTaken && F.hasAddressTaken())
966     return false;
967   // Only safe to do if this function may be discarded if it is not used
968   // in the compilation unit.
969   if (!GlobalValue::isDiscardableIfUnused(F.getLinkage()))
970     return false;
971 
972   // For AvailableExternallyLinkage functions.
973   if (!F.hasComdat()) {
974     assert(F.getLinkage() == GlobalValue::AvailableExternallyLinkage);
975     return true;
976   }
977   return true;
978 }
979 
980 // Parse the value profile options.
981 void getMemOPSizeRangeFromOption(std::string MemOPSizeRange,
982                                  int64_t &RangeStart, int64_t &RangeLast) {
983   static const int64_t DefaultMemOPSizeRangeStart = 0;
984   static const int64_t DefaultMemOPSizeRangeLast = 8;
985   RangeStart = DefaultMemOPSizeRangeStart;
986   RangeLast = DefaultMemOPSizeRangeLast;
987 
988   if (!MemOPSizeRange.empty()) {
989     auto Pos = MemOPSizeRange.find(":");
990     if (Pos != std::string::npos) {
991       if (Pos > 0)
992         RangeStart = atoi(MemOPSizeRange.substr(0, Pos).c_str());
993       if (Pos < MemOPSizeRange.size() - 1)
994         RangeLast = atoi(MemOPSizeRange.substr(Pos + 1).c_str());
995     } else
996       RangeLast = atoi(MemOPSizeRange.c_str());
997   }
998   assert(RangeLast >= RangeStart);
999 }
1000 
1001 } // end namespace llvm
1002