1 //===- SampleProfReader.cpp - Read LLVM sample profile data ---------------===//
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 implements the class that reads LLVM sample profiles. It
10 // supports three file formats: text, binary and gcov.
11 //
12 // The textual representation is useful for debugging and testing purposes. The
13 // binary representation is more compact, resulting in smaller file sizes.
14 //
15 // The gcov encoding is the one generated by GCC's AutoFDO profile creation
16 // tool (https://github.com/google/autofdo)
17 //
18 // All three encodings can be used interchangeably as an input sample profile.
19 //
20 //===----------------------------------------------------------------------===//
21 
22 #include "llvm/ProfileData/SampleProfReader.h"
23 #include "llvm/ADT/DenseMap.h"
24 #include "llvm/ADT/STLExtras.h"
25 #include "llvm/ADT/StringRef.h"
26 #include "llvm/IR/ProfileSummary.h"
27 #include "llvm/ProfileData/ProfileCommon.h"
28 #include "llvm/ProfileData/SampleProf.h"
29 #include "llvm/Support/Compression.h"
30 #include "llvm/Support/ErrorOr.h"
31 #include "llvm/Support/LEB128.h"
32 #include "llvm/Support/LineIterator.h"
33 #include "llvm/Support/MD5.h"
34 #include "llvm/Support/MemoryBuffer.h"
35 #include "llvm/Support/raw_ostream.h"
36 #include <algorithm>
37 #include <cstddef>
38 #include <cstdint>
39 #include <limits>
40 #include <memory>
41 #include <system_error>
42 #include <vector>
43 
44 using namespace llvm;
45 using namespace sampleprof;
46 
47 /// Dump the function profile for \p FName.
48 ///
49 /// \param FName Name of the function to print.
50 /// \param OS Stream to emit the output to.
51 void SampleProfileReader::dumpFunctionProfile(StringRef FName,
52                                               raw_ostream &OS) {
53   OS << "Function: " << FName << ": " << Profiles[FName];
54 }
55 
56 /// Dump all the function profiles found on stream \p OS.
57 void SampleProfileReader::dump(raw_ostream &OS) {
58   for (const auto &I : Profiles)
59     dumpFunctionProfile(I.getKey(), OS);
60 }
61 
62 /// Parse \p Input as function head.
63 ///
64 /// Parse one line of \p Input, and update function name in \p FName,
65 /// function's total sample count in \p NumSamples, function's entry
66 /// count in \p NumHeadSamples.
67 ///
68 /// \returns true if parsing is successful.
69 static bool ParseHead(const StringRef &Input, StringRef &FName,
70                       uint64_t &NumSamples, uint64_t &NumHeadSamples) {
71   if (Input[0] == ' ')
72     return false;
73   size_t n2 = Input.rfind(':');
74   size_t n1 = Input.rfind(':', n2 - 1);
75   FName = Input.substr(0, n1);
76   if (Input.substr(n1 + 1, n2 - n1 - 1).getAsInteger(10, NumSamples))
77     return false;
78   if (Input.substr(n2 + 1).getAsInteger(10, NumHeadSamples))
79     return false;
80   return true;
81 }
82 
83 /// Returns true if line offset \p L is legal (only has 16 bits).
84 static bool isOffsetLegal(unsigned L) { return (L & 0xffff) == L; }
85 
86 /// Parse \p Input as line sample.
87 ///
88 /// \param Input input line.
89 /// \param IsCallsite true if the line represents an inlined callsite.
90 /// \param Depth the depth of the inline stack.
91 /// \param NumSamples total samples of the line/inlined callsite.
92 /// \param LineOffset line offset to the start of the function.
93 /// \param Discriminator discriminator of the line.
94 /// \param TargetCountMap map from indirect call target to count.
95 ///
96 /// returns true if parsing is successful.
97 static bool ParseLine(const StringRef &Input, bool &IsCallsite, uint32_t &Depth,
98                       uint64_t &NumSamples, uint32_t &LineOffset,
99                       uint32_t &Discriminator, StringRef &CalleeName,
100                       DenseMap<StringRef, uint64_t> &TargetCountMap) {
101   for (Depth = 0; Input[Depth] == ' '; Depth++)
102     ;
103   if (Depth == 0)
104     return false;
105 
106   size_t n1 = Input.find(':');
107   StringRef Loc = Input.substr(Depth, n1 - Depth);
108   size_t n2 = Loc.find('.');
109   if (n2 == StringRef::npos) {
110     if (Loc.getAsInteger(10, LineOffset) || !isOffsetLegal(LineOffset))
111       return false;
112     Discriminator = 0;
113   } else {
114     if (Loc.substr(0, n2).getAsInteger(10, LineOffset))
115       return false;
116     if (Loc.substr(n2 + 1).getAsInteger(10, Discriminator))
117       return false;
118   }
119 
120   StringRef Rest = Input.substr(n1 + 2);
121   if (Rest[0] >= '0' && Rest[0] <= '9') {
122     IsCallsite = false;
123     size_t n3 = Rest.find(' ');
124     if (n3 == StringRef::npos) {
125       if (Rest.getAsInteger(10, NumSamples))
126         return false;
127     } else {
128       if (Rest.substr(0, n3).getAsInteger(10, NumSamples))
129         return false;
130     }
131     // Find call targets and their sample counts.
132     // Note: In some cases, there are symbols in the profile which are not
133     // mangled. To accommodate such cases, use colon + integer pairs as the
134     // anchor points.
135     // An example:
136     // _M_construct<char *>:1000 string_view<std::allocator<char> >:437
137     // ":1000" and ":437" are used as anchor points so the string above will
138     // be interpreted as
139     // target: _M_construct<char *>
140     // count: 1000
141     // target: string_view<std::allocator<char> >
142     // count: 437
143     while (n3 != StringRef::npos) {
144       n3 += Rest.substr(n3).find_first_not_of(' ');
145       Rest = Rest.substr(n3);
146       n3 = Rest.find_first_of(':');
147       if (n3 == StringRef::npos || n3 == 0)
148         return false;
149 
150       StringRef Target;
151       uint64_t count, n4;
152       while (true) {
153         // Get the segment after the current colon.
154         StringRef AfterColon = Rest.substr(n3 + 1);
155         // Get the target symbol before the current colon.
156         Target = Rest.substr(0, n3);
157         // Check if the word after the current colon is an integer.
158         n4 = AfterColon.find_first_of(' ');
159         n4 = (n4 != StringRef::npos) ? n3 + n4 + 1 : Rest.size();
160         StringRef WordAfterColon = Rest.substr(n3 + 1, n4 - n3 - 1);
161         if (!WordAfterColon.getAsInteger(10, count))
162           break;
163 
164         // Try to find the next colon.
165         uint64_t n5 = AfterColon.find_first_of(':');
166         if (n5 == StringRef::npos)
167           return false;
168         n3 += n5 + 1;
169       }
170 
171       // An anchor point is found. Save the {target, count} pair
172       TargetCountMap[Target] = count;
173       if (n4 == Rest.size())
174         break;
175       // Change n3 to the next blank space after colon + integer pair.
176       n3 = n4;
177     }
178   } else {
179     IsCallsite = true;
180     size_t n3 = Rest.find_last_of(':');
181     CalleeName = Rest.substr(0, n3);
182     if (Rest.substr(n3 + 1).getAsInteger(10, NumSamples))
183       return false;
184   }
185   return true;
186 }
187 
188 /// Load samples from a text file.
189 ///
190 /// See the documentation at the top of the file for an explanation of
191 /// the expected format.
192 ///
193 /// \returns true if the file was loaded successfully, false otherwise.
194 std::error_code SampleProfileReaderText::readImpl() {
195   line_iterator LineIt(*Buffer, /*SkipBlanks=*/true, '#');
196   sampleprof_error Result = sampleprof_error::success;
197 
198   InlineCallStack InlineStack;
199 
200   for (; !LineIt.is_at_eof(); ++LineIt) {
201     if ((*LineIt)[(*LineIt).find_first_not_of(' ')] == '#')
202       continue;
203     // Read the header of each function.
204     //
205     // Note that for function identifiers we are actually expecting
206     // mangled names, but we may not always get them. This happens when
207     // the compiler decides not to emit the function (e.g., it was inlined
208     // and removed). In this case, the binary will not have the linkage
209     // name for the function, so the profiler will emit the function's
210     // unmangled name, which may contain characters like ':' and '>' in its
211     // name (member functions, templates, etc).
212     //
213     // The only requirement we place on the identifier, then, is that it
214     // should not begin with a number.
215     if ((*LineIt)[0] != ' ') {
216       uint64_t NumSamples, NumHeadSamples;
217       StringRef FName;
218       if (!ParseHead(*LineIt, FName, NumSamples, NumHeadSamples)) {
219         reportError(LineIt.line_number(),
220                     "Expected 'mangled_name:NUM:NUM', found " + *LineIt);
221         return sampleprof_error::malformed;
222       }
223       Profiles[FName] = FunctionSamples();
224       FunctionSamples &FProfile = Profiles[FName];
225       FProfile.setName(FName);
226       MergeResult(Result, FProfile.addTotalSamples(NumSamples));
227       MergeResult(Result, FProfile.addHeadSamples(NumHeadSamples));
228       InlineStack.clear();
229       InlineStack.push_back(&FProfile);
230     } else {
231       uint64_t NumSamples;
232       StringRef FName;
233       DenseMap<StringRef, uint64_t> TargetCountMap;
234       bool IsCallsite;
235       uint32_t Depth, LineOffset, Discriminator;
236       if (!ParseLine(*LineIt, IsCallsite, Depth, NumSamples, LineOffset,
237                      Discriminator, FName, TargetCountMap)) {
238         reportError(LineIt.line_number(),
239                     "Expected 'NUM[.NUM]: NUM[ mangled_name:NUM]*', found " +
240                         *LineIt);
241         return sampleprof_error::malformed;
242       }
243       if (IsCallsite) {
244         while (InlineStack.size() > Depth) {
245           InlineStack.pop_back();
246         }
247         FunctionSamples &FSamples = InlineStack.back()->functionSamplesAt(
248             LineLocation(LineOffset, Discriminator))[std::string(FName)];
249         FSamples.setName(FName);
250         MergeResult(Result, FSamples.addTotalSamples(NumSamples));
251         InlineStack.push_back(&FSamples);
252       } else {
253         while (InlineStack.size() > Depth) {
254           InlineStack.pop_back();
255         }
256         FunctionSamples &FProfile = *InlineStack.back();
257         for (const auto &name_count : TargetCountMap) {
258           MergeResult(Result, FProfile.addCalledTargetSamples(
259                                   LineOffset, Discriminator, name_count.first,
260                                   name_count.second));
261         }
262         MergeResult(Result, FProfile.addBodySamples(LineOffset, Discriminator,
263                                                     NumSamples));
264       }
265     }
266   }
267   if (Result == sampleprof_error::success)
268     computeSummary();
269 
270   return Result;
271 }
272 
273 bool SampleProfileReaderText::hasFormat(const MemoryBuffer &Buffer) {
274   bool result = false;
275 
276   // Check that the first non-comment line is a valid function header.
277   line_iterator LineIt(Buffer, /*SkipBlanks=*/true, '#');
278   if (!LineIt.is_at_eof()) {
279     if ((*LineIt)[0] != ' ') {
280       uint64_t NumSamples, NumHeadSamples;
281       StringRef FName;
282       result = ParseHead(*LineIt, FName, NumSamples, NumHeadSamples);
283     }
284   }
285 
286   return result;
287 }
288 
289 template <typename T> ErrorOr<T> SampleProfileReaderBinary::readNumber() {
290   unsigned NumBytesRead = 0;
291   std::error_code EC;
292   uint64_t Val = decodeULEB128(Data, &NumBytesRead);
293 
294   if (Val > std::numeric_limits<T>::max())
295     EC = sampleprof_error::malformed;
296   else if (Data + NumBytesRead > End)
297     EC = sampleprof_error::truncated;
298   else
299     EC = sampleprof_error::success;
300 
301   if (EC) {
302     reportError(0, EC.message());
303     return EC;
304   }
305 
306   Data += NumBytesRead;
307   return static_cast<T>(Val);
308 }
309 
310 ErrorOr<StringRef> SampleProfileReaderBinary::readString() {
311   std::error_code EC;
312   StringRef Str(reinterpret_cast<const char *>(Data));
313   if (Data + Str.size() + 1 > End) {
314     EC = sampleprof_error::truncated;
315     reportError(0, EC.message());
316     return EC;
317   }
318 
319   Data += Str.size() + 1;
320   return Str;
321 }
322 
323 template <typename T>
324 ErrorOr<T> SampleProfileReaderBinary::readUnencodedNumber() {
325   std::error_code EC;
326 
327   if (Data + sizeof(T) > End) {
328     EC = sampleprof_error::truncated;
329     reportError(0, EC.message());
330     return EC;
331   }
332 
333   using namespace support;
334   T Val = endian::readNext<T, little, unaligned>(Data);
335   return Val;
336 }
337 
338 template <typename T>
339 inline ErrorOr<uint32_t> SampleProfileReaderBinary::readStringIndex(T &Table) {
340   std::error_code EC;
341   auto Idx = readNumber<uint32_t>();
342   if (std::error_code EC = Idx.getError())
343     return EC;
344   if (*Idx >= Table.size())
345     return sampleprof_error::truncated_name_table;
346   return *Idx;
347 }
348 
349 ErrorOr<StringRef> SampleProfileReaderBinary::readStringFromTable() {
350   auto Idx = readStringIndex(NameTable);
351   if (std::error_code EC = Idx.getError())
352     return EC;
353 
354   return NameTable[*Idx];
355 }
356 
357 ErrorOr<StringRef> SampleProfileReaderCompactBinary::readStringFromTable() {
358   auto Idx = readStringIndex(NameTable);
359   if (std::error_code EC = Idx.getError())
360     return EC;
361 
362   return StringRef(NameTable[*Idx]);
363 }
364 
365 std::error_code
366 SampleProfileReaderBinary::readProfile(FunctionSamples &FProfile) {
367   auto NumSamples = readNumber<uint64_t>();
368   if (std::error_code EC = NumSamples.getError())
369     return EC;
370   FProfile.addTotalSamples(*NumSamples);
371 
372   // Read the samples in the body.
373   auto NumRecords = readNumber<uint32_t>();
374   if (std::error_code EC = NumRecords.getError())
375     return EC;
376 
377   for (uint32_t I = 0; I < *NumRecords; ++I) {
378     auto LineOffset = readNumber<uint64_t>();
379     if (std::error_code EC = LineOffset.getError())
380       return EC;
381 
382     if (!isOffsetLegal(*LineOffset)) {
383       return std::error_code();
384     }
385 
386     auto Discriminator = readNumber<uint64_t>();
387     if (std::error_code EC = Discriminator.getError())
388       return EC;
389 
390     auto NumSamples = readNumber<uint64_t>();
391     if (std::error_code EC = NumSamples.getError())
392       return EC;
393 
394     auto NumCalls = readNumber<uint32_t>();
395     if (std::error_code EC = NumCalls.getError())
396       return EC;
397 
398     for (uint32_t J = 0; J < *NumCalls; ++J) {
399       auto CalledFunction(readStringFromTable());
400       if (std::error_code EC = CalledFunction.getError())
401         return EC;
402 
403       auto CalledFunctionSamples = readNumber<uint64_t>();
404       if (std::error_code EC = CalledFunctionSamples.getError())
405         return EC;
406 
407       FProfile.addCalledTargetSamples(*LineOffset, *Discriminator,
408                                       *CalledFunction, *CalledFunctionSamples);
409     }
410 
411     FProfile.addBodySamples(*LineOffset, *Discriminator, *NumSamples);
412   }
413 
414   // Read all the samples for inlined function calls.
415   auto NumCallsites = readNumber<uint32_t>();
416   if (std::error_code EC = NumCallsites.getError())
417     return EC;
418 
419   for (uint32_t J = 0; J < *NumCallsites; ++J) {
420     auto LineOffset = readNumber<uint64_t>();
421     if (std::error_code EC = LineOffset.getError())
422       return EC;
423 
424     auto Discriminator = readNumber<uint64_t>();
425     if (std::error_code EC = Discriminator.getError())
426       return EC;
427 
428     auto FName(readStringFromTable());
429     if (std::error_code EC = FName.getError())
430       return EC;
431 
432     FunctionSamples &CalleeProfile = FProfile.functionSamplesAt(
433         LineLocation(*LineOffset, *Discriminator))[std::string(*FName)];
434     CalleeProfile.setName(*FName);
435     if (std::error_code EC = readProfile(CalleeProfile))
436       return EC;
437   }
438 
439   return sampleprof_error::success;
440 }
441 
442 std::error_code
443 SampleProfileReaderBinary::readFuncProfile(const uint8_t *Start) {
444   Data = Start;
445   auto NumHeadSamples = readNumber<uint64_t>();
446   if (std::error_code EC = NumHeadSamples.getError())
447     return EC;
448 
449   auto FName(readStringFromTable());
450   if (std::error_code EC = FName.getError())
451     return EC;
452 
453   Profiles[*FName] = FunctionSamples();
454   FunctionSamples &FProfile = Profiles[*FName];
455   FProfile.setName(*FName);
456 
457   FProfile.addHeadSamples(*NumHeadSamples);
458 
459   if (std::error_code EC = readProfile(FProfile))
460     return EC;
461   return sampleprof_error::success;
462 }
463 
464 std::error_code SampleProfileReaderBinary::readImpl() {
465   while (!at_eof()) {
466     if (std::error_code EC = readFuncProfile(Data))
467       return EC;
468   }
469 
470   return sampleprof_error::success;
471 }
472 
473 std::error_code SampleProfileReaderExtBinaryBase::readOneSection(
474     const uint8_t *Start, uint64_t Size, const SecHdrTableEntry &Entry) {
475   Data = Start;
476   End = Start + Size;
477   switch (Entry.Type) {
478   case SecProfSummary:
479     if (std::error_code EC = readSummary())
480       return EC;
481     if (hasSecFlag(Entry, SecProfSummaryFlags::SecFlagPartial))
482       Summary->setPartialProfile(true);
483     break;
484   case SecNameTable:
485     if (std::error_code EC = readNameTableSec(
486             hasSecFlag(Entry, SecNameTableFlags::SecFlagMD5Name)))
487       return EC;
488     break;
489   case SecLBRProfile:
490     if (std::error_code EC = readFuncProfiles())
491       return EC;
492     break;
493   case SecFuncOffsetTable:
494     if (std::error_code EC = readFuncOffsetTable())
495       return EC;
496     break;
497   case SecProfileSymbolList:
498     if (std::error_code EC = readProfileSymbolList())
499       return EC;
500     break;
501   default:
502     if (std::error_code EC = readCustomSection(Entry))
503       return EC;
504     break;
505   }
506   return sampleprof_error::success;
507 }
508 
509 void SampleProfileReaderExtBinaryBase::collectFuncsFrom(const Module &M) {
510   UseAllFuncs = false;
511   FuncsToUse.clear();
512   for (auto &F : M)
513     FuncsToUse.insert(FunctionSamples::getCanonicalFnName(F));
514 }
515 
516 std::error_code SampleProfileReaderExtBinaryBase::readFuncOffsetTable() {
517   auto Size = readNumber<uint64_t>();
518   if (std::error_code EC = Size.getError())
519     return EC;
520 
521   FuncOffsetTable.reserve(*Size);
522   for (uint32_t I = 0; I < *Size; ++I) {
523     auto FName(readStringFromTable());
524     if (std::error_code EC = FName.getError())
525       return EC;
526 
527     auto Offset = readNumber<uint64_t>();
528     if (std::error_code EC = Offset.getError())
529       return EC;
530 
531     FuncOffsetTable[*FName] = *Offset;
532   }
533   return sampleprof_error::success;
534 }
535 
536 std::error_code SampleProfileReaderExtBinaryBase::readFuncProfiles() {
537   const uint8_t *Start = Data;
538   if (UseAllFuncs) {
539     while (Data < End) {
540       if (std::error_code EC = readFuncProfile(Data))
541         return EC;
542     }
543     assert(Data == End && "More data is read than expected");
544     return sampleprof_error::success;
545   }
546 
547   if (Remapper) {
548     for (auto Name : FuncsToUse) {
549       Remapper->insert(Name);
550     }
551   }
552 
553   if (useMD5()) {
554     for (auto Name : FuncsToUse) {
555       auto GUID = std::to_string(MD5Hash(Name));
556       auto iter = FuncOffsetTable.find(StringRef(GUID));
557       if (iter == FuncOffsetTable.end())
558         continue;
559       const uint8_t *FuncProfileAddr = Start + iter->second;
560       assert(FuncProfileAddr < End && "out of LBRProfile section");
561       if (std::error_code EC = readFuncProfile(FuncProfileAddr))
562         return EC;
563     }
564   } else {
565     for (auto NameOffset : FuncOffsetTable) {
566       auto FuncName = NameOffset.first;
567       if (!FuncsToUse.count(FuncName) &&
568           (!Remapper || !Remapper->exist(FuncName)))
569         continue;
570       const uint8_t *FuncProfileAddr = Start + NameOffset.second;
571       assert(FuncProfileAddr < End && "out of LBRProfile section");
572       if (std::error_code EC = readFuncProfile(FuncProfileAddr))
573         return EC;
574     }
575   }
576 
577   Data = End;
578   return sampleprof_error::success;
579 }
580 
581 std::error_code SampleProfileReaderExtBinaryBase::readProfileSymbolList() {
582   if (!ProfSymList)
583     ProfSymList = std::make_unique<ProfileSymbolList>();
584 
585   if (std::error_code EC = ProfSymList->read(Data, End - Data))
586     return EC;
587 
588   Data = End;
589   return sampleprof_error::success;
590 }
591 
592 std::error_code SampleProfileReaderExtBinaryBase::decompressSection(
593     const uint8_t *SecStart, const uint64_t SecSize,
594     const uint8_t *&DecompressBuf, uint64_t &DecompressBufSize) {
595   Data = SecStart;
596   End = SecStart + SecSize;
597   auto DecompressSize = readNumber<uint64_t>();
598   if (std::error_code EC = DecompressSize.getError())
599     return EC;
600   DecompressBufSize = *DecompressSize;
601 
602   auto CompressSize = readNumber<uint64_t>();
603   if (std::error_code EC = CompressSize.getError())
604     return EC;
605 
606   if (!llvm::zlib::isAvailable())
607     return sampleprof_error::zlib_unavailable;
608 
609   StringRef CompressedStrings(reinterpret_cast<const char *>(Data),
610                               *CompressSize);
611   char *Buffer = Allocator.Allocate<char>(DecompressBufSize);
612   size_t UCSize = DecompressBufSize;
613   llvm::Error E =
614       zlib::uncompress(CompressedStrings, Buffer, UCSize);
615   if (E)
616     return sampleprof_error::uncompress_failed;
617   DecompressBuf = reinterpret_cast<const uint8_t *>(Buffer);
618   return sampleprof_error::success;
619 }
620 
621 std::error_code SampleProfileReaderExtBinaryBase::readImpl() {
622   const uint8_t *BufStart =
623       reinterpret_cast<const uint8_t *>(Buffer->getBufferStart());
624 
625   for (auto &Entry : SecHdrTable) {
626     // Skip empty section.
627     if (!Entry.Size)
628       continue;
629 
630     const uint8_t *SecStart = BufStart + Entry.Offset;
631     uint64_t SecSize = Entry.Size;
632 
633     // If the section is compressed, decompress it into a buffer
634     // DecompressBuf before reading the actual data. The pointee of
635     // 'Data' will be changed to buffer hold by DecompressBuf
636     // temporarily when reading the actual data.
637     bool isCompressed = hasSecFlag(Entry, SecCommonFlags::SecFlagCompress);
638     if (isCompressed) {
639       const uint8_t *DecompressBuf;
640       uint64_t DecompressBufSize;
641       if (std::error_code EC = decompressSection(
642               SecStart, SecSize, DecompressBuf, DecompressBufSize))
643         return EC;
644       SecStart = DecompressBuf;
645       SecSize = DecompressBufSize;
646     }
647 
648     if (std::error_code EC = readOneSection(SecStart, SecSize, Entry))
649       return EC;
650     if (Data != SecStart + SecSize)
651       return sampleprof_error::malformed;
652 
653     // Change the pointee of 'Data' from DecompressBuf to original Buffer.
654     if (isCompressed) {
655       Data = BufStart + Entry.Offset;
656       End = BufStart + Buffer->getBufferSize();
657     }
658   }
659 
660   return sampleprof_error::success;
661 }
662 
663 std::error_code SampleProfileReaderCompactBinary::readImpl() {
664   std::vector<uint64_t> OffsetsToUse;
665   if (UseAllFuncs) {
666     for (auto FuncEntry : FuncOffsetTable) {
667       OffsetsToUse.push_back(FuncEntry.second);
668     }
669   }
670   else {
671     for (auto Name : FuncsToUse) {
672       auto GUID = std::to_string(MD5Hash(Name));
673       auto iter = FuncOffsetTable.find(StringRef(GUID));
674       if (iter == FuncOffsetTable.end())
675         continue;
676       OffsetsToUse.push_back(iter->second);
677     }
678   }
679 
680   for (auto Offset : OffsetsToUse) {
681     const uint8_t *SavedData = Data;
682     if (std::error_code EC = readFuncProfile(
683             reinterpret_cast<const uint8_t *>(Buffer->getBufferStart()) +
684             Offset))
685       return EC;
686     Data = SavedData;
687   }
688   return sampleprof_error::success;
689 }
690 
691 std::error_code SampleProfileReaderRawBinary::verifySPMagic(uint64_t Magic) {
692   if (Magic == SPMagic())
693     return sampleprof_error::success;
694   return sampleprof_error::bad_magic;
695 }
696 
697 std::error_code SampleProfileReaderExtBinary::verifySPMagic(uint64_t Magic) {
698   if (Magic == SPMagic(SPF_Ext_Binary))
699     return sampleprof_error::success;
700   return sampleprof_error::bad_magic;
701 }
702 
703 std::error_code
704 SampleProfileReaderCompactBinary::verifySPMagic(uint64_t Magic) {
705   if (Magic == SPMagic(SPF_Compact_Binary))
706     return sampleprof_error::success;
707   return sampleprof_error::bad_magic;
708 }
709 
710 std::error_code SampleProfileReaderBinary::readNameTable() {
711   auto Size = readNumber<uint32_t>();
712   if (std::error_code EC = Size.getError())
713     return EC;
714   NameTable.reserve(*Size);
715   for (uint32_t I = 0; I < *Size; ++I) {
716     auto Name(readString());
717     if (std::error_code EC = Name.getError())
718       return EC;
719     NameTable.push_back(*Name);
720   }
721 
722   return sampleprof_error::success;
723 }
724 
725 std::error_code SampleProfileReaderExtBinaryBase::readMD5NameTable() {
726   auto Size = readNumber<uint64_t>();
727   if (std::error_code EC = Size.getError())
728     return EC;
729   NameTable.reserve(*Size);
730   MD5StringBuf = std::make_unique<std::vector<std::string>>();
731   MD5StringBuf->reserve(*Size);
732   for (uint32_t I = 0; I < *Size; ++I) {
733     auto FID = readNumber<uint64_t>();
734     if (std::error_code EC = FID.getError())
735       return EC;
736     MD5StringBuf->push_back(std::to_string(*FID));
737     // NameTable is a vector of StringRef. Here it is pushing back a
738     // StringRef initialized with the last string in MD5stringBuf.
739     NameTable.push_back(MD5StringBuf->back());
740   }
741   return sampleprof_error::success;
742 }
743 
744 std::error_code SampleProfileReaderExtBinaryBase::readNameTableSec(bool IsMD5) {
745   if (IsMD5)
746     return readMD5NameTable();
747   return SampleProfileReaderBinary::readNameTable();
748 }
749 
750 std::error_code SampleProfileReaderCompactBinary::readNameTable() {
751   auto Size = readNumber<uint64_t>();
752   if (std::error_code EC = Size.getError())
753     return EC;
754   NameTable.reserve(*Size);
755   for (uint32_t I = 0; I < *Size; ++I) {
756     auto FID = readNumber<uint64_t>();
757     if (std::error_code EC = FID.getError())
758       return EC;
759     NameTable.push_back(std::to_string(*FID));
760   }
761   return sampleprof_error::success;
762 }
763 
764 std::error_code SampleProfileReaderExtBinaryBase::readSecHdrTableEntry() {
765   SecHdrTableEntry Entry;
766   auto Type = readUnencodedNumber<uint64_t>();
767   if (std::error_code EC = Type.getError())
768     return EC;
769   Entry.Type = static_cast<SecType>(*Type);
770 
771   auto Flags = readUnencodedNumber<uint64_t>();
772   if (std::error_code EC = Flags.getError())
773     return EC;
774   Entry.Flags = *Flags;
775 
776   auto Offset = readUnencodedNumber<uint64_t>();
777   if (std::error_code EC = Offset.getError())
778     return EC;
779   Entry.Offset = *Offset;
780 
781   auto Size = readUnencodedNumber<uint64_t>();
782   if (std::error_code EC = Size.getError())
783     return EC;
784   Entry.Size = *Size;
785 
786   SecHdrTable.push_back(std::move(Entry));
787   return sampleprof_error::success;
788 }
789 
790 std::error_code SampleProfileReaderExtBinaryBase::readSecHdrTable() {
791   auto EntryNum = readUnencodedNumber<uint64_t>();
792   if (std::error_code EC = EntryNum.getError())
793     return EC;
794 
795   for (uint32_t i = 0; i < (*EntryNum); i++)
796     if (std::error_code EC = readSecHdrTableEntry())
797       return EC;
798 
799   return sampleprof_error::success;
800 }
801 
802 std::error_code SampleProfileReaderExtBinaryBase::readHeader() {
803   const uint8_t *BufStart =
804       reinterpret_cast<const uint8_t *>(Buffer->getBufferStart());
805   Data = BufStart;
806   End = BufStart + Buffer->getBufferSize();
807 
808   if (std::error_code EC = readMagicIdent())
809     return EC;
810 
811   if (std::error_code EC = readSecHdrTable())
812     return EC;
813 
814   return sampleprof_error::success;
815 }
816 
817 uint64_t SampleProfileReaderExtBinaryBase::getSectionSize(SecType Type) {
818   for (auto &Entry : SecHdrTable) {
819     if (Entry.Type == Type)
820       return Entry.Size;
821   }
822   return 0;
823 }
824 
825 uint64_t SampleProfileReaderExtBinaryBase::getFileSize() {
826   // Sections in SecHdrTable is not necessarily in the same order as
827   // sections in the profile because section like FuncOffsetTable needs
828   // to be written after section LBRProfile but needs to be read before
829   // section LBRProfile, so we cannot simply use the last entry in
830   // SecHdrTable to calculate the file size.
831   uint64_t FileSize = 0;
832   for (auto &Entry : SecHdrTable) {
833     FileSize = std::max(Entry.Offset + Entry.Size, FileSize);
834   }
835   return FileSize;
836 }
837 
838 static std::string getSecFlagsStr(const SecHdrTableEntry &Entry) {
839   std::string Flags;
840   if (hasSecFlag(Entry, SecCommonFlags::SecFlagCompress))
841     Flags.append("{compressed,");
842   else
843     Flags.append("{");
844 
845   switch (Entry.Type) {
846   case SecNameTable:
847     if (hasSecFlag(Entry, SecNameTableFlags::SecFlagMD5Name))
848       Flags.append("md5,");
849     break;
850   case SecProfSummary:
851     if (hasSecFlag(Entry, SecProfSummaryFlags::SecFlagPartial))
852       Flags.append("partial,");
853     break;
854   default:
855     break;
856   }
857   char &last = Flags.back();
858   if (last == ',')
859     last = '}';
860   else
861     Flags.append("}");
862   return Flags;
863 }
864 
865 bool SampleProfileReaderExtBinaryBase::dumpSectionInfo(raw_ostream &OS) {
866   uint64_t TotalSecsSize = 0;
867   for (auto &Entry : SecHdrTable) {
868     OS << getSecName(Entry.Type) << " - Offset: " << Entry.Offset
869        << ", Size: " << Entry.Size << ", Flags: " << getSecFlagsStr(Entry)
870        << "\n";
871     ;
872     TotalSecsSize += getSectionSize(Entry.Type);
873   }
874   uint64_t HeaderSize = SecHdrTable.front().Offset;
875   assert(HeaderSize + TotalSecsSize == getFileSize() &&
876          "Size of 'header + sections' doesn't match the total size of profile");
877 
878   OS << "Header Size: " << HeaderSize << "\n";
879   OS << "Total Sections Size: " << TotalSecsSize << "\n";
880   OS << "File Size: " << getFileSize() << "\n";
881   return true;
882 }
883 
884 std::error_code SampleProfileReaderBinary::readMagicIdent() {
885   // Read and check the magic identifier.
886   auto Magic = readNumber<uint64_t>();
887   if (std::error_code EC = Magic.getError())
888     return EC;
889   else if (std::error_code EC = verifySPMagic(*Magic))
890     return EC;
891 
892   // Read the version number.
893   auto Version = readNumber<uint64_t>();
894   if (std::error_code EC = Version.getError())
895     return EC;
896   else if (*Version != SPVersion())
897     return sampleprof_error::unsupported_version;
898 
899   return sampleprof_error::success;
900 }
901 
902 std::error_code SampleProfileReaderBinary::readHeader() {
903   Data = reinterpret_cast<const uint8_t *>(Buffer->getBufferStart());
904   End = Data + Buffer->getBufferSize();
905 
906   if (std::error_code EC = readMagicIdent())
907     return EC;
908 
909   if (std::error_code EC = readSummary())
910     return EC;
911 
912   if (std::error_code EC = readNameTable())
913     return EC;
914   return sampleprof_error::success;
915 }
916 
917 std::error_code SampleProfileReaderCompactBinary::readHeader() {
918   SampleProfileReaderBinary::readHeader();
919   if (std::error_code EC = readFuncOffsetTable())
920     return EC;
921   return sampleprof_error::success;
922 }
923 
924 std::error_code SampleProfileReaderCompactBinary::readFuncOffsetTable() {
925   auto TableOffset = readUnencodedNumber<uint64_t>();
926   if (std::error_code EC = TableOffset.getError())
927     return EC;
928 
929   const uint8_t *SavedData = Data;
930   const uint8_t *TableStart =
931       reinterpret_cast<const uint8_t *>(Buffer->getBufferStart()) +
932       *TableOffset;
933   Data = TableStart;
934 
935   auto Size = readNumber<uint64_t>();
936   if (std::error_code EC = Size.getError())
937     return EC;
938 
939   FuncOffsetTable.reserve(*Size);
940   for (uint32_t I = 0; I < *Size; ++I) {
941     auto FName(readStringFromTable());
942     if (std::error_code EC = FName.getError())
943       return EC;
944 
945     auto Offset = readNumber<uint64_t>();
946     if (std::error_code EC = Offset.getError())
947       return EC;
948 
949     FuncOffsetTable[*FName] = *Offset;
950   }
951   End = TableStart;
952   Data = SavedData;
953   return sampleprof_error::success;
954 }
955 
956 void SampleProfileReaderCompactBinary::collectFuncsFrom(const Module &M) {
957   UseAllFuncs = false;
958   FuncsToUse.clear();
959   for (auto &F : M)
960     FuncsToUse.insert(FunctionSamples::getCanonicalFnName(F));
961 }
962 
963 std::error_code SampleProfileReaderBinary::readSummaryEntry(
964     std::vector<ProfileSummaryEntry> &Entries) {
965   auto Cutoff = readNumber<uint64_t>();
966   if (std::error_code EC = Cutoff.getError())
967     return EC;
968 
969   auto MinBlockCount = readNumber<uint64_t>();
970   if (std::error_code EC = MinBlockCount.getError())
971     return EC;
972 
973   auto NumBlocks = readNumber<uint64_t>();
974   if (std::error_code EC = NumBlocks.getError())
975     return EC;
976 
977   Entries.emplace_back(*Cutoff, *MinBlockCount, *NumBlocks);
978   return sampleprof_error::success;
979 }
980 
981 std::error_code SampleProfileReaderBinary::readSummary() {
982   auto TotalCount = readNumber<uint64_t>();
983   if (std::error_code EC = TotalCount.getError())
984     return EC;
985 
986   auto MaxBlockCount = readNumber<uint64_t>();
987   if (std::error_code EC = MaxBlockCount.getError())
988     return EC;
989 
990   auto MaxFunctionCount = readNumber<uint64_t>();
991   if (std::error_code EC = MaxFunctionCount.getError())
992     return EC;
993 
994   auto NumBlocks = readNumber<uint64_t>();
995   if (std::error_code EC = NumBlocks.getError())
996     return EC;
997 
998   auto NumFunctions = readNumber<uint64_t>();
999   if (std::error_code EC = NumFunctions.getError())
1000     return EC;
1001 
1002   auto NumSummaryEntries = readNumber<uint64_t>();
1003   if (std::error_code EC = NumSummaryEntries.getError())
1004     return EC;
1005 
1006   std::vector<ProfileSummaryEntry> Entries;
1007   for (unsigned i = 0; i < *NumSummaryEntries; i++) {
1008     std::error_code EC = readSummaryEntry(Entries);
1009     if (EC != sampleprof_error::success)
1010       return EC;
1011   }
1012   Summary = std::make_unique<ProfileSummary>(
1013       ProfileSummary::PSK_Sample, Entries, *TotalCount, *MaxBlockCount, 0,
1014       *MaxFunctionCount, *NumBlocks, *NumFunctions);
1015 
1016   return sampleprof_error::success;
1017 }
1018 
1019 bool SampleProfileReaderRawBinary::hasFormat(const MemoryBuffer &Buffer) {
1020   const uint8_t *Data =
1021       reinterpret_cast<const uint8_t *>(Buffer.getBufferStart());
1022   uint64_t Magic = decodeULEB128(Data);
1023   return Magic == SPMagic();
1024 }
1025 
1026 bool SampleProfileReaderExtBinary::hasFormat(const MemoryBuffer &Buffer) {
1027   const uint8_t *Data =
1028       reinterpret_cast<const uint8_t *>(Buffer.getBufferStart());
1029   uint64_t Magic = decodeULEB128(Data);
1030   return Magic == SPMagic(SPF_Ext_Binary);
1031 }
1032 
1033 bool SampleProfileReaderCompactBinary::hasFormat(const MemoryBuffer &Buffer) {
1034   const uint8_t *Data =
1035       reinterpret_cast<const uint8_t *>(Buffer.getBufferStart());
1036   uint64_t Magic = decodeULEB128(Data);
1037   return Magic == SPMagic(SPF_Compact_Binary);
1038 }
1039 
1040 std::error_code SampleProfileReaderGCC::skipNextWord() {
1041   uint32_t dummy;
1042   if (!GcovBuffer.readInt(dummy))
1043     return sampleprof_error::truncated;
1044   return sampleprof_error::success;
1045 }
1046 
1047 template <typename T> ErrorOr<T> SampleProfileReaderGCC::readNumber() {
1048   if (sizeof(T) <= sizeof(uint32_t)) {
1049     uint32_t Val;
1050     if (GcovBuffer.readInt(Val) && Val <= std::numeric_limits<T>::max())
1051       return static_cast<T>(Val);
1052   } else if (sizeof(T) <= sizeof(uint64_t)) {
1053     uint64_t Val;
1054     if (GcovBuffer.readInt64(Val) && Val <= std::numeric_limits<T>::max())
1055       return static_cast<T>(Val);
1056   }
1057 
1058   std::error_code EC = sampleprof_error::malformed;
1059   reportError(0, EC.message());
1060   return EC;
1061 }
1062 
1063 ErrorOr<StringRef> SampleProfileReaderGCC::readString() {
1064   StringRef Str;
1065   if (!GcovBuffer.readString(Str))
1066     return sampleprof_error::truncated;
1067   return Str;
1068 }
1069 
1070 std::error_code SampleProfileReaderGCC::readHeader() {
1071   // Read the magic identifier.
1072   if (!GcovBuffer.readGCDAFormat())
1073     return sampleprof_error::unrecognized_format;
1074 
1075   // Read the version number. Note - the GCC reader does not validate this
1076   // version, but the profile creator generates v704.
1077   GCOV::GCOVVersion version;
1078   if (!GcovBuffer.readGCOVVersion(version))
1079     return sampleprof_error::unrecognized_format;
1080 
1081   if (version != GCOV::V407)
1082     return sampleprof_error::unsupported_version;
1083 
1084   // Skip the empty integer.
1085   if (std::error_code EC = skipNextWord())
1086     return EC;
1087 
1088   return sampleprof_error::success;
1089 }
1090 
1091 std::error_code SampleProfileReaderGCC::readSectionTag(uint32_t Expected) {
1092   uint32_t Tag;
1093   if (!GcovBuffer.readInt(Tag))
1094     return sampleprof_error::truncated;
1095 
1096   if (Tag != Expected)
1097     return sampleprof_error::malformed;
1098 
1099   if (std::error_code EC = skipNextWord())
1100     return EC;
1101 
1102   return sampleprof_error::success;
1103 }
1104 
1105 std::error_code SampleProfileReaderGCC::readNameTable() {
1106   if (std::error_code EC = readSectionTag(GCOVTagAFDOFileNames))
1107     return EC;
1108 
1109   uint32_t Size;
1110   if (!GcovBuffer.readInt(Size))
1111     return sampleprof_error::truncated;
1112 
1113   for (uint32_t I = 0; I < Size; ++I) {
1114     StringRef Str;
1115     if (!GcovBuffer.readString(Str))
1116       return sampleprof_error::truncated;
1117     Names.push_back(std::string(Str));
1118   }
1119 
1120   return sampleprof_error::success;
1121 }
1122 
1123 std::error_code SampleProfileReaderGCC::readFunctionProfiles() {
1124   if (std::error_code EC = readSectionTag(GCOVTagAFDOFunction))
1125     return EC;
1126 
1127   uint32_t NumFunctions;
1128   if (!GcovBuffer.readInt(NumFunctions))
1129     return sampleprof_error::truncated;
1130 
1131   InlineCallStack Stack;
1132   for (uint32_t I = 0; I < NumFunctions; ++I)
1133     if (std::error_code EC = readOneFunctionProfile(Stack, true, 0))
1134       return EC;
1135 
1136   computeSummary();
1137   return sampleprof_error::success;
1138 }
1139 
1140 std::error_code SampleProfileReaderGCC::readOneFunctionProfile(
1141     const InlineCallStack &InlineStack, bool Update, uint32_t Offset) {
1142   uint64_t HeadCount = 0;
1143   if (InlineStack.size() == 0)
1144     if (!GcovBuffer.readInt64(HeadCount))
1145       return sampleprof_error::truncated;
1146 
1147   uint32_t NameIdx;
1148   if (!GcovBuffer.readInt(NameIdx))
1149     return sampleprof_error::truncated;
1150 
1151   StringRef Name(Names[NameIdx]);
1152 
1153   uint32_t NumPosCounts;
1154   if (!GcovBuffer.readInt(NumPosCounts))
1155     return sampleprof_error::truncated;
1156 
1157   uint32_t NumCallsites;
1158   if (!GcovBuffer.readInt(NumCallsites))
1159     return sampleprof_error::truncated;
1160 
1161   FunctionSamples *FProfile = nullptr;
1162   if (InlineStack.size() == 0) {
1163     // If this is a top function that we have already processed, do not
1164     // update its profile again.  This happens in the presence of
1165     // function aliases.  Since these aliases share the same function
1166     // body, there will be identical replicated profiles for the
1167     // original function.  In this case, we simply not bother updating
1168     // the profile of the original function.
1169     FProfile = &Profiles[Name];
1170     FProfile->addHeadSamples(HeadCount);
1171     if (FProfile->getTotalSamples() > 0)
1172       Update = false;
1173   } else {
1174     // Otherwise, we are reading an inlined instance. The top of the
1175     // inline stack contains the profile of the caller. Insert this
1176     // callee in the caller's CallsiteMap.
1177     FunctionSamples *CallerProfile = InlineStack.front();
1178     uint32_t LineOffset = Offset >> 16;
1179     uint32_t Discriminator = Offset & 0xffff;
1180     FProfile = &CallerProfile->functionSamplesAt(
1181         LineLocation(LineOffset, Discriminator))[std::string(Name)];
1182   }
1183   FProfile->setName(Name);
1184 
1185   for (uint32_t I = 0; I < NumPosCounts; ++I) {
1186     uint32_t Offset;
1187     if (!GcovBuffer.readInt(Offset))
1188       return sampleprof_error::truncated;
1189 
1190     uint32_t NumTargets;
1191     if (!GcovBuffer.readInt(NumTargets))
1192       return sampleprof_error::truncated;
1193 
1194     uint64_t Count;
1195     if (!GcovBuffer.readInt64(Count))
1196       return sampleprof_error::truncated;
1197 
1198     // The line location is encoded in the offset as:
1199     //   high 16 bits: line offset to the start of the function.
1200     //   low 16 bits: discriminator.
1201     uint32_t LineOffset = Offset >> 16;
1202     uint32_t Discriminator = Offset & 0xffff;
1203 
1204     InlineCallStack NewStack;
1205     NewStack.push_back(FProfile);
1206     NewStack.insert(NewStack.end(), InlineStack.begin(), InlineStack.end());
1207     if (Update) {
1208       // Walk up the inline stack, adding the samples on this line to
1209       // the total sample count of the callers in the chain.
1210       for (auto CallerProfile : NewStack)
1211         CallerProfile->addTotalSamples(Count);
1212 
1213       // Update the body samples for the current profile.
1214       FProfile->addBodySamples(LineOffset, Discriminator, Count);
1215     }
1216 
1217     // Process the list of functions called at an indirect call site.
1218     // These are all the targets that a function pointer (or virtual
1219     // function) resolved at runtime.
1220     for (uint32_t J = 0; J < NumTargets; J++) {
1221       uint32_t HistVal;
1222       if (!GcovBuffer.readInt(HistVal))
1223         return sampleprof_error::truncated;
1224 
1225       if (HistVal != HIST_TYPE_INDIR_CALL_TOPN)
1226         return sampleprof_error::malformed;
1227 
1228       uint64_t TargetIdx;
1229       if (!GcovBuffer.readInt64(TargetIdx))
1230         return sampleprof_error::truncated;
1231       StringRef TargetName(Names[TargetIdx]);
1232 
1233       uint64_t TargetCount;
1234       if (!GcovBuffer.readInt64(TargetCount))
1235         return sampleprof_error::truncated;
1236 
1237       if (Update)
1238         FProfile->addCalledTargetSamples(LineOffset, Discriminator,
1239                                          TargetName, TargetCount);
1240     }
1241   }
1242 
1243   // Process all the inlined callers into the current function. These
1244   // are all the callsites that were inlined into this function.
1245   for (uint32_t I = 0; I < NumCallsites; I++) {
1246     // The offset is encoded as:
1247     //   high 16 bits: line offset to the start of the function.
1248     //   low 16 bits: discriminator.
1249     uint32_t Offset;
1250     if (!GcovBuffer.readInt(Offset))
1251       return sampleprof_error::truncated;
1252     InlineCallStack NewStack;
1253     NewStack.push_back(FProfile);
1254     NewStack.insert(NewStack.end(), InlineStack.begin(), InlineStack.end());
1255     if (std::error_code EC = readOneFunctionProfile(NewStack, Update, Offset))
1256       return EC;
1257   }
1258 
1259   return sampleprof_error::success;
1260 }
1261 
1262 /// Read a GCC AutoFDO profile.
1263 ///
1264 /// This format is generated by the Linux Perf conversion tool at
1265 /// https://github.com/google/autofdo.
1266 std::error_code SampleProfileReaderGCC::readImpl() {
1267   // Read the string table.
1268   if (std::error_code EC = readNameTable())
1269     return EC;
1270 
1271   // Read the source profile.
1272   if (std::error_code EC = readFunctionProfiles())
1273     return EC;
1274 
1275   return sampleprof_error::success;
1276 }
1277 
1278 bool SampleProfileReaderGCC::hasFormat(const MemoryBuffer &Buffer) {
1279   StringRef Magic(reinterpret_cast<const char *>(Buffer.getBufferStart()));
1280   return Magic == "adcg*704";
1281 }
1282 
1283 void SampleProfileReaderItaniumRemapper::applyRemapping(LLVMContext &Ctx) {
1284   // If the reader uses MD5 to represent string, we can't remap it because
1285   // we don't know what the original function names were.
1286   if (Reader.useMD5()) {
1287     Ctx.diagnose(DiagnosticInfoSampleProfile(
1288         Reader.getBuffer()->getBufferIdentifier(),
1289         "Profile data remapping cannot be applied to profile data "
1290         "in compact format (original mangled names are not available).",
1291         DS_Warning));
1292     return;
1293   }
1294 
1295   assert(Remappings && "should be initialized while creating remapper");
1296   for (auto &Sample : Reader.getProfiles()) {
1297     DenseSet<StringRef> NamesInSample;
1298     Sample.second.findAllNames(NamesInSample);
1299     for (auto &Name : NamesInSample)
1300       if (auto Key = Remappings->insert(Name))
1301         NameMap.insert({Key, Name});
1302   }
1303 
1304   RemappingApplied = true;
1305 }
1306 
1307 Optional<StringRef>
1308 SampleProfileReaderItaniumRemapper::lookUpNameInProfile(StringRef Fname) {
1309   if (auto Key = Remappings->lookup(Fname))
1310     return NameMap.lookup(Key);
1311   return None;
1312 }
1313 
1314 /// Prepare a memory buffer for the contents of \p Filename.
1315 ///
1316 /// \returns an error code indicating the status of the buffer.
1317 static ErrorOr<std::unique_ptr<MemoryBuffer>>
1318 setupMemoryBuffer(const Twine &Filename) {
1319   auto BufferOrErr = MemoryBuffer::getFileOrSTDIN(Filename);
1320   if (std::error_code EC = BufferOrErr.getError())
1321     return EC;
1322   auto Buffer = std::move(BufferOrErr.get());
1323 
1324   // Sanity check the file.
1325   if (uint64_t(Buffer->getBufferSize()) > std::numeric_limits<uint32_t>::max())
1326     return sampleprof_error::too_large;
1327 
1328   return std::move(Buffer);
1329 }
1330 
1331 /// Create a sample profile reader based on the format of the input file.
1332 ///
1333 /// \param Filename The file to open.
1334 ///
1335 /// \param C The LLVM context to use to emit diagnostics.
1336 ///
1337 /// \param RemapFilename The file used for profile remapping.
1338 ///
1339 /// \returns an error code indicating the status of the created reader.
1340 ErrorOr<std::unique_ptr<SampleProfileReader>>
1341 SampleProfileReader::create(const std::string Filename, LLVMContext &C,
1342                             const std::string RemapFilename) {
1343   auto BufferOrError = setupMemoryBuffer(Filename);
1344   if (std::error_code EC = BufferOrError.getError())
1345     return EC;
1346   return create(BufferOrError.get(), C, RemapFilename);
1347 }
1348 
1349 /// Create a sample profile remapper from the given input, to remap the
1350 /// function names in the given profile data.
1351 ///
1352 /// \param Filename The file to open.
1353 ///
1354 /// \param Reader The profile reader the remapper is going to be applied to.
1355 ///
1356 /// \param C The LLVM context to use to emit diagnostics.
1357 ///
1358 /// \returns an error code indicating the status of the created reader.
1359 ErrorOr<std::unique_ptr<SampleProfileReaderItaniumRemapper>>
1360 SampleProfileReaderItaniumRemapper::create(const std::string Filename,
1361                                            SampleProfileReader &Reader,
1362                                            LLVMContext &C) {
1363   auto BufferOrError = setupMemoryBuffer(Filename);
1364   if (std::error_code EC = BufferOrError.getError())
1365     return EC;
1366   return create(BufferOrError.get(), Reader, C);
1367 }
1368 
1369 /// Create a sample profile remapper from the given input, to remap the
1370 /// function names in the given profile data.
1371 ///
1372 /// \param B The memory buffer to create the reader from (assumes ownership).
1373 ///
1374 /// \param C The LLVM context to use to emit diagnostics.
1375 ///
1376 /// \param Reader The profile reader the remapper is going to be applied to.
1377 ///
1378 /// \returns an error code indicating the status of the created reader.
1379 ErrorOr<std::unique_ptr<SampleProfileReaderItaniumRemapper>>
1380 SampleProfileReaderItaniumRemapper::create(std::unique_ptr<MemoryBuffer> &B,
1381                                            SampleProfileReader &Reader,
1382                                            LLVMContext &C) {
1383   auto Remappings = std::make_unique<SymbolRemappingReader>();
1384   if (Error E = Remappings->read(*B.get())) {
1385     handleAllErrors(
1386         std::move(E), [&](const SymbolRemappingParseError &ParseError) {
1387           C.diagnose(DiagnosticInfoSampleProfile(B->getBufferIdentifier(),
1388                                                  ParseError.getLineNum(),
1389                                                  ParseError.getMessage()));
1390         });
1391     return sampleprof_error::malformed;
1392   }
1393 
1394   return std::make_unique<SampleProfileReaderItaniumRemapper>(
1395       std::move(B), std::move(Remappings), Reader);
1396 }
1397 
1398 /// Create a sample profile reader based on the format of the input data.
1399 ///
1400 /// \param B The memory buffer to create the reader from (assumes ownership).
1401 ///
1402 /// \param C The LLVM context to use to emit diagnostics.
1403 ///
1404 /// \param RemapFilename The file used for profile remapping.
1405 ///
1406 /// \returns an error code indicating the status of the created reader.
1407 ErrorOr<std::unique_ptr<SampleProfileReader>>
1408 SampleProfileReader::create(std::unique_ptr<MemoryBuffer> &B, LLVMContext &C,
1409                             const std::string RemapFilename) {
1410   std::unique_ptr<SampleProfileReader> Reader;
1411   if (SampleProfileReaderRawBinary::hasFormat(*B))
1412     Reader.reset(new SampleProfileReaderRawBinary(std::move(B), C));
1413   else if (SampleProfileReaderExtBinary::hasFormat(*B))
1414     Reader.reset(new SampleProfileReaderExtBinary(std::move(B), C));
1415   else if (SampleProfileReaderCompactBinary::hasFormat(*B))
1416     Reader.reset(new SampleProfileReaderCompactBinary(std::move(B), C));
1417   else if (SampleProfileReaderGCC::hasFormat(*B))
1418     Reader.reset(new SampleProfileReaderGCC(std::move(B), C));
1419   else if (SampleProfileReaderText::hasFormat(*B))
1420     Reader.reset(new SampleProfileReaderText(std::move(B), C));
1421   else
1422     return sampleprof_error::unrecognized_format;
1423 
1424   if (!RemapFilename.empty()) {
1425     auto ReaderOrErr =
1426         SampleProfileReaderItaniumRemapper::create(RemapFilename, *Reader, C);
1427     if (std::error_code EC = ReaderOrErr.getError()) {
1428       std::string Msg = "Could not create remapper: " + EC.message();
1429       C.diagnose(DiagnosticInfoSampleProfile(RemapFilename, Msg));
1430       return EC;
1431     }
1432     Reader->Remapper = std::move(ReaderOrErr.get());
1433   }
1434 
1435   FunctionSamples::Format = Reader->getFormat();
1436   if (std::error_code EC = Reader->readHeader()) {
1437     return EC;
1438   }
1439 
1440   return std::move(Reader);
1441 }
1442 
1443 // For text and GCC file formats, we compute the summary after reading the
1444 // profile. Binary format has the profile summary in its header.
1445 void SampleProfileReader::computeSummary() {
1446   SampleProfileSummaryBuilder Builder(ProfileSummaryBuilder::DefaultCutoffs);
1447   for (const auto &I : Profiles) {
1448     const FunctionSamples &Profile = I.second;
1449     Builder.addRecord(Profile);
1450   }
1451   Summary = Builder.getSummary();
1452 }
1453