[llvm] r257985 - [libFuzzer] replace vector with a simpler data structure in the Dictionaries to avoid memory allocations on hot path

David Blaikie via llvm-commits llvm-commits at lists.llvm.org
Fri Jan 15 20:10:20 PST 2016


On Fri, Jan 15, 2016 at 7:53 PM, Kostya Serebryany via llvm-commits <
llvm-commits at lists.llvm.org> wrote:

> Author: kcc
> Date: Fri Jan 15 21:53:32 2016
> New Revision: 257985
>
> URL: http://llvm.org/viewvc/llvm-project?rev=257985&view=rev
> Log:
> [libFuzzer] replace vector with a simpler data structure in the
> Dictionaries to avoid memory allocations on hot path
>
> Modified:
>     llvm/trunk/lib/Fuzzer/FuzzerDriver.cpp
>     llvm/trunk/lib/Fuzzer/FuzzerInternal.h
>     llvm/trunk/lib/Fuzzer/FuzzerMutate.cpp
>     llvm/trunk/lib/Fuzzer/FuzzerTraceState.cpp
>     llvm/trunk/lib/Fuzzer/FuzzerUtil.cpp
>     llvm/trunk/lib/Fuzzer/test/FuzzerUnittest.cpp
>
> Modified: llvm/trunk/lib/Fuzzer/FuzzerDriver.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Fuzzer/FuzzerDriver.cpp?rev=257985&r1=257984&r2=257985&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Fuzzer/FuzzerDriver.cpp (original)
> +++ llvm/trunk/lib/Fuzzer/FuzzerDriver.cpp Fri Jan 15 21:53:32 2016
> @@ -301,7 +301,8 @@ int FuzzerDriver(const std::vector<std::
>    Fuzzer F(USF, Options);
>
>    for (auto &U: Dictionary)
> -    USF.GetMD().AddWordToManualDictionary(U);
> +    if (U.size() <= Word::GetMaxSize())
> +      USF.GetMD().AddWordToManualDictionary(Word(U.data(), U.size()));
>
>    // Timer
>    if (Flags.timeout > 0)
>
> Modified: llvm/trunk/lib/Fuzzer/FuzzerInternal.h
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Fuzzer/FuzzerInternal.h?rev=257985&r1=257984&r2=257985&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Fuzzer/FuzzerInternal.h (original)
> +++ llvm/trunk/lib/Fuzzer/FuzzerInternal.h Fri Jan 15 21:53:32 2016
> @@ -18,6 +18,7 @@
>  #include <cstddef>
>  #include <cstdlib>
>  #include <string>
> +#include <string.h>
>  #include <vector>
>  #include <unordered_set>
>
> @@ -27,6 +28,40 @@ namespace fuzzer {
>  using namespace std::chrono;
>  typedef std::vector<uint8_t> Unit;
>
> +// A simple POD sized array of bytes.
> +template<size_t kMaxSize>
> +class FixedWord {
> + public:
> +
> +  FixedWord() : Size(0) {}
> +  FixedWord(const uint8_t *B, uint8_t S) { Set(B, S); }
> +
> +  void Set(const uint8_t *B, uint8_t S) {
> +    assert(S <= kMaxSize);
> +    memcpy(Data, B, S);
> +    Size = S;
> +  }
> +
> +  bool operator == (const FixedWord<kMaxSize> &w) const {
> +    return Size == w.Size && 0 == memcmp(Data, w.Data, Size);
> +  }
> +
> +  bool operator < (const FixedWord<kMaxSize> &w) const {
> +    if (Size != w.Size) return Size < w.Size;
> +    return memcmp(Data, w.Data, Size) < 0;
> +  }
> +
> +  static size_t GetMaxSize() { return kMaxSize; }
> +  const uint8_t *data() const { return Data; }
> +  uint8_t size() const { return Size; }
> +
> + private:
> +  uint8_t Size;
> +  uint8_t Data[kMaxSize];
> +};
> +
> +typedef FixedWord<27> Word;  // 28 bytes.
> +
>  std::string FileToString(const std::string &Path);
>  Unit FileToVector(const std::string &Path);
>  void ReadDirToVectorOfUnits(const char *Path, std::vector<Unit> *V,
> @@ -43,6 +78,7 @@ void PrintHexArray(const uint8_t *Data,
>                     const char *PrintAfter = "");
>  void PrintASCII(const uint8_t *Data, size_t Size, const char *PrintAfter
> = "");
>  void PrintASCII(const Unit &U, const char *PrintAfter = "");
> +void PrintASCII(const Word &W, const char *PrintAfter = "");
>  std::string Hash(const Unit &U);
>  void SetTimer(int Seconds);
>  std::string Base64(const Unit &U);
> @@ -118,9 +154,9 @@ class MutationDispatcher {
>    size_t CrossOver(const uint8_t *Data1, size_t Size1, const uint8_t
> *Data2,
>                     size_t Size2, uint8_t *Out, size_t MaxOutSize);
>
> -  void AddWordToManualDictionary(const Unit &Word);
> +  void AddWordToManualDictionary(const Word &W);
>
> -  void AddWordToAutoDictionary(const Unit &Word, size_t PositionHint);
> +  void AddWordToAutoDictionary(const Word &W, size_t PositionHint);
>    void ClearAutoDictionary();
>    void PrintRecommendedDictionary();
>
>
> Modified: llvm/trunk/lib/Fuzzer/FuzzerMutate.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Fuzzer/FuzzerMutate.cpp?rev=257985&r1=257984&r2=257985&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Fuzzer/FuzzerMutate.cpp (original)
> +++ llvm/trunk/lib/Fuzzer/FuzzerMutate.cpp Fri Jan 15 21:53:32 2016
> @@ -23,16 +23,15 @@ struct Mutator {
>  };
>
>  struct DictionaryEntry {
> -  Unit Word;
> +  Word W;
>    size_t PositionHint;
>  };
>
>  struct Dictionary : public std::vector<DictionaryEntry>{
> -  bool ContainsWord(const Unit &W) const {
> +  bool ContainsWord(const Word &W) const {
>      return end() !=
> -           std::find_if(begin(), end(), [&](const DictionaryEntry &DE) {
> -             return DE.Word == W;
> -           });
> +           std::find_if(begin(), end(),
> +                        [&](const DictionaryEntry &DE) { return DE.W ==
> W; });
>

This ^ looks like std::any_of, perhaps? (maybe even llvm::any_of, for the
ultimate convenience)


>    }
>  };
>
> @@ -161,20 +160,20 @@ size_t MutationDispatcher::Impl::AddWord
>      size_t MaxSize) {
>    if (D.empty()) return 0;
>    const DictionaryEntry &DE = D[Rand(D.size())];
> -  const Unit &Word = DE.Word;
> +  const Word &W = DE.W;
>    size_t PositionHint = DE.PositionHint;
>    bool UsePositionHint = PositionHint !=
> std::numeric_limits<size_t>::max() &&
> -                         PositionHint + Word.size() < Size &&
> Rand.RandBool();
> -  if (Rand.RandBool()) {  // Insert Word.
> -    if (Size + Word.size() > MaxSize) return 0;
> +                         PositionHint + W.size() < Size &&
> Rand.RandBool();
> +  if (Rand.RandBool()) {  // Insert W.
> +    if (Size + W.size() > MaxSize) return 0;
>      size_t Idx = UsePositionHint ? PositionHint : Rand(Size + 1);
> -    memmove(Data + Idx + Word.size(), Data + Idx, Size - Idx);
> -    memcpy(Data + Idx, Word.data(), Word.size());
> -    Size += Word.size();
> -  } else {  // Overwrite some bytes with Word.
> -    if (Word.size() > Size) return 0;
> -    size_t Idx = UsePositionHint ? PositionHint : Rand(Size -
> Word.size());
> -    memcpy(Data + Idx, Word.data(), Word.size());
> +    memmove(Data + Idx + W.size(), Data + Idx, Size - Idx);
> +    memcpy(Data + Idx, W.data(), W.size());
> +    Size += W.size();
> +  } else {  // Overwrite some bytes with W.
> +    if (W.size() > Size) return 0;
> +    size_t Idx = UsePositionHint ? PositionHint : Rand(Size - W.size());
> +    memcpy(Data + Idx, W.data(), W.size());
>    }
>    CurrentDictionaryEntrySequence.push_back(DE);
>    return Size;
> @@ -238,16 +237,16 @@ void MutationDispatcher::StartMutationSe
>  void MutationDispatcher::RecordSuccessfulMutationSequence() {
>    for (auto &DE : MDImpl->CurrentDictionaryEntrySequence)
>      // Linear search is fine here as this happens seldom.
> -    if (!MDImpl->PersistentAutoDictionary.ContainsWord(DE.Word))
> +    if (!MDImpl->PersistentAutoDictionary.ContainsWord(DE.W))
>        MDImpl->PersistentAutoDictionary.push_back(
> -          {DE.Word, std::numeric_limits<size_t>::max()});
> +          {DE.W, std::numeric_limits<size_t>::max()});
>  }
>
>  void MutationDispatcher::PrintRecommendedDictionary() {
> -  std::vector<Unit> V;
> +  std::vector<Word> V;
>    for (auto &DE : MDImpl->PersistentAutoDictionary)
> -    if (!MDImpl->ManualDictionary.ContainsWord(DE.Word))
> -      V.push_back(DE.Word);
> +    if (!MDImpl->ManualDictionary.ContainsWord(DE.W))
> +      V.push_back(DE.W);
>    if (V.empty()) return;
>    Printf("###### Recommended dictionary. ######\n");
>    for (auto &U: V) {
> @@ -265,7 +264,7 @@ void MutationDispatcher::PrintMutationSe
>      Printf(" DE: ");
>      for (auto &DE : MDImpl->CurrentDictionaryEntrySequence) {
>        Printf("\"");
> -      PrintASCII(DE.Word, "\"-");
> +      PrintASCII(DE.W, "\"-");
>      }
>    }
>  }
> @@ -299,16 +298,16 @@ void MutationDispatcher::SetCorpus(const
>    MDImpl->SetCorpus(Corpus);
>  }
>
> -void MutationDispatcher::AddWordToManualDictionary(const Unit &Word) {
> +void MutationDispatcher::AddWordToManualDictionary(const Word &W) {
>    MDImpl->ManualDictionary.push_back(
> -      {Word, std::numeric_limits<size_t>::max()});
> +      {W, std::numeric_limits<size_t>::max()});
>  }
>
> -void MutationDispatcher::AddWordToAutoDictionary(const Unit &Word,
> +void MutationDispatcher::AddWordToAutoDictionary(const Word &W,
>                                                   size_t PositionHint) {
>    static const size_t kMaxAutoDictSize = 1 << 14;
>    if (MDImpl->TempAutoDictionary.size() >= kMaxAutoDictSize) return;
> -  MDImpl->TempAutoDictionary.push_back({Word, PositionHint});
> +  MDImpl->TempAutoDictionary.push_back({W, PositionHint});
>  }
>
>  void MutationDispatcher::ClearAutoDictionary() {
>
> Modified: llvm/trunk/lib/Fuzzer/FuzzerTraceState.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Fuzzer/FuzzerTraceState.cpp?rev=257985&r1=257984&r2=257985&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Fuzzer/FuzzerTraceState.cpp (original)
> +++ llvm/trunk/lib/Fuzzer/FuzzerTraceState.cpp Fri Jan 15 21:53:32 2016
> @@ -164,14 +164,10 @@ struct LabelRange {
>
>  // For now, very simple: put Size bytes of Data at position Pos.
>  struct TraceBasedMutation {
> -  static const size_t kMaxSize = 28;
> -  uint32_t Pos : 24;
> -  uint32_t Size : 8;
> -  uint8_t  Data[kMaxSize];
> +  uint32_t Pos;
> +  Word W;
>  };
>
> -const size_t TraceBasedMutation::kMaxSize;
> -
>  // Declared as static globals for faster checks inside the hooks.
>  static bool RecordingTraces = false;
>  static bool RecordingMemcmp = false;
> @@ -223,12 +219,11 @@ class TraceState {
>      RecordingMemcmp = false;
>      for (size_t i = 0; i < NumMutations; i++) {
>        auto &M = Mutations[i];
> -      Unit U(M.Data, M.Data + M.Size);
>        if (Options.Verbosity >= 2) {
> -        AutoDictUnitCounts[U]++;
> +        AutoDictUnitCounts[M.W]++;
>          AutoDictAdds++;
>          if ((AutoDictAdds & (AutoDictAdds - 1)) == 0) {
> -          typedef std::pair<size_t, Unit> CU;
> +          typedef std::pair<size_t, Word> CU;
>            std::vector<CU> CountedUnits;
>            for (auto &I : AutoDictUnitCounts)
>              CountedUnits.push_back(std::make_pair(I.second, I.first));
> @@ -242,17 +237,15 @@ class TraceState {
>            }
>          }
>        }
> -      USF.GetMD().AddWordToAutoDictionary(U, M.Pos);
> +      USF.GetMD().AddWordToAutoDictionary(M.W, M.Pos);
>      }
>    }
>
>    void AddMutation(uint32_t Pos, uint32_t Size, const uint8_t *Data) {
>      if (NumMutations >= kMaxMutations) return;
> -    assert(Size <= TraceBasedMutation::kMaxSize);
>      auto &M = Mutations[NumMutations++];
>      M.Pos = Pos;
> -    M.Size = Size;
> -    memcpy(M.Data, Data, Size);
> +    M.W.Set(Data, Size);
>    }
>
>    void AddMutation(uint32_t Pos, uint32_t Size, uint64_t Data) {
> @@ -274,7 +267,7 @@ class TraceState {
>    const Fuzzer::FuzzingOptions &Options;
>    uint8_t **CurrentUnitData;
>    size_t *CurrentUnitSize;
> -  std::map<Unit, size_t> AutoDictUnitCounts;
> +  std::map<Word, size_t> AutoDictUnitCounts;
>    size_t AutoDictAdds = 0;
>    static thread_local bool IsMyThread;
>  };
> @@ -423,7 +416,7 @@ void TraceState::TraceCmpCallback(uintpt
>  void TraceState::TraceMemcmpCallback(size_t CmpSize, const uint8_t *Data1,
>                                       const uint8_t *Data2) {
>    if (!RecordingMemcmp || !IsMyThread) return;
> -  CmpSize = std::min(CmpSize, TraceBasedMutation::kMaxSize);
> +  CmpSize = std::min(CmpSize, Word::GetMaxSize());
>    int Added2 = TryToAddDesiredData(Data1, Data2, CmpSize);
>    int Added1 = TryToAddDesiredData(Data2, Data1, CmpSize);
>    if ((Added1 || Added2) && Options.Verbosity >= 3) {
>
> Modified: llvm/trunk/lib/Fuzzer/FuzzerUtil.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Fuzzer/FuzzerUtil.cpp?rev=257985&r1=257984&r2=257985&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Fuzzer/FuzzerUtil.cpp (original)
> +++ llvm/trunk/lib/Fuzzer/FuzzerUtil.cpp Fri Jan 15 21:53:32 2016
> @@ -49,10 +49,12 @@ void PrintASCII(const uint8_t *Data, siz
>    Printf("%s", PrintAfter);
>  }
>
> +void PrintASCII(const Word &W, const char *PrintAfter) {
> +  PrintASCII(W.data(), W.size(), PrintAfter);
> +}
> +
>  void PrintASCII(const Unit &U, const char *PrintAfter) {
> -  for (auto X : U)
> -    PrintASCIIByte(X);
> -  Printf("%s", PrintAfter);
> +  PrintASCII(U.data(), U.size(), PrintAfter);
>  }
>
>  std::string Hash(const Unit &U) {
>
> Modified: llvm/trunk/lib/Fuzzer/test/FuzzerUnittest.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Fuzzer/test/FuzzerUnittest.cpp?rev=257985&r1=257984&r2=257985&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Fuzzer/test/FuzzerUnittest.cpp (original)
> +++ llvm/trunk/lib/Fuzzer/test/FuzzerUnittest.cpp Fri Jan 15 21:53:32 2016
> @@ -247,8 +247,8 @@ void TestAddWordFromDictionary(Mutator M
>    MutationDispatcher MD(Rand);
>    uint8_t Word1[4] = {0xAA, 0xBB, 0xCC, 0xDD};
>    uint8_t Word2[3] = {0xFF, 0xEE, 0xEF};
> -  MD.AddWordToManualDictionary(Unit(Word1, Word1 + sizeof(Word1)));
> -  MD.AddWordToManualDictionary(Unit(Word2, Word2 + sizeof(Word2)));
> +  MD.AddWordToManualDictionary(Word(Word1, sizeof(Word1)));
> +  MD.AddWordToManualDictionary(Word(Word2, sizeof(Word2)));
>    int FoundMask = 0;
>    uint8_t CH0[7] = {0x00, 0x11, 0x22, 0xAA, 0xBB, 0xCC, 0xDD};
>    uint8_t CH1[7] = {0x00, 0x11, 0xAA, 0xBB, 0xCC, 0xDD, 0x22};
> @@ -285,16 +285,16 @@ TEST(FuzzerMutate, AddWordFromDictionary
>  void TestAddWordFromDictionaryWithHint(Mutator M, int NumIter) {
>    FuzzerRandomLibc Rand(0);
>    MutationDispatcher MD(Rand);
> -  uint8_t Word[] = {0xAA, 0xBB, 0xCC, 0xDD, 0xFF, 0xEE, 0xEF};
> +  uint8_t W[] = {0xAA, 0xBB, 0xCC, 0xDD, 0xFF, 0xEE, 0xEF};
>    size_t PosHint = 7777;
> -  MD.AddWordToAutoDictionary(Unit(Word, Word + sizeof(Word)), PosHint);
> +  MD.AddWordToAutoDictionary(Word(W, sizeof(W)), PosHint);
>    int FoundMask = 0;
>    for (int i = 0; i < NumIter; i++) {
>      uint8_t T[10000];
>      memset(T, 0, sizeof(T));
>      size_t NewSize = (MD.*M)(T, 9000, 10000);
> -    if (NewSize >= PosHint + sizeof(Word) &&
> -        !memcmp(Word, T + PosHint, sizeof(Word)))
> +    if (NewSize >= PosHint + sizeof(W) &&
> +        !memcmp(W, T + PosHint, sizeof(W)))
>        FoundMask = 1;
>    }
>    EXPECT_EQ(FoundMask, 1);
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160115/7b702f76/attachment.html>


More information about the llvm-commits mailing list