Merge pull request #2519 from sesse/master

Speed up parsing

src/build_log.cc

@@ -53,63 +53,14 @@ using namespace std;
 namespace {
 
 const char kFileSignature[] = "# ninja log v%d\n";
-const int kOldestSupportedVersion = 6;
-const int kCurrentVersion = 6;
-
-// 64bit MurmurHash2, by Austin Appleby
-#if defined(_MSC_VER)
-#define BIG_CONSTANT(x) (x)
-#else   // defined(_MSC_VER)
-#define BIG_CONSTANT(x) (x##LLU)
-#endif // !defined(_MSC_VER)
-inline
-uint64_t MurmurHash64A(const void* key, size_t len) {
-  static const uint64_t seed = 0xDECAFBADDECAFBADull;
-  const uint64_t m = BIG_CONSTANT(0xc6a4a7935bd1e995);
-  const int r = 47;
-  uint64_t h = seed ^ (len * m);
-  const unsigned char* data = static_cast<const unsigned char*>(key);
-  while (len >= 8) {
-    uint64_t k;
-    memcpy(&k, data, sizeof k);
-    k *= m;
-    k ^= k >> r;
-    k *= m;
-    h ^= k;
-    h *= m;
-    data += 8;
-    len -= 8;
-  }
-  switch (len & 7)
-  {
-  case 7: h ^= uint64_t(data[6]) << 48;
-          NINJA_FALLTHROUGH;
-  case 6: h ^= uint64_t(data[5]) << 40;
-          NINJA_FALLTHROUGH;
-  case 5: h ^= uint64_t(data[4]) << 32;
-          NINJA_FALLTHROUGH;
-  case 4: h ^= uint64_t(data[3]) << 24;
-          NINJA_FALLTHROUGH;
-  case 3: h ^= uint64_t(data[2]) << 16;
-          NINJA_FALLTHROUGH;
-  case 2: h ^= uint64_t(data[1]) << 8;
-          NINJA_FALLTHROUGH;
-  case 1: h ^= uint64_t(data[0]);
-          h *= m;
-  };
-  h ^= h >> r;
-  h *= m;
-  h ^= h >> r;
-  return h;
-}
-#undef BIG_CONSTANT
-
+const int kOldestSupportedVersion = 7;
+const int kCurrentVersion = 7;
 
 }  // namespace
 
 // static
 uint64_t BuildLog::LogEntry::HashCommand(StringPiece command) {
-  return MurmurHash64A(command.str_, command.len_);
+  return rapidhash(command.str_, command.len_);
 }
 
 BuildLog::LogEntry::LogEntry(const string& output)

src/build_log_test.cc

@@ -104,7 +104,7 @@ TEST_F(BuildLogTest, FirstWriteAddsSignature) {
 
 TEST_F(BuildLogTest, DoubleEntry) {
   FILE* f = fopen(kTestFilename, "wb");
-  fprintf(f, "# ninja log v6\n");
+  fprintf(f, "# ninja log v7\n");
   fprintf(f, "0\t1\t2\tout\t%" PRIx64 "\n",
       BuildLog::LogEntry::HashCommand("command abc"));
   fprintf(f, "0\t1\t2\tout\t%" PRIx64 "\n",
@@ -177,7 +177,7 @@ TEST_F(BuildLogTest, ObsoleteOldVersion) {
 
 TEST_F(BuildLogTest, SpacesInOutput) {
   FILE* f = fopen(kTestFilename, "wb");
-  fprintf(f, "# ninja log v6\n");
+  fprintf(f, "# ninja log v7\n");
   fprintf(f, "123\t456\t456\tout with space\t%" PRIx64 "\n",
       BuildLog::LogEntry::HashCommand("command"));
   fclose(f);
@@ -200,10 +200,10 @@ TEST_F(BuildLogTest, DuplicateVersionHeader) {
   // build log on Windows. This shouldn't crash, and the second version header
   // should be ignored.
   FILE* f = fopen(kTestFilename, "wb");
-  fprintf(f, "# ninja log v6\n");
+  fprintf(f, "# ninja log v7\n");
   fprintf(f, "123\t456\t456\tout\t%" PRIx64 "\n",
       BuildLog::LogEntry::HashCommand("command"));
-  fprintf(f, "# ninja log v6\n");
+  fprintf(f, "# ninja log v7\n");
   fprintf(f, "456\t789\t789\tout2\t%" PRIx64 "\n",
       BuildLog::LogEntry::HashCommand("command2"));
   fclose(f);
@@ -252,7 +252,7 @@ struct TestDiskInterface : public DiskInterface {
 
 TEST_F(BuildLogTest, Restat) {
   FILE* f = fopen(kTestFilename, "wb");
-  fprintf(f, "# ninja log v6\n"
+  fprintf(f, "# ninja log v7\n"
              "1\t2\t3\tout\tcommand\n");
   fclose(f);
   std::string err;
@@ -280,7 +280,7 @@ TEST_F(BuildLogTest, VeryLongInputLine) {
   // Ninja's build log buffer is currently 256kB. Lines longer than that are
   // silently ignored, but don't affect parsing of other lines.
   FILE* f = fopen(kTestFilename, "wb");
-  fprintf(f, "# ninja log v6\n");
+  fprintf(f, "# ninja log v7\n");
   fprintf(f, "123\t456\t456\tout\tcommand start");
   for (size_t i = 0; i < (512 << 10) / strlen(" more_command"); ++i)
     fputs(" more_command", f);

src/deps_log.cc

@@ -185,15 +185,13 @@ LoadStatus DepsLog::Load(const string& path, State* state, string* err) {
     return LOAD_SUCCESS;
   }
 
-  long offset;
+  long offset = ftell(f);
   bool read_failed = false;
   int unique_dep_record_count = 0;
   int total_dep_record_count = 0;
   for (;;) {
-    offset = ftell(f);
-
     unsigned size;
-    if (fread(&size, 4, 1, f) < 1) {
+    if (fread(&size, sizeof(size), 1, f) < 1) {
       if (!feof(f))
         read_failed = true;
       break;
@@ -205,6 +203,7 @@ LoadStatus DepsLog::Load(const string& path, State* state, string* err) {
       read_failed = true;
       break;
     }
+    offset += size + sizeof(size);
 
     if (is_deps) {
       if ((size % 4) != 0) {

src/eval_env.cc

@@ -99,6 +99,10 @@ string BindingEnv::LookupWithFallback(const string& var,
 }
 
 string EvalString::Evaluate(Env* env) const {
+  if (parsed_.empty()) {
+    return single_token_;
+  }
+
   string result;
   for (TokenList::const_iterator i = parsed_.begin(); i != parsed_.end(); ++i) {
     if (i->second == RAW)
@@ -110,40 +114,57 @@ string EvalString::Evaluate(Env* env) const {
 }
 
 void EvalString::AddText(StringPiece text) {
-  // Add it to the end of an existing RAW token if possible.
-  if (!parsed_.empty() && parsed_.back().second == RAW) {
-    parsed_.back().first.append(text.str_, text.len_);
+  if (parsed_.empty()) {
+    single_token_.append(text.begin(), text.end());
+  } else if (!parsed_.empty() && parsed_.back().second == RAW) {
+    parsed_.back().first.append(text.begin(), text.end());
   } else {
-    parsed_.push_back(make_pair(text.AsString(), RAW));
+    parsed_.push_back(std::make_pair(text.AsString(), RAW));
   }
 }
+
 void EvalString::AddSpecial(StringPiece text) {
-  parsed_.push_back(make_pair(text.AsString(), SPECIAL));
+  if (parsed_.empty() && !single_token_.empty()) {
+    // Going from one to two tokens, so we can no longer apply
+    // our single_token_ optimization and need to push everything
+    // onto the vector.
+    parsed_.push_back(std::make_pair(std::move(single_token_), RAW));
+  }
+  parsed_.push_back(std::make_pair(text.AsString(), SPECIAL));
 }
 
 string EvalString::Serialize() const {
   string result;
-  for (TokenList::const_iterator i = parsed_.begin();
-       i != parsed_.end(); ++i) {
+  if (parsed_.empty() && !single_token_.empty()) {
     result.append("[");
-    if (i->second == SPECIAL)
-      result.append("$");
-    result.append(i->first);
+    result.append(single_token_);
     result.append("]");
+  } else {
+    for (const auto& pair : parsed_) {
+      result.append("[");
+      if (pair.second == SPECIAL)
+        result.append("$");
+      result.append(pair.first.begin(), pair.first.end());
+      result.append("]");
+    }
   }
   return result;
 }
 
 string EvalString::Unparse() const {
   string result;
-  for (TokenList::const_iterator i = parsed_.begin();
-       i != parsed_.end(); ++i) {
-    bool special = (i->second == SPECIAL);
-    if (special)
-      result.append("${");
-    result.append(i->first);
-    if (special)
-      result.append("}");
+  if (parsed_.empty() && !single_token_.empty()) {
+    result.append(single_token_.begin(), single_token_.end());
+  } else {
+    for (TokenList::const_iterator i = parsed_.begin();
+         i != parsed_.end(); ++i) {
+      bool special = (i->second == SPECIAL);
+      if (special)
+        result.append("${");
+      result.append(i->first.begin(), i->first.end());
+      if (special)
+        result.append("}");
+    }
   }
   return result;
 }

src/eval_env.h

@@ -39,8 +39,8 @@ struct EvalString {
   /// @return The string with variables not expanded.
   std::string Unparse() const;
 
-  void Clear() { parsed_.clear(); }
-  bool empty() const { return parsed_.empty(); }
+  void Clear() { parsed_.clear(); single_token_.clear(); }
+  bool empty() const { return parsed_.empty() && single_token_.empty(); }
 
   void AddText(StringPiece text);
   void AddSpecial(StringPiece text);
@@ -53,6 +53,12 @@ struct EvalString {
   enum TokenType { RAW, SPECIAL };
   typedef std::vector<std::pair<std::string, TokenType> > TokenList;
   TokenList parsed_;
+
+  // If we hold only a single RAW token, then we keep it here instead of
+  // pushing it on TokenList. This saves a bunch of allocations for
+  // what is a common case. If parsed_ is nonempty, then this value
+  // must be ignored.
+  std::string single_token_;
 };
 
 /// An invocable build command and associated metadata (description, etc.).

src/graph.cc

@@ -740,12 +740,13 @@ bool ImplicitDepLoader::LoadDepsFromLog(Edge* edge, string* err) {
     return false;
   }
 
-  vector<Node*>::iterator implicit_dep =
-      PreallocateSpace(edge, deps->node_count);
-  for (int i = 0; i < deps->node_count; ++i, ++implicit_dep) {
-    Node* node = deps->nodes[i];
-    *implicit_dep = node;
-    node->AddOutEdge(edge);
+  Node** nodes = deps->nodes;
+  size_t node_count = deps->node_count;
+  edge->inputs_.insert(edge->inputs_.end() - edge->order_only_deps_,
+                       nodes, nodes + node_count);
+  edge->implicit_deps_ += node_count;
+  for (size_t i = 0; i < node_count; ++i) {
+    nodes[i]->AddOutEdge(edge);
   }
   return true;
 }

src/hash_map.h

@@ -20,40 +20,8 @@
 #include "string_piece.h"
 #include "util.h"
 
-// MurmurHash2, by Austin Appleby
-static inline
-unsigned int MurmurHash2(const void* key, size_t len) {
-  static const unsigned int seed = 0xDECAFBAD;
-  const unsigned int m = 0x5bd1e995;
-  const int r = 24;
-  unsigned int h = seed ^ len;
-  const unsigned char* data = static_cast<const unsigned char*>(key);
-  while (len >= 4) {
-    unsigned int k;
-    memcpy(&k, data, sizeof k);
-    k *= m;
-    k ^= k >> r;
-    k *= m;
-    h *= m;
-    h ^= k;
-    data += 4;
-    len -= 4;
-  }
-  switch (len) {
-  case 3: h ^= data[2] << 16;
-          NINJA_FALLTHROUGH;
-  case 2: h ^= data[1] << 8;
-          NINJA_FALLTHROUGH;
-  case 1: h ^= data[0];
-    h *= m;
-  };
-  h ^= h >> 13;
-  h *= m;
-  h ^= h >> 15;
-  return h;
-}
-
-#include <unordered_map>
+#include "third_party/emhash/hash_table8.hpp"
+#include "third_party/rapidhash/rapidhash.h"
 
 namespace std {
 template<>
@@ -62,7 +30,7 @@ struct hash<StringPiece> {
   typedef size_t result_type;
 
   size_t operator()(StringPiece key) const {
-    return MurmurHash2(key.str_, key.len_);
+    return rapidhash(key.str_, key.len_);
   }
 };
 }
@@ -73,7 +41,7 @@ struct hash<StringPiece> {
 /// mapping StringPiece => Foo*.
 template<typename V>
 struct ExternalStringHashMap {
-  typedef std::unordered_map<StringPiece, V> Type;
+  typedef emhash8::HashMap<StringPiece, V> Type;
 };
 
 #endif // NINJA_MAP_H_