Introduce a caching mechanism for macro expanded tokens.

Previously macro expanded tokens were added to Preprocessor's bump allocator and never released,
even after the TokenLexer that were lexing them was finished, thus they were wasting memory.
A very "useful" boost library was causing clang to eat 1 GB just for the expanded macro tokens.

Introduce a special cache that works like a stack; a TokenLexer can add the macro expanded tokens
in the cache, and when it finishes, the tokens are removed from the end of the cache.

Now consumed memory by expanded tokens for that library is ~ 1.5 MB.

Part of rdar://9327049.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@134105 91177308-0d34-0410-b5e6-96231b3b80d8

include/clang/Lex/Preprocessor.h

@@ -29,6 +29,7 @@
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/OwningPtr.h"
 #include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/ArrayRef.h"
 #include "llvm/Support/Allocator.h"
 #include <vector>
 
@@ -240,6 +241,14 @@ class Preprocessor : public llvm::RefCountedBase<Preprocessor> {
   unsigned NumCachedTokenLexers;
   TokenLexer *TokenLexerCache[TokenLexerCacheSize];
 
+  /// \brief Keeps macro expanded tokens for TokenLexers.
+  //
+  /// Works like a stack; a TokenLexer adds the macro expanded tokens that is
+  /// going to lex in the cache and when it finishes the tokens are removed
+  /// from the end of the cache.
+  llvm::SmallVector<Token, 16> MacroExpandedTokens;
+  std::vector<std::pair<TokenLexer *, size_t> > MacroExpandingLexersStack;
+
   /// \brief A record of the macro definitions and instantiations that
   /// occurred during preprocessing. 
   ///
@@ -979,6 +988,16 @@ private:
   /// the macro should not be expanded return true, otherwise return false.
   bool HandleMacroExpandedIdentifier(Token &Tok, MacroInfo *MI);
 
+  /// \brief Cache macro expanded tokens for TokenLexers.
+  //
+  /// Works like a stack; a TokenLexer adds the macro expanded tokens that is
+  /// going to lex in the cache and when it finishes the tokens are removed
+  /// from the end of the cache.
+  Token *cacheMacroExpandedTokens(TokenLexer *tokLexer,
+                                  llvm::ArrayRef<Token> tokens);
+  void removeCachedMacroExpandedTokensOfLastLexer();
+  friend void TokenLexer::ExpandFunctionArguments();
+
   /// isNextPPTokenLParen - Determine whether the next preprocessor token to be
   /// lexed is a '('.  If so, consume the token and return true, if not, this
   /// method should have no observable side-effect on the lexed tokens.

include/clang/Lex/TokenLexer.h

@@ -43,10 +43,13 @@ class TokenLexer {
   /// Tokens - This is the pointer to an array of tokens that the macro is
   /// defined to, with arguments expanded for function-like macros.  If this is
   /// a token stream, these are the tokens we are returning.  This points into
-  /// the macro definition we are lexing from, a scratch buffer allocated from
-  /// the preprocessor's bump pointer allocator, or some other buffer that we
-  /// may or may not own (depending on OwnsTokens).
+  /// the macro definition we are lexing from, a cache buffer that is owned by
+  /// the preprocessor, or some other buffer that we may or may not own
+  /// (depending on OwnsTokens).
+  /// Note that if it points into Preprocessor's cache buffer, the Preprocessor
+  /// may update the pointer as needed.
   const Token *Tokens;
+  friend class Preprocessor;
 
   /// NumTokens - This is the length of the Tokens array.
   ///

lib/Lex/PPLexerChange.cpp

@@ -265,6 +265,10 @@ bool Preprocessor::HandleEndOfTokenLexer(Token &Result) {
   assert(CurTokenLexer && !CurPPLexer &&
          "Ending a macro when currently in a #include file!");
 
+  if (!MacroExpandingLexersStack.empty() &&
+      MacroExpandingLexersStack.back().first == CurTokenLexer.get())
+    removeCachedMacroExpandedTokensOfLastLexer();
+
   // Delete or cache the now-dead macro expander.
   if (NumCachedTokenLexers == TokenLexerCacheSize)
     CurTokenLexer.reset();

lib/Lex/PPMacroExpansion.cpp

@@ -22,6 +22,7 @@
 #include "clang/Lex/CodeCompletionHandler.h"
 #include "clang/Lex/ExternalPreprocessorSource.h"
 #include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/Config/config.h"
 #include "llvm/Support/raw_ostream.h"
 #include <cstdio>
@@ -490,6 +491,46 @@ MacroArgs *Preprocessor::ReadFunctionLikeMacroArgs(Token &MacroName,
                            isVarargsElided, *this);
 }
 
+/// \brief Keeps macro expanded tokens for TokenLexers.
+//
+/// Works like a stack; a TokenLexer adds the macro expanded tokens that is
+/// going to lex in the cache and when it finishes the tokens are removed
+/// from the end of the cache.
+Token *Preprocessor::cacheMacroExpandedTokens(TokenLexer *tokLexer,
+                                              llvm::ArrayRef<Token> tokens) {
+  assert(tokLexer);
+  if (tokens.empty())
+    return 0;
+
+  size_t newIndex = MacroExpandedTokens.size();
+  bool cacheNeedsToGrow = tokens.size() >
+                      MacroExpandedTokens.capacity()-MacroExpandedTokens.size(); 
+  MacroExpandedTokens.append(tokens.begin(), tokens.end());
+
+  if (cacheNeedsToGrow) {
+    // Go through all the TokenLexers whose 'Tokens' pointer points in the
+    // buffer and update the pointers to the (potential) new buffer array.
+    for (unsigned i = 0, e = MacroExpandingLexersStack.size(); i != e; ++i) {
+      TokenLexer *prevLexer;
+      size_t tokIndex;
+      llvm::tie(prevLexer, tokIndex) = MacroExpandingLexersStack[i];
+      prevLexer->Tokens = MacroExpandedTokens.data() + tokIndex;
+    }
+  }
+
+  MacroExpandingLexersStack.push_back(std::make_pair(tokLexer, newIndex));
+  return MacroExpandedTokens.data() + newIndex;
+}
+
+void Preprocessor::removeCachedMacroExpandedTokensOfLastLexer() {
+  assert(!MacroExpandingLexersStack.empty());
+  size_t tokIndex = MacroExpandingLexersStack.back().second;
+  assert(tokIndex < MacroExpandedTokens.size());
+  // Pop the cached macro expanded tokens from the end.
+  MacroExpandedTokens.resize(tokIndex);
+  MacroExpandingLexersStack.pop_back();
+}
+
 /// ComputeDATE_TIME - Compute the current time, enter it into the specified
 /// scratch buffer, then return DATELoc/TIMELoc locations with the position of
 /// the identifier tokens inserted.

lib/Lex/Preprocessor.cpp

@@ -118,6 +118,8 @@ Preprocessor::Preprocessor(Diagnostic &diags, const LangOptions &opts,
 
 Preprocessor::~Preprocessor() {
   assert(BacktrackPositions.empty() && "EnableBacktrack/Backtrack imbalance!");
+  assert(MacroExpandingLexersStack.empty() && MacroExpandedTokens.empty() &&
+         "Preprocessor::HandleEndOfTokenLexer should have cleared those");
 
   while (!IncludeMacroStack.empty()) {
     delete IncludeMacroStack.back().TheLexer;
@@ -226,7 +228,7 @@ Preprocessor::macro_begin(bool IncludeExternalMacros) const {
 }
 
 size_t Preprocessor::getTotalMemory() const {
-  return BP.getTotalMemory();
+  return BP.getTotalMemory() + MacroExpandedTokens.capacity()*sizeof(Token);
 }
 
 Preprocessor::macro_iterator

lib/Lex/TokenLexer.cpp

@@ -284,15 +284,11 @@ void TokenLexer::ExpandFunctionArguments() {
     assert(!OwnsTokens && "This would leak if we already own the token list");
     // This is deleted in the dtor.
     NumTokens = ResultToks.size();
-    llvm::BumpPtrAllocator &Alloc = PP.getPreprocessorAllocator();
-    Token *Res =
-      static_cast<Token *>(Alloc.Allocate(sizeof(Token)*ResultToks.size(),
-                                          llvm::alignOf<Token>()));
-    if (NumTokens)
-      memcpy(Res, &ResultToks[0], NumTokens*sizeof(Token));
-    Tokens = Res;
+    // The tokens will be added to Preprocessor's cache and will be removed
+    // when this TokenLexer finishes lexing them.
+    Tokens = PP.cacheMacroExpandedTokens(this, ResultToks);
 
-    // The preprocessor bump pointer owns these tokens, not us.
+    // The preprocessor cache of macro expanded tokens owns these tokens,not us.
     OwnsTokens = false;
   }
 }