[llvm-commits] CVS: llvm/lib/Bytecode/Reader/Reader.cpp Reader.h

[Chris Lattner]

Changes in directory llvm/lib/Bytecode/Reader:

Reader.cpp updated: 1.166 -> 1.167
Reader.h updated: 1.24 -> 1.25
---
Log message:

Use a map to cache the ModuleType information, so we can do logarithmic
lookups instead of linear time lookups.  This speeds up bc parsing of a
large file from

137.834u 118.256s 4:27.96
to
132.611u 114.436s 4:08.53

with a release build.


---
Diffs of the changes:  (+47 -8)

 Reader.cpp |   49 +++++++++++++++++++++++++++++++++++++++++--------
 Reader.h   |    6 ++++++
 2 files changed, 47 insertions(+), 8 deletions(-)


Index: llvm/lib/Bytecode/Reader/Reader.cpp
diff -u llvm/lib/Bytecode/Reader/Reader.cpp:1.166 llvm/lib/Bytecode/Reader/Reader.cpp:1.167
--- llvm/lib/Bytecode/Reader/Reader.cpp:1.166	Wed Aug 17 14:22:05 2005
+++ llvm/lib/Bytecode/Reader/Reader.cpp	Mon Oct  3 16:26:53 2005
@@ -347,11 +347,25 @@
     return Type::FirstDerivedTyID + ModuleTypes.size() +
            (&*I - &FunctionTypes[0]);
 
-  // Check the module level types now...
-  I = std::find(ModuleTypes.begin(), ModuleTypes.end(), Ty);
-  if (I == ModuleTypes.end())
+  // If we don't have our cache yet, build it now.
+  if (ModuleTypeIDCache.empty()) {
+    unsigned N = 0;
+    ModuleTypeIDCache.reserve(ModuleTypes.size());
+    for (TypeListTy::iterator I = ModuleTypes.begin(), E = ModuleTypes.end();
+         I != E; ++I, ++N)
+      ModuleTypeIDCache.push_back(std::make_pair(*I, N));
+    
+    std::sort(ModuleTypeIDCache.begin(), ModuleTypeIDCache.end());
+  }
+  
+  // Binary search the cache for the entry.
+  std::vector<std::pair<const Type*, unsigned> >::iterator IT =
+    std::lower_bound(ModuleTypeIDCache.begin(), ModuleTypeIDCache.end(),
+                     std::make_pair(Ty, 0U));
+  if (IT == ModuleTypeIDCache.end() || IT->first != Ty)
     error("Didn't find type in ModuleTypes.");
-  return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
+    
+  return Type::FirstDerivedTyID + IT->second;
 }
 
 /// This is just like getType, but when a compaction table is in use, it is
@@ -375,11 +389,26 @@
 unsigned BytecodeReader::getGlobalTableTypeSlot(const Type *Ty) {
   if (Ty->isPrimitiveType())
     return Ty->getTypeID();
-  TypeListTy::iterator I = std::find(ModuleTypes.begin(),
-                                      ModuleTypes.end(), Ty);
-  if (I == ModuleTypes.end())
+  
+  // If we don't have our cache yet, build it now.
+  if (ModuleTypeIDCache.empty()) {
+    unsigned N = 0;
+    ModuleTypeIDCache.reserve(ModuleTypes.size());
+    for (TypeListTy::iterator I = ModuleTypes.begin(), E = ModuleTypes.end();
+         I != E; ++I, ++N)
+      ModuleTypeIDCache.push_back(std::make_pair(*I, N));
+    
+    std::sort(ModuleTypeIDCache.begin(), ModuleTypeIDCache.end());
+  }
+  
+  // Binary search the cache for the entry.
+  std::vector<std::pair<const Type*, unsigned> >::iterator IT =
+    std::lower_bound(ModuleTypeIDCache.begin(), ModuleTypeIDCache.end(),
+                     std::make_pair(Ty, 0U));
+  if (IT == ModuleTypeIDCache.end() || IT->first != Ty)
     error("Didn't find type in ModuleTypes.");
-  return Type::FirstDerivedTyID + (&*I - &ModuleTypes[0]);
+  
+  return Type::FirstDerivedTyID + IT->second;
 }
 
 /// Retrieve a value of a given type and slot number, possibly creating
@@ -1309,6 +1338,10 @@
   if (Handler)
     Handler->handleTypeList(NumEntries);
 
+  // If we are about to resolve types, make sure the type cache is clear.
+  if (NumEntries)
+    ModuleTypeIDCache.clear();
+  
   // Loop through reading all of the types.  Forward types will make use of the
   // opaque types just inserted.
   //


Index: llvm/lib/Bytecode/Reader/Reader.h
diff -u llvm/lib/Bytecode/Reader/Reader.h:1.24 llvm/lib/Bytecode/Reader/Reader.h:1.25
--- llvm/lib/Bytecode/Reader/Reader.h:1.24	Tue Aug 16 16:59:52 2005
+++ llvm/lib/Bytecode/Reader/Reader.h	Mon Oct  3 16:26:53 2005
@@ -333,6 +333,12 @@
   /// @brief This vector is used to deal with forward references to types in
   /// a module.
   TypeListTy ModuleTypes;
+  
+  /// @brief This is an inverse mapping of ModuleTypes from the type to an
+  /// index.  Because refining types causes the index of this map to be
+  /// invalidated, any time we refine a type, we clear this cache and recompute
+  /// it next time we need it.  These entries are ordered by the pointer value.
+  std::vector<std::pair<const Type*, unsigned> > ModuleTypeIDCache;
 
   /// @brief This vector is used to deal with forward references to types in
   /// a function.