Make PredIteratorCache size() logically const. Do not require copying predecessors to get size.

[• dberlin]

Summary

Every single benchmark i can run, on large and small cfgs, fully
connected, etc, across 3 different platforms (x86, arm., and PPC) says
that the current pred iterator cache is a losing proposition.

I can't find a case where it's faster than just walking preds, and in some cases, it's 5-10% slower.

This is due to copying the preds.
It also degrades into copying the entire cfg.

The one operation that is occasionally faster is the cached size.
This makes that operation faster by not relying on having the copies available.

I'm not even sure that is faster enough to be worth it. I, again, have
trouble finding cases where this takes long enough in a pass to be
worth caching compared to a million other things they could cache or
improve.

My suggestion:
We next remove the get() interface.
We do stronger benchmarking of size().
We probably end up killing this entire cache.
/

[davide]

LGTM. I personally like your plan of removing the cache altogether supported by benchmarks.

include/llvm/IR/PredIteratorCache.h

 
 namespace llvm {
 
 /// PredIteratorCache - This class is an extremely trivial cache for
 /// predecessor iterator queries.  This is useful for code that repeatedly
 /// wants the predecessor list for the same blocks.
 class PredIteratorCache {
   /// BlockToPredsMap - Pointer to null-terminated list.
-  DenseMap<BasicBlock *, BasicBlock **> BlockToPredsMap;
-  DenseMap<BasicBlock *, unsigned> BlockToPredCountMap;
+  mutable DenseMap<BasicBlock *, BasicBlock **> BlockToPredsMap;
+  mutable DenseMap<BasicBlock *, unsigned> BlockToPredCountMap;
 
   /// Memory - This is the space that holds cached preds.
   BumpPtrAllocator Memory;
 
 private:
   /// GetPreds - Get a cached list for the null-terminated predecessor list of
   /// the specified block.  This can be used in a loop like this:
   ///   for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI)
BasicBlock **GetPreds(BasicBlock *BB) {
 
     BlockToPredCountMap[BB] = PredCache.size() - 1;
 
     Entry = Memory.Allocate<BasicBlock *>(PredCache.size());
     std::copy(PredCache.begin(), PredCache.end(), Entry);
     return Entry;
   }
 
-  unsigned GetNumPreds(BasicBlock *BB) {
-    GetPreds(BB);
-    return BlockToPredCountMap[BB];
+  unsigned GetNumPreds(BasicBlock *BB) const {
+    auto Result = BlockToPredCountMap.find(BB);
+    if (Result != BlockToPredCountMap.end())
+      return Result->second;
+    return BlockToPredCountMap[BB] = std::distance(pred_begin(BB), pred_end(BB));
   }
 
 public:
-  size_t size(BasicBlock *BB) { return GetNumPreds(BB); }
+  size_t size(BasicBlock *BB) const { return GetNumPreds(BB); }
   ArrayRef<BasicBlock *> get(BasicBlock *BB) {
     return makeArrayRef(GetPreds(BB), GetNumPreds(BB));
   }
 
   /// clear - Remove all information.
   void clear() {
     BlockToPredsMap.clear();
     BlockToPredCountMap.clear();
     Memory.Reset();
   }
 };
 
 } // end namespace llvm
 
 #endif