use hte new pred cache to speed up the new non-local memdep

queries.  This speeds up GVN using the new queries (not yet
checked in) by just over 10%.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@60743 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 31 insertions, 14 deletions

diff --git a/include/llvm/Analysis/MemoryDependenceAnalysis.h b/include/llvm/Analysis/MemoryDependenceAnalysis.h
index f46a286..da82d7a 100644
--- a/include/llvm/Analysis/MemoryDependenceAnalysis.h
+++ b/include/llvm/Analysis/MemoryDependenceAnalysis.h
@@ -18,6 +18,7 @@
 #include "llvm/Pass.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/OwningPtr.h"
 #include "llvm/ADT/PointerIntPair.h"
 
 namespace llvm {
@@ -28,6 +29,7 @@ namespace llvm {
   class AliasAnalysis;
   class TargetData;
   class MemoryDependenceAnalysis;
+  class PredIteratorCache;
   
   /// MemDepResult - A memory dependence query can return one of three different
   /// answers, described below.
@@ -193,23 +195,18 @@ namespace llvm {
     /// Current AA implementation, just a cache.
     AliasAnalysis *AA;
     TargetData *TD;
+    OwningPtr<PredIteratorCache> PredCache;
   public:
-    MemoryDependenceAnalysis() : FunctionPass(&ID) {}
+    MemoryDependenceAnalysis();
+    ~MemoryDependenceAnalysis();
     static char ID;
 
     /// Pass Implementation stuff.  This doesn't do any analysis eagerly.
     bool runOnFunction(Function &);
     
     /// Clean up memory in between runs
-    void releaseMemory() {
-      LocalDeps.clear();
-      NonLocalDeps.clear();
-      NonLocalPointerDeps.clear();
-      ReverseLocalDeps.clear();
-      ReverseNonLocalDeps.clear();
-      ReverseNonLocalPtrDeps.clear();
-    }
-
+    void releaseMemory();
+    
     /// getAnalysisUsage - Does not modify anything.  It uses Value Numbering
     /// and Alias Analysis.
     ///
diff --git a/lib/Analysis/MemoryDependenceAnalysis.cpp b/lib/Analysis/MemoryDependenceAnalysis.cpp
index c42633b..9439502 100644
--- a/lib/Analysis/MemoryDependenceAnalysis.cpp
+++ b/lib/Analysis/MemoryDependenceAnalysis.cpp
@@ -21,7 +21,7 @@
 #include "llvm/Function.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/ADT/Statistic.h"
-#include "llvm/Support/CFG.h"
+#include "llvm/Support/PredIteratorCache.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Target/TargetData.h"
 using namespace llvm;
@@ -45,6 +45,25 @@ char MemoryDependenceAnalysis::ID = 0;
 static RegisterPass<MemoryDependenceAnalysis> X("memdep",
                                      "Memory Dependence Analysis", false, true);
 
+MemoryDependenceAnalysis::MemoryDependenceAnalysis()
+: FunctionPass(&ID), PredCache(0) {
+}
+MemoryDependenceAnalysis::~MemoryDependenceAnalysis() {
+}
+
+/// Clean up memory in between runs
+void MemoryDependenceAnalysis::releaseMemory() {
+  LocalDeps.clear();
+  NonLocalDeps.clear();
+  NonLocalPointerDeps.clear();
+  ReverseLocalDeps.clear();
+  ReverseNonLocalDeps.clear();
+  ReverseNonLocalPtrDeps.clear();
+  PredCache->clear();
+}
+
+
+
 /// getAnalysisUsage - Does not modify anything.  It uses Alias Analysis.
 ///
 void MemoryDependenceAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
@@ -56,6 +75,8 @@ void MemoryDependenceAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
 bool MemoryDependenceAnalysis::runOnFunction(Function &) {
   AA = &getAnalysis<AliasAnalysis>();
   TD = &getAnalysis<TargetData>();
+  if (PredCache == 0)
+    PredCache.reset(new PredIteratorCache());
   return false;
 }
 
@@ -468,8 +489,7 @@ getNonLocalPointerDependency(Value *Pointer, bool isLoad, BasicBlock *FromBB,
   // While we have blocks to analyze, get their values.
   SmallPtrSet<BasicBlock*, 64> Visited;
   
-  for (pred_iterator PI = pred_begin(FromBB), E = pred_end(FromBB); PI != E;
-       ++PI) {
+  for (BasicBlock **PI = PredCache->GetPreds(FromBB); *PI; ++PI) {
     // TODO: PHI TRANSLATE.
     getNonLocalPointerDepInternal(Pointer, PointeeSize, isLoad, *PI,
                                   Result, Visited);
@@ -592,7 +612,7 @@ getNonLocalPointerDepInternal(Value *Pointer, uint64_t PointeeSize,
     
     // Otherwise, we have to process all the predecessors of this block to scan
     // them as well.
-    for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
+    for (BasicBlock **PI = PredCache->GetPreds(BB); *PI; ++PI) {
       // TODO: PHI TRANSLATE.
       Worklist.push_back(*PI);
     }