Enable the new LoopInfo algorithm by default.

The primary advantage is that loop optimizations will be applied in a
stable order. This helps debugging and unit test creation. It is also
a better overall implementation without pathologically bad performance
on deep functions.

On large functions (llvm-stress --size=200000 | opt -loops)
Before: 0.1263s
After:  0.0225s

On deep functions (after tweaking llvm-stress, thanks Nadav):
Before: 0.2281s
After:  0.0227s

See r158790 for more comments.

The loop tree is now consistently generated in forward order, but loop
passes are applied in reverse order over the program. If we have a
loop optimization that prefers forward order, that can easily be
achieved by adding a different type of LoopPassManager.

llvm-svn: 159183

2 files changed, 8 insertions, 11 deletions

diff --git a/llvm/lib/Analysis/LoopInfo.cpp b/llvm/lib/Analysis/LoopInfo.cpp
index c5e5bbe266d..20c33a3d9d6 100644
--- a/llvm/lib/Analysis/LoopInfo.cpp
+++ b/llvm/lib/Analysis/LoopInfo.cpp
@@ -44,10 +44,6 @@ static cl::opt<bool,true>
 VerifyLoopInfoX("verify-loop-info", cl::location(VerifyLoopInfo),
                 cl::desc("Verify loop info (time consuming)"));
 
-static cl::opt<bool>
-StableLoopInfo("stable-loops", cl::Hidden, cl::init(false),
-               cl::desc("Compute a stable loop tree."));
-
 char LoopInfo::ID = 0;
 INITIALIZE_PASS_BEGIN(LoopInfo, "loops", "Natural Loop Information", true, true)
 INITIALIZE_PASS_DEPENDENCY(DominatorTree)
@@ -516,10 +512,7 @@ Loop *UnloopUpdater::getNearestLoop(BasicBlock *BB, Loop *BBLoop) {
 //
 bool LoopInfo::runOnFunction(Function &) {
   releaseMemory();
-  if (StableLoopInfo)
-    LI.Analyze(getAnalysis<DominatorTree>().getBase());
-  else
-    LI.Calculate(getAnalysis<DominatorTree>().getBase());    // Update
+  LI.Analyze(getAnalysis<DominatorTree>().getBase());
   return false;
 }
 
diff --git a/llvm/lib/Analysis/LoopPass.cpp b/llvm/lib/Analysis/LoopPass.cpp
index aba700ac5c3..1540112fe10 100644
--- a/llvm/lib/Analysis/LoopPass.cpp
+++ b/llvm/lib/Analysis/LoopPass.cpp
@@ -162,7 +162,7 @@ void LPPassManager::deleteSimpleAnalysisValue(Value *V, Loop *L) {
 // Recurse through all subloops and all loops  into LQ.
 static void addLoopIntoQueue(Loop *L, std::deque<Loop *> &LQ) {
   LQ.push_back(L);
-  for (Loop::iterator I = L->begin(), E = L->end(); I != E; ++I)
+  for (Loop::reverse_iterator I = L->rbegin(), E = L->rend(); I != E; ++I)
     addLoopIntoQueue(*I, LQ);
 }
 
@@ -183,8 +183,12 @@ bool LPPassManager::runOnFunction(Function &F) {
   // Collect inherited analysis from Module level pass manager.
   populateInheritedAnalysis(TPM->activeStack);
 
-  // Populate Loop Queue
-  for (LoopInfo::iterator I = LI->begin(), E = LI->end(); I != E; ++I)
+  // Populate the loop queue in reverse program order. There is no clear need to
+  // process sibling loops in either forward or reverse order. There may be some
+  // advantage in deleting uses in a later loop before optimizing the
+  // definitions in an earlier loop. If we find a clear reason to process in
+  // forward order, then a forward variant of LoopPassManager should be created.
+  for (LoopInfo::reverse_iterator I = LI->rbegin(), E = LI->rend(); I != E; ++I)
     addLoopIntoQueue(*I, LQ);
 
   if (LQ.empty()) // No loops, skip calling finalizers