Speed up the mem2reg transform for allocas which are only read/written in a single

basic block.  This is amazingly common in code generated by the C/C++ front-ends.
This change makes it not have to insert ANY phi nodes, whereas before it would insert
a ton of dead ones which DCE would have to clean up.

Thus, this fix improves compile-time performance of these trivial allocas in two ways:
  1. It doesn't have to do the walking and book-keeping for renaming
  2. It does not insert dead phi nodes for them which would have to
     subsequently be cleaned up.

On my favorite testcase from 252.eon, this special case handles 305 out of
356 promoted allocas in the function.  It speeds up the mem2reg pass from 7.5256s
to 1.2505s.  It inserts 677 fewer dead PHI nodes, which speeds up a subsequent
-dce pass from 18.7524s to 2.4806s.

There are still 120 trivially dead PHI nodes being inserted for variables used
in multiple basic blocks, but they are not handled by this patch.

llvm-svn: 8881

lib/Transforms/Utils/PromoteMemoryToRegister.cpp

@@ -38,11 +38,10 @@ bool isAllocaPromotable(const AllocaInst *AI, const TargetData &TD) {
   return true;
 }
 
-
 namespace {
   struct PromoteMem2Reg {
     // Allocas - The alloca instructions being promoted
-    const std::vector<AllocaInst*> &Allocas;
+    std::vector<AllocaInst*> Allocas;
     DominanceFrontier &DF;
     const TargetData &TD;
 
@@ -62,6 +61,8 @@ namespace {
     void run();
 
   private:
+    void PromoteLocallyUsedAlloca(AllocaInst *AI);
+
     void RenamePass(BasicBlock *BB, BasicBlock *Pred,
                     std::vector<Value*> &IncVals);
     bool QueuePhiNode(BasicBlock *BB, unsigned AllocaIdx, unsigned &Version);
@@ -79,13 +80,53 @@ void PromoteMem2Reg::run() {
     assert(Allocas[i]->getParent()->getParent() == &F &&
            "All allocas should be in the same function, which is same as DF!");
 
+    if (AI->use_empty()) {
+      // If there are no uses of the alloca, just delete it now.
+      AI->getParent()->getInstList().erase(AI);
+
+      // Remove the alloca from the Allocas list, since it has been processed
+      Allocas[i] = Allocas.back();
+      Allocas.pop_back();
+      --i;
+      continue;
+    }
+
     // Calculate the set of write-locations for each alloca.  This is analogous
     // to counting the number of 'redefinitions' of each variable.
     std::vector<BasicBlock*> DefiningBlocks;
-    for (Value::use_iterator U =AI->use_begin(), E = AI->use_end(); U != E; ++U)
-      if (StoreInst *SI = dyn_cast<StoreInst>(cast<Instruction>(*U)))
-        // jot down the basic-block it came from
+
+    BasicBlock *OnlyBlock = 0;
+    bool OnlyUsedInOneBlock = true;
+
+    // As we scan the uses of the alloca instruction, keep track of stores, and
+    // decide whether all of the loads and stores to the alloca are within the
+    // same basic block.
+    for (Value::use_iterator U =AI->use_begin(), E = AI->use_end(); U != E;++U){
+      Instruction *User = cast<Instruction>(*U);
+      if (StoreInst *SI = dyn_cast<StoreInst>(User)) {
+        // Remember the basic blocks which define new values for the alloca
         DefiningBlocks.push_back(SI->getParent());
+      }
+
+      if (OnlyUsedInOneBlock) {
+        if (OnlyBlock == 0)
+          OnlyBlock = User->getParent();
+        else if (OnlyBlock != User->getParent())
+          OnlyUsedInOneBlock = false;
+      }
+    }
+
+    // If the alloca is only read and written in one basic block, just perform a
+    // linear sweep over the block to eliminate it.
+    if (OnlyUsedInOneBlock) {
+      PromoteLocallyUsedAlloca(AI);
+
+      // Remove the alloca from the Allocas list, since it has been processed
+      Allocas[i] = Allocas.back();
+      Allocas.pop_back();
+      --i;
+      continue;
+    }
 
     AllocaLookup[Allocas[i]] = i;
     
@@ -113,6 +154,9 @@ void PromoteMem2Reg::run() {
     }
   }
   
+  if (Allocas.empty())
+    return; // All of the allocas must have been trivial!
+
   // Set the incoming values for the basic block to be null values for all of
   // the alloca's.  We do this in case there is a load of a value that has not
   // been stored yet.  In this case, it will get this null value.
@@ -194,6 +238,41 @@ void PromoteMem2Reg::run() {
   }
 }
 
+// PromoteLocallyUsedAlloca - Many allocas are only used within a single basic
+// block.  If this is the case, avoid traversing the CFG and inserting a lot of
+// potentially useless PHI nodes by just performing a single linear pass over
+// the basic block using the Alloca.
+//
+void PromoteMem2Reg::PromoteLocallyUsedAlloca(AllocaInst *AI) {
+  assert(!AI->use_empty() && "There are no uses of the alloca!");
+
+  // Uses of the uninitialized memory location shall get zero...
+  Value *CurVal = Constant::getNullValue(AI->getAllocatedType());
+  
+  BasicBlock *BB = cast<Instruction>(AI->use_back())->getParent();
+
+  for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ) {
+    Instruction *Inst = I++;
+    if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
+      if (LI->getOperand(0) == AI) {
+        // Loads just return the "current value"...
+        LI->replaceAllUsesWith(CurVal);
+        BB->getInstList().erase(LI);
+      }
+    } else if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
+      if (SI->getOperand(1) == AI) {
+        // Loads just update the "current value"...
+        CurVal = SI->getOperand(0);
+        BB->getInstList().erase(SI);
+      }
+    }
+  }
+
+  // After traversing the basic block, there should be no more uses of the
+  // alloca, remove it now.
+  assert(AI->use_empty() && "Uses of alloca from more than one BB??");
+  AI->getParent()->getInstList().erase(AI);
+}
 
 // QueuePhiNode - queues a phi-node to be added to a basic-block for a specific
 // Alloca returns true if there wasn't already a phi-node for that variable