Implement PR1786 by iterating between dead cycle elimination

and simplifycfg in the rare cases when it is needed.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44044 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Transforms/Scalar/SimplifyCFG.cpp

@@ -27,6 +27,7 @@
 #include "llvm/Support/CFG.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Pass.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/Statistic.h"
 using namespace llvm;
@@ -50,9 +51,9 @@ FunctionPass *llvm::createCFGSimplificationPass() {
 }
 
 static bool MarkAliveBlocks(BasicBlock *BB,
-                            SmallPtrSet<BasicBlock*, 16> &Reachable) {
+                            SmallPtrSet<BasicBlock*, 128> &Reachable) {
   
-  std::vector<BasicBlock*> Worklist;
+  SmallVector<BasicBlock*, 128> Worklist;
   Worklist.push_back(BB);
   bool Changed = false;
   while (!Worklist.empty()) {
@@ -93,38 +94,42 @@ static bool MarkAliveBlocks(BasicBlock *BB,
   return Changed;
 }
 
-
+/// RemoveUnreachableBlocks - Remove blocks that are not reachable, even if they
-// It is possible that we may require multiple passes over the code to fully
+/// are in a dead cycle.  Return true if a change was made, false otherwise.
-// simplify the CFG.
+static bool RemoveUnreachableBlocks(Function &F) {
-//
+  SmallPtrSet<BasicBlock*, 128> Reachable;
-bool CFGSimplifyPass::runOnFunction(Function &F) {
-  SmallPtrSet<BasicBlock*, 16> Reachable;
   bool Changed = MarkAliveBlocks(F.begin(), Reachable);
   
   // If there are unreachable blocks in the CFG...
-  if (Reachable.size() != F.size()) {
+  if (Reachable.size() == F.size())
-    assert(Reachable.size() < F.size());
+    return Changed;
-    NumSimpl += F.size()-Reachable.size();
   
-    // Loop over all of the basic blocks that are not reachable, dropping all of
+  assert(Reachable.size() < F.size());
-    // their internal references...
+  NumSimpl += F.size()-Reachable.size();
-    for (Function::iterator BB = ++F.begin(), E = F.end(); BB != E; ++BB)
-      if (!Reachable.count(BB)) {
-        for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB); SI!=SE; ++SI)
-          if (Reachable.count(*SI))
-            (*SI)->removePredecessor(BB);
-        BB->dropAllReferences();
-      }
   
-    for (Function::iterator I = ++F.begin(); I != F.end();)
+  // Loop over all of the basic blocks that are not reachable, dropping all of
-      if (!Reachable.count(I))
+  // their internal references...
-        I = F.getBasicBlockList().erase(I);
+  for (Function::iterator BB = ++F.begin(), E = F.end(); BB != E; ++BB)
-      else
+    if (!Reachable.count(BB)) {
-        ++I;
+      for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB); SI!=SE; ++SI)
+        if (Reachable.count(*SI))
+          (*SI)->removePredecessor(BB);
+      BB->dropAllReferences();
+    }
   
-    Changed = true;
+  for (Function::iterator I = ++F.begin(); I != F.end();)
-  }
+    if (!Reachable.count(I))
+      I = F.getBasicBlockList().erase(I);
+    else
+      ++I;
   
+  return true;
+}
+
+/// IterativeSimplifyCFG - Call SimplifyCFG on all the blocks in the function,
+/// iterating until no more changes are made.
+static bool IterativeSimplifyCFG(Function &F) {
+  bool Changed = false;
   bool LocalChange = true;
   while (LocalChange) {
     LocalChange = false;
@@ -140,6 +145,31 @@ bool CFGSimplifyPass::runOnFunction(Function &F) {
     }
     Changed |= LocalChange;
   }
-
   return Changed;
 }
+
+// It is possible that we may require multiple passes over the code to fully
+// simplify the CFG.
+//
+bool CFGSimplifyPass::runOnFunction(Function &F) {
+  bool EverChanged = RemoveUnreachableBlocks(F);
+  EverChanged |= IterativeSimplifyCFG(F);
+  
+  // If neither pass changed anything, we're done.
+  if (!EverChanged) return false;
+
+  // IterativeSimplifyCFG can (rarely) make some loops dead.  If this happens,
+  // RemoveUnreachableBlocks is needed to nuke them, which means we should
+  // iterate between the two optimizations.  We structure the code like this to
+  // avoid reruning IterativeSimplifyCFG if the second pass of 
+  // RemoveUnreachableBlocks doesn't do anything.
+  if (!RemoveUnreachableBlocks(F))
+    return true;
+  
+  do {
+    EverChanged = IterativeSimplifyCFG(F);
+    EverChanged |= RemoveUnreachableBlocks(F);
+  } while (EverChanged);
+  
+  return true;
+}

test/Transforms/SimplifyCFG/iterative-simplify.ll

@@ -0,0 +1,100 @@
+; RUN: llvm-as < %s | opt -simplifycfg | llvm-dis | not grep bb17
+; PR1786
+
+define i32 @main() {
+entry:
+	%retval = alloca i32, align 4		; <i32*> [#uses=1]
+	%i = alloca i32, align 4		; <i32*> [#uses=7]
+	%z = alloca i32, align 4		; <i32*> [#uses=4]
+	%z16 = alloca i32, align 4		; <i32*> [#uses=4]
+	%"alloca point" = bitcast i32 0 to i32		; <i32> [#uses=0]
+	store i32 0, i32* %i
+	%toBool = icmp ne i8 1, 0		; <i1> [#uses=1]
+	br i1 %toBool, label %cond_true, label %cond_false
+
+cond_true:		; preds = %entry
+	store i32 0, i32* %z
+	br label %bb
+
+bb:		; preds = %cond_next, %cond_true
+	%tmp = load i32* %z		; <i32> [#uses=1]
+	%tmp1 = sub i32 %tmp, 16384		; <i32> [#uses=1]
+	store i32 %tmp1, i32* %z
+	%tmp2 = load i32* %i		; <i32> [#uses=1]
+	%tmp3 = add i32 %tmp2, 1		; <i32> [#uses=1]
+	store i32 %tmp3, i32* %i
+	%tmp4 = load i32* %i		; <i32> [#uses=1]
+	%tmp5 = icmp sgt i32 %tmp4, 262144		; <i1> [#uses=1]
+	%tmp56 = zext i1 %tmp5 to i8		; <i8> [#uses=1]
+	%toBool7 = icmp ne i8 %tmp56, 0		; <i1> [#uses=1]
+	br i1 %toBool7, label %cond_true8, label %cond_next
+
+cond_true8:		; preds = %bb
+	call void @abort( )
+	unreachable
+
+cond_next:		; preds = %bb
+	%tmp9 = load i32* %z		; <i32> [#uses=1]
+	%tmp10 = icmp ne i32 %tmp9, 0		; <i1> [#uses=1]
+	%tmp1011 = zext i1 %tmp10 to i8		; <i8> [#uses=1]
+	%toBool12 = icmp ne i8 %tmp1011, 0		; <i1> [#uses=1]
+	br i1 %toBool12, label %bb, label %bb13
+
+bb13:		; preds = %cond_next
+	call void @exit( i32 0 )
+	unreachable
+
+cond_false:		; preds = %entry
+	%toBool14 = icmp ne i8 1, 0		; <i1> [#uses=1]
+	br i1 %toBool14, label %cond_true15, label %cond_false33
+
+cond_true15:		; preds = %cond_false
+	store i32 0, i32* %z16
+	br label %bb17
+
+bb17:		; preds = %cond_next27, %cond_true15
+	%tmp18 = load i32* %z16		; <i32> [#uses=1]
+	%tmp19 = sub i32 %tmp18, 16384		; <i32> [#uses=1]
+	store i32 %tmp19, i32* %z16
+	%tmp20 = load i32* %i		; <i32> [#uses=1]
+	%tmp21 = add i32 %tmp20, 1		; <i32> [#uses=1]
+	store i32 %tmp21, i32* %i
+	%tmp22 = load i32* %i		; <i32> [#uses=1]
+	%tmp23 = icmp sgt i32 %tmp22, 262144		; <i1> [#uses=1]
+	%tmp2324 = zext i1 %tmp23 to i8		; <i8> [#uses=1]
+	%toBool25 = icmp ne i8 %tmp2324, 0		; <i1> [#uses=1]
+	br i1 %toBool25, label %cond_true26, label %cond_next27
+
+cond_true26:		; preds = %bb17
+	call void @abort( )
+	unreachable
+
+cond_next27:		; preds = %bb17
+	%tmp28 = load i32* %z16		; <i32> [#uses=1]
+	%tmp29 = icmp ne i32 %tmp28, 0		; <i1> [#uses=1]
+	%tmp2930 = zext i1 %tmp29 to i8		; <i8> [#uses=1]
+	%toBool31 = icmp ne i8 %tmp2930, 0		; <i1> [#uses=1]
+	br i1 %toBool31, label %bb17, label %bb32
+
+bb32:		; preds = %cond_next27
+	call void @exit( i32 0 )
+	unreachable
+
+cond_false33:		; preds = %cond_false
+	call void @exit( i32 0 )
+	unreachable
+
+cond_next34:		; No predecessors!
+	br label %cond_next35
+
+cond_next35:		; preds = %cond_next34
+	br label %return
+
+return:		; preds = %cond_next35
+	%retval36 = load i32* %retval		; <i32> [#uses=1]
+	ret i32 %retval36
+}
+
+declare void @abort()
+
+declare void @exit(i32)