5 files changed, 38 insertions, 6 deletions
diff --git a/lib/CodeGen/BranchFolding.cpp b/lib/CodeGen/BranchFolding.cpp
index d8ec340692..7cc4c2db79 100644
--- a/lib/CodeGen/BranchFolding.cpp
+++ b/lib/CodeGen/BranchFolding.cpp
@@ -35,12 +35,19 @@ using namespace llvm;
 STATISTIC(NumDeadBlocks, "Number of dead blocks removed");
 STATISTIC(NumBranchOpts, "Number of branches optimized");
 STATISTIC(NumTailMerge , "Number of block tails merged");
-static cl::opt<bool> EnableTailMerge("enable-tail-merge", cl::Hidden);
-
+static cl::opt<cl::boolOrDefault> FlagEnableTailMerge("enable-tail-merge", 
+                              cl::init(cl::BOU_UNSET), cl::Hidden);
 namespace {
   struct BranchFolder : public MachineFunctionPass {
     static char ID;
-    BranchFolder() : MachineFunctionPass((intptr_t)&ID) {}
+    BranchFolder(bool defaultEnableTailMerge) : 
+        MachineFunctionPass((intptr_t)&ID) {
+          switch (FlagEnableTailMerge) {
+          case cl::BOU_UNSET: EnableTailMerge = defaultEnableTailMerge; break;
+          case cl::BOU_TRUE: EnableTailMerge = true; break;
+          case cl::BOU_FALSE: EnableTailMerge = false; break;
+          }
+    }
 
     virtual bool runOnMachineFunction(MachineFunction &MF);
     virtual const char *getPassName() const { return "Control Flow Optimizer"; }
@@ -49,6 +56,7 @@ namespace {
     bool MadeChange;
   private:
     // Tail Merging.
+    bool EnableTailMerge;
     bool TailMergeBlocks(MachineFunction &MF);
     bool TryMergeBlocks(MachineBasicBlock* SuccBB,
                         MachineBasicBlock* PredBB);
@@ -79,7 +87,8 @@ static bool CorrectExtraCFGEdges(MachineBasicBlock &MBB,
                                  bool isCond, 
                                  MachineFunction::iterator FallThru);
 
-FunctionPass *llvm::createBranchFoldingPass() { return new BranchFolder(); }
+FunctionPass *llvm::createBranchFoldingPass(bool DefaultEnableTailMerge) { 
+      return new BranchFolder(DefaultEnableTailMerge); }
 
 /// RemoveDeadBlock - Remove the specified dead machine basic block from the
 /// function, updating the CFG.
diff --git a/lib/CodeGen/LLVMTargetMachine.cpp b/lib/CodeGen/LLVMTargetMachine.cpp
index 34c45f3135..41f7e199a0 100644
--- a/lib/CodeGen/LLVMTargetMachine.cpp
+++ b/lib/CodeGen/LLVMTargetMachine.cpp
@@ -78,7 +78,7 @@ LLVMTargetMachine::addPassesToEmitFile(FunctionPassManager &PM,
   
   // Branch folding must be run after regalloc and prolog/epilog insertion.
   if (!Fast)
-    PM.add(createBranchFoldingPass());
+    PM.add(createBranchFoldingPass(DoTailMergeDefault()));
     
   // Fold redundant debug labels.
   PM.add(createDebugLabelFoldingPass());
@@ -181,7 +181,7 @@ bool LLVMTargetMachine::addPassesToEmitMachineCode(FunctionPassManager &PM,
   
   // Branch folding must be run after regalloc and prolog/epilog insertion.
   if (!Fast)
-    PM.add(createBranchFoldingPass());
+    PM.add(createBranchFoldingPass(DoTailMergeDefault()));
   
   if (addPreEmitPass(PM, Fast) && PrintMachineCode)
     PM.add(createMachineFunctionPrinterPass(cerr));
diff --git a/lib/Support/CommandLine.cpp b/lib/Support/CommandLine.cpp
index 2b0798aa77..1f5008a3fa 100644
--- a/lib/Support/CommandLine.cpp
+++ b/lib/Support/CommandLine.cpp
@@ -36,6 +36,7 @@ using namespace cl;
 // Template instantiations and anchors.
 //
 TEMPLATE_INSTANTIATION(class basic_parser<bool>);
+TEMPLATE_INSTANTIATION(class basic_parser<boolOrDefault>);
 TEMPLATE_INSTANTIATION(class basic_parser<int>);
 TEMPLATE_INSTANTIATION(class basic_parser<unsigned>);
 TEMPLATE_INSTANTIATION(class basic_parser<double>);
@@ -50,6 +51,7 @@ TEMPLATE_INSTANTIATION(class opt<bool>);
 void Option::anchor() {}
 void basic_parser_impl::anchor() {}
 void parser<bool>::anchor() {}
+void parser<boolOrDefault>::anchor() {}
 void parser<int>::anchor() {}
 void parser<unsigned>::anchor() {}
 void parser<double>::anchor() {}
@@ -767,6 +769,22 @@ bool parser<bool>::parse(Option &O, const char *ArgName,
   return false;
 }
 
+// parser<boolOrDefault> implementation
+//
+bool parser<boolOrDefault>::parse(Option &O, const char *ArgName,
+                         const std::string &Arg, boolOrDefault &Value) {
+  if (Arg == "" || Arg == "true" || Arg == "TRUE" || Arg == "True" ||
+      Arg == "1") {
+    Value = BOU_TRUE;
+  } else if (Arg == "false" || Arg == "FALSE" || Arg == "False" || Arg == "0") {
+    Value = BOU_FALSE;
+  } else {
+    return O.error(": '" + Arg +
+                   "' is invalid value for boolean argument! Try 0 or 1");
+  }
+  return false;
+}
+
 // parser<int> implementation
 //
 bool parser<int>::parse(Option &O, const char *ArgName,
diff --git a/lib/Target/PowerPC/PPCTargetMachine.cpp b/lib/Target/PowerPC/PPCTargetMachine.cpp
index df3e3d2257..6d5e050773 100644
--- a/lib/Target/PowerPC/PPCTargetMachine.cpp
+++ b/lib/Target/PowerPC/PPCTargetMachine.cpp
@@ -96,6 +96,10 @@ PPCTargetMachine::PPCTargetMachine(const Module &M, const std::string &FS,
       setRelocationModel(Reloc::Static);
 }
 
+/// Override this for PowerPC.  Tail merging happily breaks up instruction issue
+/// groups, which typically degrades performance.
+const bool PPCTargetMachine::DoTailMergeDefault() const { return false; }
+
 PPC32TargetMachine::PPC32TargetMachine(const Module &M, const std::string &FS) 
   : PPCTargetMachine(M, FS, false) {
 }
diff --git a/lib/Target/PowerPC/PPCTargetMachine.h b/lib/Target/PowerPC/PPCTargetMachine.h
index a249952aa1..fdf9d35e57 100644
--- a/lib/Target/PowerPC/PPCTargetMachine.h
+++ b/lib/Target/PowerPC/PPCTargetMachine.h
@@ -73,6 +73,7 @@ public:
                               MachineCodeEmitter &MCE);
   virtual bool addSimpleCodeEmitter(FunctionPassManager &PM, bool Fast,
                                     MachineCodeEmitter &MCE);
+  virtual const bool DoTailMergeDefault() const;
 };
 
 /// PPC32TargetMachine - PowerPC 32-bit target machine.