Run code placement optimization for targets that want it (arm and x86 for now).

llvm-svn: 71726

include/llvm/Target/TargetLowering.h

@@ -620,6 +620,13 @@ public:
     return allowUnalignedMemoryAccesses;
   }
 
+  /// This function returns true if the target would benefit from code placement
+  /// optimization.
+  /// @brief Determine if the target should perform code placement optimization.
+  bool shouldOptimizeCodePlacement() const {
+    return benefitFromCodePlacementOpt;
+  }
+
   /// getOptimalMemOpType - Returns the target specific optimal type for load
   /// and store operations as a result of memset, memcpy, and memmove lowering.
   /// It returns MVT::iAny if SelectionDAG should be responsible for
@@ -1652,6 +1659,10 @@ protected:
   /// operations when copying small arrays and other similar tasks.
   /// @brief Indicate whether the target permits unaligned memory accesses.
   bool allowUnalignedMemoryAccesses;
+
+  /// This field specifies whether the target can benefit from code placement
+  /// optimization.
+  bool benefitFromCodePlacementOpt;
 };
 } // end llvm namespace
 

lib/CodeGen/CodePlacementOpt.cpp

@@ -104,6 +104,9 @@ FunctionPass *llvm::createCodePlacementOptPass() {
 ///       jcc <cond> C, [exit]
 ///
 bool CodePlacementOpt::OptimizeIntraLoopEdges() {
+  if (!TLI->shouldOptimizeCodePlacement())
+    return false;
+
   bool Changed = false;
   for (unsigned i = 0, e = UncondJmpMBBs.size(); i != e; ++i) {
     MachineBasicBlock *MBB = UncondJmpMBBs[i].first;

lib/CodeGen/SelectionDAG/TargetLowering.cpp

@@ -483,6 +483,7 @@ TargetLowering::TargetLowering(TargetMachine &tm)
   memset(TargetDAGCombineArray, 0, array_lengthof(TargetDAGCombineArray));
   maxStoresPerMemset = maxStoresPerMemcpy = maxStoresPerMemmove = 8;
   allowUnalignedMemoryAccesses = false;
+  benefitFromCodePlacementOpt = false;
   UseUnderscoreSetJmp = false;
   UseUnderscoreLongJmp = false;
   SelectIsExpensive = false;

lib/Target/ARM/ARMISelLowering.cpp

@@ -288,6 +288,7 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
   setIfCvtDupBlockSizeLimit(Subtarget->isThumb() ? 0 : 2);
 
   maxStoresPerMemcpy = 1;   //// temporary - rewrite interface to use type
+  benefitFromCodePlacementOpt = true;
 }
 
 const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const {

lib/Target/X86/X86ISelLowering.cpp

@@ -844,6 +844,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
   maxStoresPerMemmove = 3; // For @llvm.memmove -> sequence of stores
   allowUnalignedMemoryAccesses = true; // x86 supports it!
   setPrefLoopAlignment(16);
+  benefitFromCodePlacementOpt = true;
 }