Index: llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -540,6 +540,12 @@
 
   /// Return the maximum number of waves per SIMD for kernels using \p VGPRs VGPRs
   unsigned getOccupancyWithNumVGPRs(unsigned VGPRs) const;
+
+  /// \returns True if waitcnt instruction is needed before barrier instruction,
+  /// false otherwise.
+  bool needWaitcntBeforeBarrier() const {
+    return true;
+  }
 };
 
 } // End namespace llvm
Index: llvm/trunk/lib/Target/AMDGPU/SIInsertWaits.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIInsertWaits.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SIInsertWaits.cpp
@@ -590,8 +590,9 @@
       // S_SENDMSG implicitly waits for all outstanding LGKM transfers to finish,
       // but we also want to wait for any other outstanding transfers before
       // signalling other hardware blocks
-      if (I->getOpcode() == AMDGPU::S_BARRIER ||
-          I->getOpcode() == AMDGPU::S_SENDMSG)
+      if ((I->getOpcode() == AMDGPU::S_BARRIER &&
+               ST->needWaitcntBeforeBarrier()) ||
+           I->getOpcode() == AMDGPU::S_SENDMSG)
         Required = LastIssued;
       else
         Required = handleOperands(*I);