Index: lib/Target/AMDGPU/AMDGPU.td
===================================================================
--- lib/Target/AMDGPU/AMDGPU.td
+++ lib/Target/AMDGPU/AMDGPU.td
@@ -108,6 +108,11 @@
         "true",
         "Force using DS instruction immediate offsets on SI">;
 
+def FeatureFlatForGlobal : SubtargetFeature<"flat-for-global",
+        "FlatForGlobal",
+        "true",
+        "Force to generate flat instruction for global">;
+
 def FeatureFlatAddressSpace : SubtargetFeature<"flat-address-space",
         "FlatAddressSpace",
         "true",
Index: lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -95,7 +95,7 @@
   bool SelectDS1Addr1Offset(SDValue Ptr, SDValue &Base, SDValue &Offset) const;
   bool SelectDS64Bit4ByteAligned(SDValue Ptr, SDValue &Base, SDValue &Offset0,
                                  SDValue &Offset1) const;
-  void SelectMUBUF(SDValue Addr, SDValue &SRsrc, SDValue &VAddr,
+  bool SelectMUBUF(SDValue Addr, SDValue &SRsrc, SDValue &VAddr,
                    SDValue &SOffset, SDValue &Offset, SDValue &Offen,
                    SDValue &Idxen, SDValue &Addr64, SDValue &GLC, SDValue &SLC,
                    SDValue &TFE) const;
@@ -920,12 +920,16 @@
   return isUInt<12>(Imm->getZExtValue());
 }
 
-void AMDGPUDAGToDAGISel::SelectMUBUF(SDValue Addr, SDValue &Ptr,
+bool AMDGPUDAGToDAGISel::SelectMUBUF(SDValue Addr, SDValue &Ptr,
                                      SDValue &VAddr, SDValue &SOffset,
                                      SDValue &Offset, SDValue &Offen,
                                      SDValue &Idxen, SDValue &Addr64,
                                      SDValue &GLC, SDValue &SLC,
                                      SDValue &TFE) const {
+  // Subtarget prefers to use flat instruction
+  if (Subtarget->useFlatForGlobal())
+    return false;
+
   SDLoc DL(Addr);
 
   GLC = CurDAG->getTargetConstant(0, DL, MVT::i1);
@@ -958,14 +962,14 @@
 
     if (isLegalMUBUFImmOffset(C1)) {
         Offset = CurDAG->getTargetConstant(C1->getZExtValue(), DL, MVT::i16);
-        return;
+        return true;
     } else if (isUInt<32>(C1->getZExtValue())) {
       // Illegal offset, store it in soffset.
       Offset = CurDAG->getTargetConstant(0, DL, MVT::i16);
       SOffset = SDValue(CurDAG->getMachineNode(AMDGPU::S_MOV_B32, DL, MVT::i32,
                    CurDAG->getTargetConstant(C1->getZExtValue(), DL, MVT::i32)),
                         0);
-      return;
+      return true;
     }
   }
 
@@ -977,13 +981,15 @@
     Ptr = N0;
     VAddr = N1;
     Offset = CurDAG->getTargetConstant(0, DL, MVT::i16);
-    return;
+    return true;
   }
 
   // default case -> offset
   VAddr = CurDAG->getTargetConstant(0, DL, MVT::i32);
   Ptr = Addr;
   Offset = CurDAG->getTargetConstant(0, DL, MVT::i16);
+
+  return true;
 }
 
 bool AMDGPUDAGToDAGISel::SelectMUBUFAddr64(SDValue Addr, SDValue &SRsrc,
@@ -996,8 +1002,9 @@
   if (Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
     return false;
 
-  SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
-              GLC, SLC, TFE);
+  if (!SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
+              GLC, SLC, TFE))
+    return false;
 
   ConstantSDNode *C = cast<ConstantSDNode>(Addr64);
   if (C->getSExtValue()) {
@@ -1063,8 +1070,9 @@
   const SIInstrInfo *TII =
     static_cast<const SIInstrInfo *>(Subtarget->getInstrInfo());
 
-  SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
-              GLC, SLC, TFE);
+  if (!SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
+              GLC, SLC, TFE))
+    return false;
 
   if (!cast<ConstantSDNode>(Offen)->getSExtValue() &&
       !cast<ConstantSDNode>(Idxen)->getSExtValue() &&
Index: lib/Target/AMDGPU/AMDGPUSubtarget.h
===================================================================
--- lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -70,6 +70,7 @@
   bool FastFMAF32;
   bool CaymanISA;
   bool FlatAddressSpace;
+  bool FlatForGlobal;
   bool EnableIRStructurizer;
   bool EnablePromoteAlloca;
   bool EnableIfCvt;
@@ -159,6 +160,10 @@
     return FlatAddressSpace;
   }
 
+  bool useFlatForGlobal() const {
+    return FlatForGlobal;
+  }
+
   bool hasBFE() const {
     return (getGeneration() >= EVERGREEN);
   }
Index: lib/Target/AMDGPU/AMDGPUSubtarget.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUSubtarget.cpp
+++ lib/Target/AMDGPU/AMDGPUSubtarget.cpp
@@ -68,9 +68,9 @@
       DumpCode(false), R600ALUInst(false), HasVertexCache(false),
       TexVTXClauseSize(0), Gen(AMDGPUSubtarget::R600), FP64(false),
       FP64Denormals(false), FP32Denormals(false), FastFMAF32(false),
-      CaymanISA(false), FlatAddressSpace(false), EnableIRStructurizer(true),
-      EnablePromoteAlloca(false), EnableIfCvt(true), EnableLoadStoreOpt(false),
-      EnableUnsafeDSOffsetFolding(false),
+      CaymanISA(false), FlatAddressSpace(false), FlatForGlobal(false),
+      EnableIRStructurizer(true), EnablePromoteAlloca(false), EnableIfCvt(true),
+      EnableLoadStoreOpt(false), EnableUnsafeDSOffsetFolding(false),
       WavefrontSize(0), CFALUBug(false), LocalMemorySize(0),
       EnableVGPRSpilling(false), SGPRInitBug(false), IsGCN(false),
       GCN1Encoding(false), GCN3Encoding(false), CIInsts(false), LDSBankCount(0),
@@ -78,6 +78,9 @@
       FrameLowering(nullptr),
       InstrItins(getInstrItineraryForCPU(GPU)), TargetTriple(TT) {
 
+  if (isAmdHsaOS())
+    FlatForGlobal = true;
+
   initializeSubtargetDependencies(TT, GPU, FS);
 
   const unsigned MaxStackAlign = 64 * 16; // Maximum stack alignment (long16)
Index: lib/Target/AMDGPU/CIInstructions.td
===================================================================
--- lib/Target/AMDGPU/CIInstructions.td
+++ lib/Target/AMDGPU/CIInstructions.td
@@ -33,6 +33,8 @@
 
 def HasFlatAddressSpace : Predicate<"Subtarget->hasFlatAddressSpace()">;
 
+
+
 //===----------------------------------------------------------------------===//
 // VOP1 Instructions
 //===----------------------------------------------------------------------===//
@@ -234,3 +236,64 @@
 >;
 
 } // End Predicates = [isCI]
+
+
+//===----------------------------------------------------------------------===//
+// Patterns to generate flat for global
+//===----------------------------------------------------------------------===//
+
+def useFlatForGlobal : Predicate <
+  "Subtarget->useFlatForGlobal() || "
+  "Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS">;
+
+let Predicates = [useFlatForGlobal] in {
+
+// 1. Offset as 20bit DWORD immediate
+def : Pat <
+  (SIload_constant v4i32:$sbase, IMM20bit:$offset),
+  (S_BUFFER_LOAD_DWORD_IMM $sbase, (as_i32imm $offset))
+>;
+
+// Patterns for global loads with no offset
+class FlatLoadPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (vt (node i64:$addr)),
+  (inst $addr, 0, 0, 0)
+>;
+
+def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_USHORT, az_extloadi16_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_SSHORT, sextloadi16_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORD, global_load, i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORDX2, global_load, v2i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORDX4, global_load, v4i32>;
+
+class FlatStorePat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (node vt:$data, i64:$addr),
+  (inst $data, $addr, 0, 0, 0)
+>;
+
+def : FlatStorePat <FLAT_STORE_BYTE, truncstorei8_global, i32>;
+def : FlatStorePat <FLAT_STORE_SHORT, truncstorei16_global, i32>;
+def : FlatStorePat <FLAT_STORE_DWORD, global_store, i32>;
+def : FlatStorePat <FLAT_STORE_DWORDX2, global_store, v2i32>;
+def : FlatStorePat <FLAT_STORE_DWORDX4, global_store, v4i32>;
+
+class FlatAtomicPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (vt (node i64:$addr, vt:$data)),
+  (inst $addr, $data, 0, 0)
+>;
+
+def : FlatAtomicPat <FLAT_ATOMIC_ADD_RTN, atomic_add_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_AND_RTN, atomic_and_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SUB_RTN, atomic_sub_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SMAX_RTN, atomic_max_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_UMAX_RTN, atomic_umax_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SMIN_RTN, atomic_min_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_UMIN_RTN, atomic_umin_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_OR_RTN, atomic_or_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SWAP_RTN, atomic_swap_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_XOR_RTN, atomic_xor_global, i32>;
+
+
+} // End Predicates = [useFlatForGlobal]
Index: lib/Target/AMDGPU/VIInstructions.td
===================================================================
--- lib/Target/AMDGPU/VIInstructions.td
+++ lib/Target/AMDGPU/VIInstructions.td
@@ -101,58 +101,3 @@
 
 } // End SIAssemblerPredicate = DisableInst, SubtargetPredicate = isVI
 
-//===----------------------------------------------------------------------===//
-// SMEM Patterns
-//===----------------------------------------------------------------------===//
-
-let Predicates = [isVI] in {
-
-// 1. Offset as 20bit DWORD immediate
-def : Pat <
-  (SIload_constant v4i32:$sbase, IMM20bit:$offset),
-  (S_BUFFER_LOAD_DWORD_IMM $sbase, (as_i32imm $offset))
->;
-
-// Patterns for global loads with no offset
-class FlatLoadPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
-  (vt (node i64:$addr)),
-  (inst $addr, 0, 0, 0)
->;
-
-def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_global, i32>;
-def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_global, i32>;
-def : FlatLoadPat <FLAT_LOAD_USHORT, az_extloadi16_global, i32>;
-def : FlatLoadPat <FLAT_LOAD_SSHORT, sextloadi16_global, i32>;
-def : FlatLoadPat <FLAT_LOAD_DWORD, global_load, i32>;
-def : FlatLoadPat <FLAT_LOAD_DWORDX2, global_load, v2i32>;
-def : FlatLoadPat <FLAT_LOAD_DWORDX4, global_load, v4i32>;
-
-class FlatStorePat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
-  (node vt:$data, i64:$addr),
-  (inst $data, $addr, 0, 0, 0)
->;
-
-def : FlatStorePat <FLAT_STORE_BYTE, truncstorei8_global, i32>;
-def : FlatStorePat <FLAT_STORE_SHORT, truncstorei16_global, i32>;
-def : FlatStorePat <FLAT_STORE_DWORD, global_store, i32>;
-def : FlatStorePat <FLAT_STORE_DWORDX2, global_store, v2i32>;
-def : FlatStorePat <FLAT_STORE_DWORDX4, global_store, v4i32>;
-
-class FlatAtomicPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
-  (vt (node i64:$addr, vt:$data)),
-  (inst $addr, $data, 0, 0)
->;
-
-def : FlatAtomicPat <FLAT_ATOMIC_ADD_RTN, atomic_add_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_AND_RTN, atomic_and_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_SUB_RTN, atomic_sub_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_SMAX_RTN, atomic_max_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_UMAX_RTN, atomic_umax_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_SMIN_RTN, atomic_min_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_UMIN_RTN, atomic_umin_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_OR_RTN, atomic_or_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_SWAP_RTN, atomic_swap_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_XOR_RTN, atomic_xor_global, i32>;
-
-
-} // End Predicates = [isVI]
Index: test/CodeGen/AMDGPU/ci-use-flat-for-global.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AMDGPU/ci-use-flat-for-global.ll
@@ -0,0 +1,15 @@
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | FileCheck -check-prefix=HSA-DEFAULT %s
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-flat-for-global | FileCheck -check-prefix=HSA-NODEFAULT %s
+; RUN: llc < %s -mtriple=amdgcn -mcpu=kaveri | FileCheck -check-prefix=NOHSA-DEFAULT %s
+; RUN: llc < %s -mtriple=amdgcn -mcpu=kaveri -mattr=+flat-for-global | FileCheck -check-prefix=NOHSA-NODEFAULT %s
+
+
+; HSA-DEFAULT: flat_store_dword
+; HSA-NODEFAULT: buffer_store_dword
+; NOHSA-DEFAULT: buffer_store_dword
+; NOHSA-NODEFAULT: flat_store_dword
+define void @test(i32 addrspace(1)* %out) {
+entry:
+  store i32 0, i32 addrspace(1)* %out
+  ret void
+}