diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index 881775eaa31..ea6071bc245 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -1285,7 +1285,43 @@ SDValue AMDGPUTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
     return DAG.getMergeValues(Ops, DL);
   }
 
-  return SDValue();
+  if (Load->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS ||
+      ExtType == ISD::NON_EXTLOAD || Load->getMemoryVT().bitsGE(MVT::i32))
+    return SDValue();
+
+
+  SDValue Ptr = DAG.getNode(ISD::SRL, DL, MVT::i32, Load->getBasePtr(),
+                            DAG.getConstant(2, MVT::i32));
+  SDValue Ret = DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, Op.getValueType(),
+                            Load->getChain(), Ptr,
+                            DAG.getTargetConstant(0, MVT::i32),
+                            Op.getOperand(2));
+  SDValue ByteIdx = DAG.getNode(ISD::AND, DL, MVT::i32,
+                                Load->getBasePtr(),
+                                DAG.getConstant(0x3, MVT::i32));
+  SDValue ShiftAmt = DAG.getNode(ISD::SHL, DL, MVT::i32, ByteIdx,
+                                 DAG.getConstant(3, MVT::i32));
+
+  Ret = DAG.getNode(ISD::SRL, DL, MVT::i32, Ret, ShiftAmt);
+
+  EVT MemEltVT = MemVT.getScalarType();
+  if (ExtType == ISD::SEXTLOAD) {
+    SDValue MemEltVTNode = DAG.getValueType(MemEltVT);
+
+    SDValue Ops[] = {
+      DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i32, Ret, MemEltVTNode),
+      Load->getChain()
+    };
+
+    return DAG.getMergeValues(Ops, DL);
+  }
+
+  SDValue Ops[] = {
+    DAG.getZeroExtendInReg(Ret, DL, MemEltVT),
+    Load->getChain()
+  };
+
+  return DAG.getMergeValues(Ops, DL);
 }
 
 SDValue AMDGPUTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
diff --git a/lib/Target/R600/R600ISelLowering.cpp b/lib/Target/R600/R600ISelLowering.cpp
index eb16fcc5458..b16d53fd015 100644
--- a/lib/Target/R600/R600ISelLowering.cpp
+++ b/lib/Target/R600/R600ISelLowering.cpp
@@ -1514,8 +1514,6 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
   EVT VT = Op.getValueType();
   SDLoc DL(Op);
   LoadSDNode *LoadNode = cast<LoadSDNode>(Op);
-  ISD::LoadExtType ExtType = LoadNode->getExtensionType();
-  EVT MemVT = LoadNode->getMemoryVT();
   SDValue Chain = Op.getOperand(0);
   SDValue Ptr = Op.getOperand(1);
   SDValue LoweredLoad;
@@ -1529,45 +1527,6 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
     return DAG.getMergeValues(Ops, DL);
   }
 
-  // Handle ext private loads
-  if (LoadNode->getAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS &&
-      ExtType != ISD::NON_EXTLOAD && LoadNode->getMemoryVT().bitsLT(MVT::i32)) {
-
-
-    SDValue Ptr = DAG.getNode(ISD::SRL, DL, MVT::i32, LoadNode->getBasePtr(),
-                              DAG.getConstant(2, MVT::i32));
-    SDValue Ret = DAG.getNode(AMDGPUISD::REGISTER_LOAD, DL, Op.getValueType(),
-                              LoadNode->getChain(), Ptr,
-                              DAG.getTargetConstant(0, MVT::i32),
-                              Op.getOperand(2));
-    SDValue ByteIdx = DAG.getNode(ISD::AND, DL, MVT::i32,
-                                  LoadNode->getBasePtr(),
-                                  DAG.getConstant(0x3, MVT::i32));
-    SDValue ShiftAmt = DAG.getNode(ISD::SHL, DL, MVT::i32, ByteIdx,
-                                   DAG.getConstant(3, MVT::i32));
-
-    Ret = DAG.getNode(ISD::SRL, DL, MVT::i32, Ret, ShiftAmt);
-
-    EVT MemEltVT = MemVT.getScalarType();
-    if (ExtType == ISD::SEXTLOAD) {
-      SDValue MemEltVTNode = DAG.getValueType(MemEltVT);
-
-      SDValue Ops[] = {
-        DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i32, Ret, MemEltVTNode),
-        LoadNode->getChain()
-      };
-
-      return DAG.getMergeValues(Ops, DL);
-    }
-
-    SDValue Ops[] = {
-      DAG.getZeroExtendInReg(Ret, DL, MemEltVT),
-      LoadNode->getChain()
-    };
-
-    return DAG.getMergeValues(Ops, DL);
-  }
-
   // Lower loads constant address space global variable loads
   if (LoadNode->getAddressSpace() == AMDGPUAS::CONSTANT_ADDRESS &&
       isa<GlobalVariable>(
diff --git a/test/CodeGen/R600/private-memory.ll b/test/CodeGen/R600/private-memory.ll
index 124d9fa6450..3ce8c2cb03d 100644
--- a/test/CodeGen/R600/private-memory.ll
+++ b/test/CodeGen/R600/private-memory.ll
@@ -118,8 +118,7 @@ for.end:
 
 ; SI-PROMOTE: BUFFER_STORE_SHORT v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s{{[0-9]+}}
 ; SI-PROMOTE: BUFFER_STORE_SHORT v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s{{[0-9]+}}
-; SI-PROMOTE-NOT: MOVREL
-; SI-PROMOTE: BUFFER_LOAD_SSHORT v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}] + v{{[0-9]+}} + s{{[0-9]+}}
+; SI_PROMOTE: BUFFER_LOAD_SSHORT v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}] + v{{[0-9]+}}, s{{[0-9]+}}
 define void @short_array(i32 addrspace(1)* %out, i32 %index) {
 entry:
   %0 = alloca [2 x i16]