diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index 9df835fc874..88867b6e9da 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -495,9 +495,9 @@ SDValue AMDGPUTargetLowering::SplitVectorStore(SDValue Op,
                               Store->getBasePtr(),
                             DAG.getConstant(i * (MemEltVT.getSizeInBits() / 8),
                                             PtrVT));
-    Chains.push_back(DAG.getStore(Store->getChain(), SL, Val, Ptr,
+    Chains.push_back(DAG.getTruncStore(Store->getChain(), SL, Val, Ptr,
                          MachinePointerInfo(Store->getMemOperand()->getValue()),
-                         Store->isVolatile(), Store->isNonTemporal(),
+                         MemEltVT, Store->isVolatile(), Store->isNonTemporal(),
                          Store->getAlignment()));
   }
   return DAG.getNode(ISD::TokenFactor, SL, MVT::Other, &Chains[0], NumElts);
diff --git a/lib/Target/R600/AMDGPUInstrInfo.td b/lib/Target/R600/AMDGPUInstrInfo.td
index c61993a8cda..c0d757e2758 100644
--- a/lib/Target/R600/AMDGPUInstrInfo.td
+++ b/lib/Target/R600/AMDGPUInstrInfo.td
@@ -73,6 +73,13 @@ def AMDGPUregister_store : SDNode<"AMDGPUISD::REGISTER_STORE",
                            SDTypeProfile<0, 3, [SDTCisPtrTy<1>, SDTCisInt<2>]>,
                            [SDNPHasChain, SDNPMayStore]>;
 
+// MSKOR instructions are atomic memory instructions used mainly for storing
+// 8-bit and 16-bit values.  The definition is:
+//
+// MSKOR(dst, mask, src) MEM[dst] = ((MEM[dst] & ~mask) | src)
+//
+// src0: vec4(src, 0, 0, mask)
+// src1: dst - rat offset (aka pointer) in dwords  
 def AMDGPUstore_mskor : SDNode<"AMDGPUISD::STORE_MSKOR",
                         SDTypeProfile<0, 2, []>,
                         [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
diff --git a/lib/Target/R600/AMDGPUInstructions.td b/lib/Target/R600/AMDGPUInstructions.td
index df0bade2790..3227f944e8b 100644
--- a/lib/Target/R600/AMDGPUInstructions.td
+++ b/lib/Target/R600/AMDGPUInstructions.td
@@ -156,13 +156,23 @@ def truncstorei16_global : PatFrag<(ops node:$val, node:$ptr),
   return isGlobalStore(dyn_cast<StoreSDNode>(N));
 }]>;
 
-def local_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
-    return isLocalLoad(dyn_cast<LoadSDNode>(N));
-}]>;
-
 def local_store : PatFrag<(ops node:$val, node:$ptr),
                              (store node:$val, node:$ptr), [{
-    return isLocalStore(dyn_cast<StoreSDNode>(N));
+  return isLocalStore(dyn_cast<StoreSDNode>(N));
+}]>;
+
+def truncstorei8_local : PatFrag<(ops node:$val, node:$ptr),
+                                  (truncstorei8 node:$val, node:$ptr), [{
+  return isLocalStore(dyn_cast<StoreSDNode>(N));
+}]>;
+
+def truncstorei16_local : PatFrag<(ops node:$val, node:$ptr),
+                                  (truncstorei16 node:$val, node:$ptr), [{
+  return isLocalStore(dyn_cast<StoreSDNode>(N));
+}]>;
+
+def local_load : PatFrag<(ops node:$ptr), (load node:$ptr), [{
+    return isLocalLoad(dyn_cast<LoadSDNode>(N));
 }]>;
 
 def mskor_global : PatFrag<(ops node:$val, node:$ptr),
diff --git a/lib/Target/R600/R600Defines.h b/lib/Target/R600/R600Defines.h
index 8dc9ebb0cc2..1781f2aee16 100644
--- a/lib/Target/R600/R600Defines.h
+++ b/lib/Target/R600/R600Defines.h
@@ -45,7 +45,8 @@ namespace R600_InstFlag {
     ALU_INST = (1 << 14),
     LDS_1A = (1 << 15),
     LDS_1A1D = (1 << 16),
-    IS_EXPORT = (1 << 17)
+    IS_EXPORT = (1 << 17),
+    LDS_1A2D = (1 << 18)
   };
 }
 
diff --git a/lib/Target/R600/R600InstrFormats.td b/lib/Target/R600/R600InstrFormats.td
index 2ae33114112..ae3046d62e1 100644
--- a/lib/Target/R600/R600InstrFormats.td
+++ b/lib/Target/R600/R600InstrFormats.td
@@ -30,6 +30,7 @@ class InstR600 <dag outs, dag ins, string asm, list<dag> pattern,
   bit TEXInst = 0;
   bit ALUInst = 0;
   bit IsExport = 0;
+  bit LDS_1A2D = 0;
 
   let Namespace = "AMDGPU";
   let OutOperandList = outs;
@@ -55,6 +56,7 @@ class InstR600 <dag outs, dag ins, string asm, list<dag> pattern,
   let TSFlags{15} = LDS_1A;
   let TSFlags{16} = LDS_1A1D;
   let TSFlags{17} = IsExport;
+  let TSFlags{18} = LDS_1A2D;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/R600/R600InstrInfo.cpp b/lib/Target/R600/R600InstrInfo.cpp
index 9548a34104c..4e0607fb85e 100644
--- a/lib/Target/R600/R600InstrInfo.cpp
+++ b/lib/Target/R600/R600InstrInfo.cpp
@@ -149,7 +149,8 @@ bool R600InstrInfo::isLDSInstr(unsigned Opcode) const {
   unsigned TargetFlags = get(Opcode).TSFlags;
 
   return ((TargetFlags & R600_InstFlag::LDS_1A) |
-          (TargetFlags & R600_InstFlag::LDS_1A1D));
+          (TargetFlags & R600_InstFlag::LDS_1A1D) |
+          (TargetFlags & R600_InstFlag::LDS_1A2D));
 }
 
 bool R600InstrInfo::isTransOnly(unsigned Opcode) const {
diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
index b059a8179b7..3d92278c3a6 100644
--- a/lib/Target/R600/R600Instructions.td
+++ b/lib/Target/R600/R600Instructions.td
@@ -1657,13 +1657,31 @@ class R600_LDS_1A1D <bits<6> lds_op, string name, list<dag> pattern> :
   let LDS_1A1D = 1;
 }
 
-def LDS_READ_RET : R600_LDS_1A <0x32, "LDS_READ_RET",
-  [(set (i32 R600_Reg32:$dst), (local_load R600_Reg32:$src0))]
->;
+class R600_LDS_1A2D <bits<6> lds_op, string name, list<dag> pattern> :
+    R600_LDS <
+  lds_op,
+  (outs),
+  (ins R600_Reg32:$src0, REL:$src0_rel, SEL:$src0_sel,
+       R600_Reg32:$src1, REL:$src1_rel, SEL:$src1_sel,
+       R600_Reg32:$src2, REL:$src2_rel, SEL:$src2_sel,
+       LAST:$last, R600_Pred:$pred_sel, BANK_SWIZZLE:$bank_swizzle),
+  "  "#name# "$last $src0$src0_rel, $src1$src1_rel, $src2$src2_rel, $pred_sel",
+  pattern> {
+  let LDS_1A2D = 1;
+}
 
 def LDS_WRITE : R600_LDS_1A1D <0xD, "LDS_WRITE",
   [(local_store (i32 R600_Reg32:$src1), R600_Reg32:$src0)]
 >;
+def LDS_BYTE_WRITE : R600_LDS_1A1D<0x12, "LDS_BYTE_WRITE",
+  [(truncstorei8_local i32:$src1, i32:$src0)]
+>;
+def LDS_SHORT_WRITE : R600_LDS_1A1D<0x13, "LDS_SHORT_WRITE",
+  [(truncstorei16_local i32:$src1, i32:$src0)]
+>;
+def LDS_READ_RET : R600_LDS_1A <0x32, "LDS_READ_RET",
+  [(set (i32 R600_Reg32:$dst), (local_load R600_Reg32:$src0))]
+>;
 
   // TRUNC is used for the FLT_TO_INT instructions to work around a
   // perceived problem where the rounding modes are applied differently
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 8c52a2ea46c..785dbf17e25 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -392,6 +392,8 @@ defm V_CMPX_CLASS_F64 : VOPC_64 <0x000000b8, "V_CMPX_CLASS_F64">;
 } // End isCompare = 1
 
 def DS_WRITE_B32 : DS_Store_Helper <0x0000000d, "DS_WRITE_B32", VReg_32>;
+def DS_WRITE_B8 : DS_Store_Helper <0x00000001e, "DS_WRITE_B8", VReg_32>;
+def DS_WRITE_B16 : DS_Store_Helper <0x00000001f, "DS_WRITE_B16", VReg_32>;
 def DS_READ_B32 : DS_Load_Helper <0x00000036, "DS_READ_B32", VReg_32>;
 
 //def BUFFER_LOAD_FORMAT_X : MUBUF_ <0x00000000, "BUFFER_LOAD_FORMAT_X", []>;
@@ -1750,11 +1752,15 @@ def : Pat <
     (i32 (DS_READ_B32 0, $src0, $src0, $src0, 0, 0))
 >;
 
-def : Pat <
-    (local_store i32:$src1, i32:$src0),
-    (DS_WRITE_B32 0, $src0, $src1, $src1, 0, 0)
+class DSWritePat <DS inst, ValueType vt, PatFrag frag> : Pat <
+  (frag i32:$src1, i32:$src0),
+  (inst 0, $src0, $src1, $src1, 0, 0)
 >;
 
+def : DSWritePat <DS_WRITE_B8, i32, truncstorei8_local>;
+def : DSWritePat <DS_WRITE_B16, i32, truncstorei16_local>;
+def : DSWritePat <DS_WRITE_B32, i32, local_store>;
+
 /********** ================== **********/
 /**********   SMRD Patterns    **********/
 /********** ================== **********/
diff --git a/test/CodeGen/R600/store.ll b/test/CodeGen/R600/store.ll
index 9c80798eeae..50c91b42fa1 100644
--- a/test/CodeGen/R600/store.ll
+++ b/test/CodeGen/R600/store.ll
@@ -172,6 +172,24 @@ entry:
 ; Local Address Space
 ;===------------------------------------------------------------------------===;
 
+; EG-CHECK: @store_local_i8
+; EG-CHECK: LDS_BYTE_WRITE
+; SI-CHECK: @store_local_i8
+; SI-CHECK: DS_WRITE_B8
+define void @store_local_i8(i8 addrspace(3)* %out, i8 %in) {
+  store i8 %in, i8 addrspace(3)* %out
+  ret void
+}
+
+; EG-CHECK: @store_local_i16
+; EG-CHECK: LDS_SHORT_WRITE
+; SI-CHECK: @store_local_i16
+; SI-CHECK: DS_WRITE_B16
+define void @store_local_i16(i16 addrspace(3)* %out, i16 %in) {
+  store i16 %in, i16 addrspace(3)* %out
+  ret void
+}
+
 ; EG-CHECK: @store_local_v2i16
 ; EG-CHECK: LDS_WRITE
 ; CM-CHECK: @store_local_v2i16