diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index cfecd29853b7..92ee8626e4e6 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -533,10 +533,6 @@ bool AMDGPUDAGToDAGISel::isGlobalLoad(const MemSDNode *N) const {
     if (Subtarget->getGeneration() < AMDGPUSubtarget::SOUTHERN_ISLANDS)
       return !isa<GlobalValue>(GetUnderlyingObject(
           N->getMemOperand()->getValue(), CurDAG->getDataLayout()));
-
-    //TODO: Why do we need this?
-    if (N->getMemoryVT().bitsLT(MVT::i32))
-      return true;
   }
 
   return checkType(N->getMemOperand()->getValue(), AMDGPUAS::GLOBAL_ADDRESS);
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index a113ca2a25af..1e5a63cd42e7 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -170,17 +170,21 @@ def atomic_flat_store: flat_st <atomic_store>;
 def flat_truncstorei8 : flat_st <truncstorei8>;
 def flat_truncstorei16 : flat_st <truncstorei16>;
 
+class MubufLoad <SDPatternOperator op> : PatFrag <
+  (ops node:$ptr), (op node:$ptr), [{
 
-def mubuf_load : PatFrag <(ops node:$ptr), (load node:$ptr), [{
-	return isGlobalLoad(cast<LoadSDNode>(N)) ||
-         isConstantLoad(cast<LoadSDNode>(N), -1);
+  const MemSDNode *LD = cast<MemSDNode>(N);
+  return LD->getAddressSpace() == AMDGPUAS::GLOBAL_ADDRESS ||
+         LD->getAddressSpace() == AMDGPUAS::CONSTANT_ADDRESS;
 }]>;
 
-def mubuf_load_atomic : PatFrag <(ops node:$ptr), (atomic_load node:$ptr), [{
-	return isGlobalLoad(cast<MemSDNode>(N)) ||
-         isConstantLoad(cast<MemSDNode>(N), -1);
-}]>;
+def mubuf_load : MubufLoad <load>;
+def mubuf_az_extloadi8 : MubufLoad <az_extloadi8>;
+def mubuf_sextloadi8 : MubufLoad <sextloadi8>;
+def mubuf_az_extloadi16 : MubufLoad <az_extloadi16>;
+def mubuf_sextloadi16 : MubufLoad <sextloadi16>;
 
+def mubuf_load_atomic : MubufLoad <atomic_load>;
 
 def smrd_load : PatFrag <(ops node:$ptr), (load node:$ptr), [{
   auto Ld = cast<LoadSDNode>(N);
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 1f7c7ee0e863..93098f65a09a 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -953,16 +953,16 @@ defm BUFFER_STORE_FORMAT_XYZW : MUBUF_Store_Helper <
   mubuf<0x07>, "buffer_store_format_xyzw", VReg_128
 >;
 defm BUFFER_LOAD_UBYTE : MUBUF_Load_Helper <
-  mubuf<0x08, 0x10>, "buffer_load_ubyte", VGPR_32, i32, az_extloadi8_global
+  mubuf<0x08, 0x10>, "buffer_load_ubyte", VGPR_32, i32, mubuf_az_extloadi8
 >;
 defm BUFFER_LOAD_SBYTE : MUBUF_Load_Helper <
-  mubuf<0x09, 0x11>, "buffer_load_sbyte", VGPR_32, i32, sextloadi8_global
+  mubuf<0x09, 0x11>, "buffer_load_sbyte", VGPR_32, i32, mubuf_sextloadi8
 >;
 defm BUFFER_LOAD_USHORT : MUBUF_Load_Helper <
-  mubuf<0x0a, 0x12>, "buffer_load_ushort", VGPR_32, i32, az_extloadi16_global
+  mubuf<0x0a, 0x12>, "buffer_load_ushort", VGPR_32, i32, mubuf_az_extloadi16
 >;
 defm BUFFER_LOAD_SSHORT : MUBUF_Load_Helper <
-  mubuf<0x0b, 0x13>, "buffer_load_sshort", VGPR_32, i32, sextloadi16_global
+  mubuf<0x0b, 0x13>, "buffer_load_sshort", VGPR_32, i32, mubuf_sextloadi16
 >;
 defm BUFFER_LOAD_DWORD : MUBUF_Load_Helper <
   mubuf<0x0c, 0x14>, "buffer_load_dword", VGPR_32, i32, mubuf_load