R600: Fix handling of vector kernel arguments

The SelectionDAGBuilder was promoting vector kernel arguments to legal
types, but this won't work for R600 and SI since kernel arguments are
stored in memory and can't be promoted.  In order to handle vector
arguments correctly we need to look at the original types from the LLVM IR
function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193215 91177308-0d34-0410-b5e6-96231b3b80d8

14 files changed, 633 insertions, 150 deletions

diff --git a/lib/Target/R600/AMDGPUCallingConv.td b/lib/Target/R600/AMDGPUCallingConv.td
index a194e6d3c8..3535e35f4f 100644
--- a/lib/Target/R600/AMDGPUCallingConv.td
+++ b/lib/Target/R600/AMDGPUCallingConv.td
@@ -44,11 +44,7 @@ def CC_SI : CallingConv<[
 
 // Calling convention for compute kernels
 def CC_AMDGPU_Kernel : CallingConv<[
-  CCIfType<[v4i32, v4f32],               CCAssignToStack <16, 16>>,
-  CCIfType<[i64, f64, v2f32, v2i32],     CCAssignToStack < 8, 8>>,
-  CCIfType<[i32, f32],                   CCAssignToStack < 4, 4>>,
-  CCIfType<[i16],                        CCAssignToStack < 2, 4>>,
-  CCIfType<[i8],                         CCAssignToStack < 1, 4>>
+  CCCustom<"allocateStack">
 ]>;
 
 def CC_AMDGPU : CallingConv<[
diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index f6c074a4cd..b442c25473 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -28,6 +28,14 @@
 #include "llvm/IR/DataLayout.h"
 
 using namespace llvm;
+static bool allocateStack(unsigned ValNo, MVT ValVT, MVT LocVT,
+                      CCValAssign::LocInfo LocInfo,
+                      ISD::ArgFlagsTy ArgFlags, CCState &State) {
+  unsigned Offset = State.AllocateStack(ValVT.getSizeInBits() / 8, ArgFlags.getOrigAlign());
+    State.addLoc(CCValAssign::getMem(ValNo, ValVT, Offset, LocVT, LocInfo));
+
+  return true;
+}
 
 #include "AMDGPUGenCallingConv.inc"
 
@@ -64,6 +72,12 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::STORE, MVT::v4f32, Promote);
   AddPromotedToType(ISD::STORE, MVT::v4f32, MVT::v4i32);
 
+  setOperationAction(ISD::STORE, MVT::v8f32, Promote);
+  AddPromotedToType(ISD::STORE, MVT::v8f32, MVT::v8i32);
+
+  setOperationAction(ISD::STORE, MVT::v16f32, Promote);
+  AddPromotedToType(ISD::STORE, MVT::v16f32, MVT::v16i32);
+
   setOperationAction(ISD::STORE, MVT::f64, Promote);
   AddPromotedToType(ISD::STORE, MVT::f64, MVT::i64);
 
@@ -90,6 +104,12 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::LOAD, MVT::v4f32, Promote);
   AddPromotedToType(ISD::LOAD, MVT::v4f32, MVT::v4i32);
 
+  setOperationAction(ISD::LOAD, MVT::v8f32, Promote);
+  AddPromotedToType(ISD::LOAD, MVT::v8f32, MVT::v8i32);
+
+  setOperationAction(ISD::LOAD, MVT::v16f32, Promote);
+  AddPromotedToType(ISD::LOAD, MVT::v16f32, MVT::v16i32);
+
   setOperationAction(ISD::LOAD, MVT::f64, Promote);
   AddPromotedToType(ISD::LOAD, MVT::f64, MVT::i64);
 
@@ -656,6 +676,38 @@ SDValue AMDGPUTargetLowering::LowerUDIVREM(SDValue Op,
 // Helper functions
 //===----------------------------------------------------------------------===//
 
+void AMDGPUTargetLowering::getOriginalFunctionArgs(
+                               SelectionDAG &DAG,
+                               const Function *F,
+                               const SmallVectorImpl<ISD::InputArg> &Ins,
+                               SmallVectorImpl<ISD::InputArg> &OrigIns) const {
+
+  for (unsigned i = 0, e = Ins.size(); i < e; ++i) {
+    if (Ins[i].ArgVT == Ins[i].VT) {
+      OrigIns.push_back(Ins[i]);
+      continue;
+    }
+
+    EVT VT;
+    if (Ins[i].ArgVT.isVector() && !Ins[i].VT.isVector()) {
+      // Vector has been split into scalars.
+      VT = Ins[i].ArgVT.getVectorElementType();
+    } else if (Ins[i].VT.isVector() && Ins[i].ArgVT.isVector() &&
+               Ins[i].ArgVT.getVectorElementType() !=
+               Ins[i].VT.getVectorElementType()) {
+      // Vector elements have been promoted
+      VT = Ins[i].ArgVT;
+    } else {
+      // Vector has been spilt into smaller vectors.
+      VT = Ins[i].VT;
+    }
+
+    ISD::InputArg Arg(Ins[i].Flags, VT, VT, Ins[i].Used,
+                      Ins[i].OrigArgIndex, Ins[i].PartOffset);
+    OrigIns.push_back(Arg);
+  }
+}
+
 bool AMDGPUTargetLowering::isHWTrueValue(SDValue Op) const {
   if (ConstantFPSDNode * CFP = dyn_cast<ConstantFPSDNode>(Op)) {
     return CFP->isExactlyValue(1.0);
diff --git a/lib/Target/R600/AMDGPUISelLowering.h b/lib/Target/R600/AMDGPUISelLowering.h
index 8a68356c2e..e383016722 100644
--- a/lib/Target/R600/AMDGPUISelLowering.h
+++ b/lib/Target/R600/AMDGPUISelLowering.h
@@ -36,7 +36,6 @@ private:
   SDValue MergeVectorStore(const SDValue &Op, SelectionDAG &DAG) const;
   /// \brief Split a vector store into multiple scalar stores.
   /// \returns The resulting chain. 
-  SDValue SplitVectorStore(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerUDIVREM(SDValue Op, SelectionDAG &DAG) const;
 
 protected:
@@ -52,10 +51,21 @@ protected:
                              SelectionDAG &DAG) const;
   /// \brief Split a vector load into multiple scalar loads.
   SDValue SplitVectorLoad(const SDValue &Op, SelectionDAG &DAG) const;
+  SDValue SplitVectorStore(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
   bool isHWTrueValue(SDValue Op) const;
   bool isHWFalseValue(SDValue Op) const;
 
+  /// The SelectionDAGBuilder will automatically promote function arguments
+  /// with illegal types.  However, this does not work for the AMDGPU targets
+  /// since the function arguments are stored in memory as these illegal types.
+  /// In order to handle this properly we need to get the origianl types sizes
+  /// from the LLVM IR Function and fixup the ISD:InputArg values before
+  /// passing them to AnalyzeFormalArguments()
+  void getOriginalFunctionArgs(SelectionDAG &DAG,
+                               const Function *F,
+                               const SmallVectorImpl<ISD::InputArg> &Ins,
+                               SmallVectorImpl<ISD::InputArg> &OrigIns) const;
   void AnalyzeFormalArguments(CCState &State,
                               const SmallVectorImpl<ISD::InputArg> &Ins) const;
 
diff --git a/lib/Target/R600/AMDGPUTargetMachine.cpp b/lib/Target/R600/AMDGPUTargetMachine.cpp
index 9722e7dddf..b19277d97b 100644
--- a/lib/Target/R600/AMDGPUTargetMachine.cpp
+++ b/lib/Target/R600/AMDGPUTargetMachine.cpp
@@ -59,8 +59,9 @@ AMDGPUTargetMachine::AMDGPUTargetMachine(const Target &T, StringRef TT,
   LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OptLevel),
   Subtarget(TT, CPU, FS),
   Layout(Subtarget.getDataLayout()),
-  FrameLowering(TargetFrameLowering::StackGrowsUp, 16 // Stack Alignment
-                                                 , 0),
+  FrameLowering(TargetFrameLowering::StackGrowsUp,
+                64 * 16 // Maximum stack alignment (long16)
+               , 0),
   IntrinsicInfo(this),
   InstrItins(&Subtarget.getInstrItineraryData()) {
   // TLInfo uses InstrInfo so it must be initialized after.
diff --git a/lib/Target/R600/R600ISelLowering.cpp b/lib/Target/R600/R600ISelLowering.cpp
index 3c2e3888e0..26041eda18 100644
--- a/lib/Target/R600/R600ISelLowering.cpp
+++ b/lib/Target/R600/R600ISelLowering.cpp
@@ -1209,7 +1209,7 @@ SDValue R600TargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const
   }
 
   int ConstantBlock = ConstantAddressBlock(LoadNode->getAddressSpace());
-  if (ConstantBlock > -1) {
+  if (ConstantBlock > -1 && LoadNode->getExtensionType() != ISD::SEXTLOAD) {
     SDValue Result;
     if (dyn_cast<ConstantExpr>(LoadNode->getSrcValue()) ||
         dyn_cast<Constant>(LoadNode->getSrcValue()) ||
@@ -1340,22 +1340,29 @@ SDValue R600TargetLowering::LowerFormalArguments(
   CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
                  getTargetMachine(), ArgLocs, *DAG.getContext());
 
-  AnalyzeFormalArguments(CCInfo, Ins);
+  SmallVector<ISD::InputArg, 8> LocalIns;
+
+  getOriginalFunctionArgs(DAG, DAG.getMachineFunction().getFunction(), Ins,
+                          LocalIns);
+
+  AnalyzeFormalArguments(CCInfo, LocalIns);
 
   for (unsigned i = 0, e = Ins.size(); i < e; ++i) {
     CCValAssign &VA = ArgLocs[i];
-    EVT VT = VA.getLocVT();
+    EVT VT = Ins[i].VT;
+    EVT MemVT = LocalIns[i].VT;
 
     PointerType *PtrTy = PointerType::get(VT.getTypeForEVT(*DAG.getContext()),
                                                    AMDGPUAS::CONSTANT_BUFFER_0);
 
     // The first 36 bytes of the input buffer contains information about
     // thread group and global sizes.
-    SDValue Arg = DAG.getLoad(VT, DL, Chain,
-                           DAG.getConstant(36 + VA.getLocMemOffset(), MVT::i32),
-                           MachinePointerInfo(UndefValue::get(PtrTy)), false,
-                           false, false, 4); // 4 is the prefered alignment for
-                                             // the CONSTANT memory space.
+    SDValue Arg = DAG.getExtLoad(ISD::SEXTLOAD, DL, VT, Chain,
+                                 DAG.getConstant(36 + VA.getLocMemOffset(), MVT::i32),
+                                 MachinePointerInfo(UndefValue::get(PtrTy)),
+                                 MemVT, false, false, 4);
+                                 // 4 is the prefered alignment for
+                                 // the CONSTANT memory space.
     InVals.push_back(Arg);
   }
   return Chain;
diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
index 2c9270e00f..fe08717a33 100644
--- a/lib/Target/R600/SIISelLowering.cpp
+++ b/lib/Target/R600/SIISelLowering.cpp
@@ -69,6 +69,11 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
   // We need to custom lower vector stores from local memory
   setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
+  setOperationAction(ISD::LOAD, MVT::v8i32, Custom);
+  setOperationAction(ISD::LOAD, MVT::v16i32, Custom);
+
+  setOperationAction(ISD::STORE, MVT::v8i32, Custom);
+  setOperationAction(ISD::STORE, MVT::v16i32, Custom);
 
   setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
   setOperationAction(ISD::SELECT_CC, MVT::i32, Custom);
@@ -78,6 +83,7 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::SETCC, MVT::v2i1, Expand);
   setOperationAction(ISD::SETCC, MVT::v4i1, Expand);
 
+  setOperationAction(ISD::ANY_EXTEND, MVT::i64, Custom);
   setOperationAction(ISD::SIGN_EXTEND, MVT::i64, Custom);
   setOperationAction(ISD::ZERO_EXTEND, MVT::i64, Custom);
 
@@ -89,10 +95,15 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
 
   setLoadExtAction(ISD::SEXTLOAD, MVT::i32, Expand);
+  setLoadExtAction(ISD::SEXTLOAD, MVT::v8i16, Expand);
+  setLoadExtAction(ISD::SEXTLOAD, MVT::v16i16, Expand);
 
   setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
   setTruncStoreAction(MVT::i64, MVT::i32, Expand);
+  setTruncStoreAction(MVT::i128, MVT::i64, Expand);
+  setTruncStoreAction(MVT::v8i32, MVT::v8i16, Expand);
+  setTruncStoreAction(MVT::v16i32, MVT::v16i16, Expand);
 
   setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
 
@@ -115,23 +126,22 @@ bool SITargetLowering::allowsUnalignedMemoryAccesses(EVT  VT,
 }
 
 bool SITargetLowering::shouldSplitVectorElementType(EVT VT) const {
-  return VT.bitsLE(MVT::i8);
+  return VT.bitsLE(MVT::i16);
 }
 
-SDValue SITargetLowering::LowerParameter(SelectionDAG &DAG, EVT VT,
+SDValue SITargetLowering::LowerParameter(SelectionDAG &DAG, EVT VT, EVT MemVT,
                                          SDLoc DL, SDValue Chain,
                                          unsigned Offset) const {
   MachineRegisterInfo &MRI = DAG.getMachineFunction().getRegInfo();
   PointerType *PtrTy = PointerType::get(VT.getTypeForEVT(*DAG.getContext()),
                                             AMDGPUAS::CONSTANT_ADDRESS);
-  EVT ArgVT = MVT::getIntegerVT(VT.getSizeInBits());
   SDValue BasePtr =  DAG.getCopyFromReg(Chain, DL,
                            MRI.getLiveInVirtReg(AMDGPU::SGPR0_SGPR1), MVT::i64);
   SDValue Ptr = DAG.getNode(ISD::ADD, DL, MVT::i64, BasePtr,
                                              DAG.getConstant(Offset, MVT::i64));
-  return DAG.getLoad(VT, DL, Chain, Ptr,
-                            MachinePointerInfo(UndefValue::get(PtrTy)),
-                            false, false, false, ArgVT.getSizeInBits() >> 3);
+  return DAG.getExtLoad(ISD::SEXTLOAD, DL, VT, Chain, Ptr,
+                            MachinePointerInfo(UndefValue::get(PtrTy)), MemVT,
+                            false, false, MemVT.getSizeInBits() >> 3);
 
 }
 
@@ -190,7 +200,7 @@ SDValue SITargetLowering::LowerFormalArguments(
         NewArg.PartOffset += NewArg.VT.getStoreSize();
       }
 
-    } else {
+    } else if (Info->ShaderType != ShaderType::COMPUTE) {
       Splits.push_back(Arg);
     }
   }
@@ -213,6 +223,11 @@ SDValue SITargetLowering::LowerFormalArguments(
     MF.addLiveIn(AMDGPU::SGPR0_SGPR1, &AMDGPU::SReg_64RegClass);
   }
 
+  if (Info->ShaderType == ShaderType::COMPUTE) {
+    getOriginalFunctionArgs(DAG, DAG.getMachineFunction().getFunction(), Ins,
+                            Splits);
+  }
+
   AnalyzeFormalArguments(CCInfo, Splits);
 
   for (unsigned i = 0, e = Ins.size(), ArgIdx = 0; i != e; ++i) {
@@ -227,9 +242,11 @@ SDValue SITargetLowering::LowerFormalArguments(
     EVT VT = VA.getLocVT();
 
     if (VA.isMemLoc()) {
+      VT = Ins[i].VT;
+      EVT MemVT = Splits[i].VT;
       // The first 36 bytes of the input buffer contains information about
       // thread group and global sizes.
-      SDValue Arg = LowerParameter(DAG, VT, DL, DAG.getRoot(),
+      SDValue Arg = LowerParameter(DAG, VT, MemVT,  DL, DAG.getRoot(),
                                    36 + VA.getLocMemOffset());
       InVals.push_back(Arg);
       continue;
@@ -389,8 +406,18 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
       return SDValue();
     }
   }
+  case ISD::STORE: {
+    StoreSDNode *Store = dyn_cast<StoreSDNode>(Op);
+    if (Store->getValue().getValueType().isVector() &&
+        Store->getValue().getValueType().getVectorNumElements() >= 8)
+      return SplitVectorStore(Op, DAG);
+    else
+      return AMDGPUTargetLowering::LowerOperation(Op, DAG);
+  }
+
   case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
   case ISD::SIGN_EXTEND: return LowerSIGN_EXTEND(Op, DAG);
+  case ISD::ANY_EXTEND: // Fall-through
   case ISD::ZERO_EXTEND: return LowerZERO_EXTEND(Op, DAG);
   case ISD::GlobalAddress: return LowerGlobalAddress(MFI, Op, DAG);
   case ISD::INTRINSIC_WO_CHAIN: {
@@ -403,23 +430,23 @@ SDValue SITargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
     switch (IntrinsicID) {
     default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
     case Intrinsic::r600_read_ngroups_x:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 0);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 0);
     case Intrinsic::r600_read_ngroups_y:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 4);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 4);
     case Intrinsic::r600_read_ngroups_z:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 8);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 8);
     case Intrinsic::r600_read_global_size_x:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 12);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 12);
     case Intrinsic::r600_read_global_size_y:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 16);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 16);
     case Intrinsic::r600_read_global_size_z:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 20);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 20);
     case Intrinsic::r600_read_local_size_x:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 24);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 24);
     case Intrinsic::r600_read_local_size_y:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 28);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 28);
     case Intrinsic::r600_read_local_size_z:
-      return LowerParameter(DAG, VT, DL, DAG.getEntryNode(), 32);
+      return LowerParameter(DAG, VT, VT, DL, DAG.getEntryNode(), 32);
     case Intrinsic::r600_read_tgid_x:
       return CreateLiveInRegister(DAG, &AMDGPU::SReg_32RegClass,
                      AMDGPU::SReg_32RegClass.getRegister(NumUserSGPRs + 0), VT);
diff --git a/lib/Target/R600/SIISelLowering.h b/lib/Target/R600/SIISelLowering.h
index 9c54a6f48a..1ac6e0dfcd 100644
--- a/lib/Target/R600/SIISelLowering.h
+++ b/lib/Target/R600/SIISelLowering.h
@@ -21,7 +21,7 @@
 namespace llvm {
 
 class SITargetLowering : public AMDGPUTargetLowering {
-  SDValue LowerParameter(SelectionDAG &DAG, EVT VT, SDLoc DL,
+  SDValue LowerParameter(SelectionDAG &DAG, EVT VT, EVT MemVT, SDLoc DL,
                          SDValue Chain, unsigned Offset) const;
   SDValue LowerSampleIntrinsic(unsigned Opcode, const SDValue &Op,
                                SelectionDAG &DAG) const;
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index ab6193fa4c..92a53ddeb9 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -1560,9 +1560,12 @@ def : BitConvert <f64, i64, VReg_64>;
 
 def : BitConvert <v2f32, v2i32, VReg_64>;
 def : BitConvert <v2i32, v2f32, VReg_64>;
+def : BitConvert <v2i32, i64, VReg_64>;
 
 def : BitConvert <v4f32, v4i32, VReg_128>;
 def : BitConvert <v4i32, v4f32, VReg_128>;
+def : BitConvert <v4i32, i128,  VReg_128>;
+def : BitConvert <i128, v4i32,  VReg_128>;
 
 def : BitConvert <v8i32, v32i8, SReg_256>;
 def : BitConvert <v32i8, v8i32, SReg_256>;
diff --git a/lib/Target/R600/SIRegisterInfo.td b/lib/Target/R600/SIRegisterInfo.td
index c8e32953c7..49bdbc9fdb 100644
--- a/lib/Target/R600/SIRegisterInfo.td
+++ b/lib/Target/R600/SIRegisterInfo.td
@@ -174,7 +174,7 @@ def VReg_96 : RegisterClass<"AMDGPU", [untyped], 96, (add VGPR_96)> {
   let Size = 96;
 }
 
-def VReg_128 : RegisterClass<"AMDGPU", [v4i32, v4f32], 128, (add VGPR_128)>;
+def VReg_128 : RegisterClass<"AMDGPU", [v4i32, v4f32, i128], 128, (add VGPR_128)>;
 
 def VReg_256 : RegisterClass<"AMDGPU", [v32i8, v8i32, v8f32], 256, (add VGPR_256)>;
 
diff --git a/test/CodeGen/R600/kernel-args.ll b/test/CodeGen/R600/kernel-args.ll
new file mode 100644
index 0000000000..8b80e26011
--- /dev/null
+++ b/test/CodeGen/R600/kernel-args.ll
@@ -0,0 +1,455 @@
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=EG-CHECK
+; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s --check-prefix=EG-CHECK
+; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck %s --check-prefix=SI-CHECK
+
+; EG-CHECK-LABEL: @i8_arg
+; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
+; SI-CHECK-LABEL: @i8_arg
+; SI-CHECK: BUFFER_LOAD_UBYTE
+
+define void @i8_arg(i32 addrspace(1)* nocapture %out, i8 %in) nounwind {
+entry:
+  %0 = zext i8 %in to i32
+  store i32 %0, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @i8_zext_arg
+; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
+; SI-CHECK-LABEL: @i8_zext_arg
+; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
+
+define void @i8_zext_arg(i32 addrspace(1)* nocapture %out, i8 zeroext %in) nounwind {
+entry:
+  %0 = zext i8 %in to i32
+  store i32 %0, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @i8_sext_arg
+; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
+; SI-CHECK-LABEL: @i8_sext_arg
+; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
+
+define void @i8_sext_arg(i32 addrspace(1)* nocapture %out, i8 signext %in) nounwind {
+entry:
+  %0 = sext i8 %in to i32
+  store i32 %0, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @i16_arg
+; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
+; SI-CHECK-LABEL: @i16_arg
+; SI-CHECK: BUFFER_LOAD_USHORT
+
+define void @i16_arg(i32 addrspace(1)* nocapture %out, i16 %in) nounwind {
+entry:
+  %0 = zext i16 %in to i32
+  store i32 %0, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @i16_zext_arg
+; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
+; SI-CHECK-LABEL: @i16_zext_arg
+; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
+
+define void @i16_zext_arg(i32 addrspace(1)* nocapture %out, i16 zeroext %in) nounwind {
+entry:
+  %0 = zext i16 %in to i32
+  store i32 %0, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @i16_sext_arg
+; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
+; SI-CHECK-LABEL: @i16_sext_arg
+; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
+
+define void @i16_sext_arg(i32 addrspace(1)* nocapture %out, i16 signext %in) nounwind {
+entry:
+  %0 = sext i16 %in to i32
+  store i32 %0, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @i32_arg
+; EG-CHECK: T{{[0-9]\.[XYZW]}}, KC0[2].Z
+; SI-CHECK-LABEL: @i32_arg
+; S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
+define void @i32_arg(i32 addrspace(1)* nocapture %out, i32 %in) nounwind {
+entry:
+  store i32 %in, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @f32_arg
+; EG-CHECK: T{{[0-9]\.[XYZW]}}, KC0[2].Z
+; SI-CHECK-LABEL: @f32_arg
+; S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
+define void @f32_arg(float addrspace(1)* nocapture %out, float %in) nounwind {
+entry:
+  store float %in, float addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v2i8_arg
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; SI-CHECK-LABEL: @v2i8_arg
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+define void @v2i8_arg(<2 x i8> addrspace(1)* %out, <2 x i8> %in) {
+entry:
+  store <2 x i8> %in, <2 x i8> addrspace(1)* %out
+  ret void
+}
+
+; EG-CHECK-LABEL: @v2i16_arg
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; SI-CHECK-LABEL: @v2i16_arg
+; SI-CHECK-DAG: BUFFER_LOAD_USHORT
+; SI-CHECK-DAG: BUFFER_LOAD_USHORT
+define void @v2i16_arg(<2 x i16> addrspace(1)* %out, <2 x i16> %in) {
+entry:
+  store <2 x i16> %in, <2 x i16> addrspace(1)* %out
+  ret void
+}
+
+; EG-CHECK-LABEL: @v2i32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[2].W
+; SI-CHECK-LABEL: @v2i32_arg
+; SI-CHECK: S_LOAD_DWORDX2 SGPR{{[0-9]}}_SGPR{{[0-9]}}, SGPR0_SGPR1, 11
+define void @v2i32_arg(<2 x i32> addrspace(1)* nocapture %out, <2 x i32> %in) nounwind {
+entry:
+  store <2 x i32> %in, <2 x i32> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v2f32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[2].W
+; SI-CHECK-LABEL: @v2f32_arg
+; SI-CHECK: S_LOAD_DWORDX2 SGPR{{[0-9]}}_SGPR{{[0-9]}}, SGPR0_SGPR1, 11
+define void @v2f32_arg(<2 x float> addrspace(1)* nocapture %out, <2 x float> %in) nounwind {
+entry:
+  store <2 x float> %in, <2 x float> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v3i8_arg
+; VTX_READ_8 T{{[0-9]}}.X, T{{[0-9]}}.X, 40
+; VTX_READ_8 T{{[0-9]}}.X, T{{[0-9]}}.X, 41
+; VTX_READ_8 T{{[0-9]}}.X, T{{[0-9]}}.X, 42
+; SI-CHECK-LABEL: @v3i8_arg
+define void @v3i8_arg(<3 x i8> addrspace(1)* nocapture %out, <3 x i8> %in) nounwind {
+entry:
+  store <3 x i8> %in, <3 x i8> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v3i16_arg
+; VTX_READ_16 T{{[0-9]}}.X, T{{[0-9]}}.X, 44
+; VTX_READ_16 T{{[0-9]}}.X, T{{[0-9]}}.X, 46
+; VTX_READ_16 T{{[0-9]}}.X, T{{[0-9]}}.X, 48
+; SI-CHECK-LABEL: @v3i16_arg
+define void @v3i16_arg(<3 x i16> addrspace(1)* nocapture %out, <3 x i16> %in) nounwind {
+entry:
+  store <3 x i16> %in, <3 x i16> addrspace(1)* %out, align 4
+  ret void
+}
+; EG-CHECK-LABEL: @v3i32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].W
+; SI-CHECK-LABEL: @v3i32_arg
+; SI-CHECK: S_LOAD_DWORDX4 SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}, SGPR0_SGPR1, 13
+define void @v3i32_arg(<3 x i32> addrspace(1)* nocapture %out, <3 x i32> %in) nounwind {
+entry:
+  store <3 x i32> %in, <3 x i32> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v3f32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].W
+; SI-CHECK-LABEL: @v3f32_arg
+; SI-CHECK: S_LOAD_DWORDX4 SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}, SGPR0_SGPR1, 13
+define void @v3f32_arg(<3 x float> addrspace(1)* nocapture %out, <3 x float> %in) nounwind {
+entry:
+  store <3 x float> %in, <3 x float> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v4i8_arg
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; SI-CHECK-LABEL: @v4i8_arg
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+define void @v4i8_arg(<4 x i8> addrspace(1)* %out, <4 x i8> %in) {
+entry:
+  store <4 x i8> %in, <4 x i8> addrspace(1)* %out
+  ret void
+}
+
+; EG-CHECK-LABEL: @v4i16_arg
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; SI-CHECK-LABEL: @v4i16_arg
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+define void @v4i16_arg(<4 x i16> addrspace(1)* %out, <4 x i16> %in) {
+entry:
+  store <4 x i16> %in, <4 x i16> addrspace(1)* %out
+  ret void
+}
+
+; EG-CHECK-LABEL: @v4i32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].X
+; SI-CHECK-LABEL: @v4i32_arg
+; SI-CHECK: S_LOAD_DWORDX4 SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}, SGPR0_SGPR1, 13
+define void @v4i32_arg(<4 x i32> addrspace(1)* nocapture %out, <4 x i32> %in) nounwind {
+entry:
+  store <4 x i32> %in, <4 x i32> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v4f32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[3].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].X
+; SI-CHECK-LABEL: @v4f32_arg
+; SI-CHECK: S_LOAD_DWORDX4 SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}, SGPR0_SGPR1, 13
+define void @v4f32_arg(<4 x float> addrspace(1)* nocapture %out, <4 x float> %in) nounwind {
+entry:
+  store <4 x float> %in, <4 x float> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v8i8_arg
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; SI-CHECK-LABEL: @v8i8_arg
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+define void @v8i8_arg(<8 x i8> addrspace(1)* %out, <8 x i8> %in) {
+entry:
+  store <8 x i8> %in, <8 x i8> addrspace(1)* %out
+  ret void
+}
+
+; EG-CHECK-LABEL: @v8i16_arg
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; SI-CHECK-LABEL: @v8i16_arg
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+define void @v8i16_arg(<8 x i16> addrspace(1)* %out, <8 x i16> %in) {
+entry:
+  store <8 x i16> %in, <8 x i16> addrspace(1)* %out
+  ret void
+}
+
+; EG-CHECK-LABEL: @v8i32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[5].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[5].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[5].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[5].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].X
+; SI-CHECK-LABEL: @v8i32_arg
+; SI-CHECK: S_LOAD_DWORDX8 SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}, SGPR0_SGPR1, 17
+define void @v8i32_arg(<8 x i32> addrspace(1)* nocapture %out, <8 x i32> %in) nounwind {
+entry:
+  store <8 x i32> %in, <8 x i32> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v8f32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[4].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[5].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[5].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[5].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[5].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].X
+; SI-CHECK-LABEL: @v8f32_arg
+; SI-CHECK: S_LOAD_DWORDX8 SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}, SGPR0_SGPR1, 17
+define void @v8f32_arg(<8 x float> addrspace(1)* nocapture %out, <8 x float> %in) nounwind {
+entry:
+  store <8 x float> %in, <8 x float> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v16i8_arg
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; EG-CHECK: VTX_READ_8
+; SI-CHECK-LABEL: @v16i8_arg
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+; SI-CHECK: BUFFER_LOAD_UBYTE
+define void @v16i8_arg(<16 x i8> addrspace(1)* %out, <16 x i8> %in) {
+entry:
+  store <16 x i8> %in, <16 x i8> addrspace(1)* %out
+  ret void
+}
+
+; EG-CHECK-LABEL: @v16i16_arg
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; EG-CHECK: VTX_READ_16
+; SI-CHECK-LABEL: @v16i16_arg
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+; SI-CHECK: BUFFER_LOAD_USHORT
+define void @v16i16_arg(<16 x i16> addrspace(1)* %out, <16 x i16> %in) {
+entry:
+  store <16 x i16> %in, <16 x i16> addrspace(1)* %out
+  ret void
+}
+
+; EG-CHECK-LABEL: @v16i32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[7].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[7].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[7].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[7].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[8].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[8].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[8].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[8].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[9].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[9].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[9].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[9].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[10].X
+; SI-CHECK-LABEL: @v16i32_arg
+; SI-CHECK: S_LOAD_DWORDX16 SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}, SGPR0_SGPR1, 25
+define void @v16i32_arg(<16 x i32> addrspace(1)* nocapture %out, <16 x i32> %in) nounwind {
+entry:
+  store <16 x i32> %in, <16 x i32> addrspace(1)* %out, align 4
+  ret void
+}
+
+; EG-CHECK-LABEL: @v16f32_arg
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[6].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[7].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[7].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[7].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[7].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[8].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[8].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[8].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[8].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[9].X
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[9].Y
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[9].Z
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[9].W
+; EG-CHECK-DAG: T{{[0-9]\.[XYZW]}}, KC0[10].X
+; SI-CHECK-LABEL: @v16f32_arg
+; SI-CHECK: S_LOAD_DWORDX16 SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}_SGPR{{[0-9]+}}, SGPR0_SGPR1, 25
+define void @v16f32_arg(<16 x float> addrspace(1)* nocapture %out, <16 x float> %in) nounwind {
+entry:
+  store <16 x float> %in, <16 x float> addrspace(1)* %out, align 4
+  ret void
+}
diff --git a/test/CodeGen/R600/or.ll b/test/CodeGen/R600/or.ll
index 6950ed0dfb..6c70469e8d 100644
--- a/test/CodeGen/R600/or.ll
+++ b/test/CodeGen/R600/or.ll
@@ -41,7 +41,7 @@ define void @or_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in)
 
 ; EG-CHECK-LABEL: @or_i64
 ; EG-CHECK-DAG: OR_INT * T{{[0-9]\.[XYZW]}}, KC0[2].W, KC0[3].Y
-; EG-CHECK-DAG: OR_INT * T{{[0-9]\.[XYZW]}}, KC0[2].Z, KC0[3].X
+; EG-CHECK-DAG: OR_INT * T{{[0-9]\.[XYZW]}}, KC0[3].X, KC0[3].Z
 ; SI-CHECK-LABEL: @or_i64
 ; SI-CHECK: V_OR_B32_e32 VGPR{{[0-9]}}
 ; SI-CHECK: V_OR_B32_e32 VGPR{{[0-9]}}
diff --git a/test/CodeGen/R600/short-args.ll b/test/CodeGen/R600/short-args.ll
deleted file mode 100644
index 8882978d75..0000000000
--- a/test/CodeGen/R600/short-args.ll
+++ /dev/null
@@ -1,69 +0,0 @@
-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=EG-CHECK
-; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s --check-prefix=EG-CHECK
-; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck %s --check-prefix=SI-CHECK
-
-; EG-CHECK: @i8_arg
-; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
-; SI-CHECK: BUFFER_LOAD_UBYTE
-
-define void @i8_arg(i32 addrspace(1)* nocapture %out, i8 %in) nounwind {
-entry:
-  %0 = zext i8 %in to i32
-  store i32 %0, i32 addrspace(1)* %out, align 4
-  ret void
-}
-
-; EG-CHECK: @i8_zext_arg
-; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
-; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
-
-define void @i8_zext_arg(i32 addrspace(1)* nocapture %out, i8 zeroext %in) nounwind {
-entry:
-  %0 = zext i8 %in to i32
-  store i32 %0, i32 addrspace(1)* %out, align 4
-  ret void
-}
-
-; EG-CHECK: @i8_sext_arg
-; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
-; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
-
-define void @i8_sext_arg(i32 addrspace(1)* nocapture %out, i8 signext %in) nounwind {
-entry:
-  %0 = sext i8 %in to i32
-  store i32 %0, i32 addrspace(1)* %out, align 4
-  ret void
-}
-
-; EG-CHECK: @i16_arg
-; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
-; SI-CHECK: BUFFER_LOAD_USHORT
-
-define void @i16_arg(i32 addrspace(1)* nocapture %out, i16 %in) nounwind {
-entry:
-  %0 = zext i16 %in to i32
-  store i32 %0, i32 addrspace(1)* %out, align 4
-  ret void
-}
-
-; EG-CHECK: @i16_zext_arg
-; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
-; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
-
-define void @i16_zext_arg(i32 addrspace(1)* nocapture %out, i16 zeroext %in) nounwind {
-entry:
-  %0 = zext i16 %in to i32
-  store i32 %0, i32 addrspace(1)* %out, align 4
-  ret void
-}
-
-; EG-CHECK: @i16_sext_arg
-; EG-CHECK: MOV {{[ *]*}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z
-; SI-CHECK: S_LOAD_DWORD SGPR{{[0-9]}}, SGPR0_SGPR1, 11
-
-define void @i16_sext_arg(i32 addrspace(1)* nocapture %out, i16 signext %in) nounwind {
-entry:
-  %0 = sext i16 %in to i32
-  store i32 %0, i32 addrspace(1)* %out, align 4
-  ret void
-}
diff --git a/test/CodeGen/R600/store.ll b/test/CodeGen/R600/store.ll
index a4c025ab42..5e51d56917 100644
--- a/test/CodeGen/R600/store.ll
+++ b/test/CodeGen/R600/store.ll
@@ -7,7 +7,7 @@
 ;===------------------------------------------------------------------------===;
 
 ; i8 store
-; EG-CHECK: @store_i8
+; EG-CHECK-LABEL: @store_i8
 ; EG-CHECK: MEM_RAT MSKOR T[[RW_GPR:[0-9]]].XW, T{{[0-9]}}.X
 ; EG-CHECK: VTX_READ_8 [[VAL:T[0-9]\.X]], [[VAL]]
 ; IG 0: Get the byte index and truncate the value
@@ -26,7 +26,7 @@
 ; EG-CHECK: MOV T[[RW_GPR]].Y, 0.0
 ; EG-CHECK: MOV * T[[RW_GPR]].Z, 0.0
 
-; SI-CHECK: @store_i8
+; SI-CHECK-LABEL: @store_i8
 ; SI-CHECK: BUFFER_STORE_BYTE
 
 define void @store_i8(i8 addrspace(1)* %out, i8 %in) {
@@ -36,7 +36,7 @@ entry:
 }
 
 ; i16 store
-; EG-CHECK: @store_i16
+; EG-CHECK-LABEL: @store_i16
 ; EG-CHECK: MEM_RAT MSKOR T[[RW_GPR:[0-9]]].XW, T{{[0-9]}}.X
 ; EG-CHECK: VTX_READ_16 [[VAL:T[0-9]\.X]], [[VAL]]
 ; IG 0: Get the byte index and truncate the value
@@ -55,7 +55,7 @@ entry:
 ; EG-CHECK: MOV T[[RW_GPR]].Y, 0.0
 ; EG-CHECK: MOV * T[[RW_GPR]].Z, 0.0
 
-; SI-CHECK: @store_i16
+; SI-CHECK-LABEL: @store_i16
 ; SI-CHECK: BUFFER_STORE_SHORT
 define void @store_i16(i16 addrspace(1)* %out, i16 %in) {
 entry:
@@ -63,10 +63,10 @@ entry:
   ret void
 }
 
-; EG-CHECK: @store_v2i8
+; EG-CHECK-LABEL: @store_v2i8
 ; EG-CHECK: MEM_RAT MSKOR
 ; EG-CHECK-NOT: MEM_RAT MSKOR
-; SI-CHECK: @store_v2i8
+; SI-CHECK-LABEL: @store_v2i8
 ; SI-CHECK: BUFFER_STORE_BYTE
 ; SI-CHECK: BUFFER_STORE_BYTE
 define void @store_v2i8(<2 x i8> addrspace(1)* %out, <2 x i32> %in) {
@@ -77,12 +77,13 @@ entry:
 }
 
 
-; EG-CHECK: @store_v2i16
+; EG-CHECK-LABEL: @store_v2i16
 ; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @store_v2i16
+; CM-CHECK-LABEL: @store_v2i16
 ; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @store_v2i16
-; SI-CHECK: BUFFER_STORE_DWORD
+; SI-CHECK-LABEL: @store_v2i16
+; SI-CHECK: BUFFER_STORE_SHORT
+; SI-CHECK: BUFFER_STORE_SHORT
 define void @store_v2i16(<2 x i16> addrspace(1)* %out, <2 x i32> %in) {
 entry:
   %0 = trunc <2 x i32> %in to <2 x i16>
@@ -90,11 +91,11 @@ entry:
   ret void
 }
 
-; EG-CHECK: @store_v4i8
+; EG-CHECK-LABEL: @store_v4i8
 ; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @store_v4i8
+; CM-CHECK-LABEL: @store_v4i8
 ; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @store_v4i8
+; SI-CHECK-LABEL: @store_v4i8
 ; SI-CHECK: BUFFER_STORE_BYTE
 ; SI-CHECK: BUFFER_STORE_BYTE
 ; SI-CHECK: BUFFER_STORE_BYTE
@@ -107,11 +108,11 @@ entry:
 }
 
 ; floating-point store
-; EG-CHECK: @store_f32
+; EG-CHECK-LABEL: @store_f32
 ; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+\.X, T[0-9]+\.X}}, 1
-; CM-CHECK: @store_f32
+; CM-CHECK-LABEL: @store_f32
 ; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD T{{[0-9]+\.X, T[0-9]+\.X}}
-; SI-CHECK: @store_f32
+; SI-CHECK-LABEL: @store_f32
 ; SI-CHECK: BUFFER_STORE_DWORD
 
 define void @store_f32(float addrspace(1)* %out, float %in) {
@@ -119,13 +120,13 @@ define void @store_f32(float addrspace(1)* %out, float %in) {
   ret void
 }
 
-; EG-CHECK: @store_v4i16
+; EG-CHECK-LABEL: @store_v4i16
 ; EG-CHECK: MEM_RAT MSKOR
 ; EG-CHECK: MEM_RAT MSKOR
 ; EG-CHECK: MEM_RAT MSKOR
 ; EG-CHECK: MEM_RAT MSKOR
 ; EG-CHECK-NOT: MEM_RAT MSKOR
-; SI-CHECK: @store_v4i16
+; SI-CHECK-LABEL: @store_v4i16
 ; SI-CHECK: BUFFER_STORE_SHORT
 ; SI-CHECK: BUFFER_STORE_SHORT
 ; SI-CHECK: BUFFER_STORE_SHORT
@@ -139,11 +140,11 @@ entry:
 }
 
 ; vec2 floating-point stores
-; EG-CHECK: @store_v2f32
+; EG-CHECK-LABEL: @store_v2f32
 ; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @store_v2f32
+; CM-CHECK-LABEL: @store_v2f32
 ; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @store_v2f32
+; SI-CHECK-LABEL: @store_v2f32
 ; SI-CHECK: BUFFER_STORE_DWORDX2
 
 define void @store_v2f32(<2 x float> addrspace(1)* %out, float %a, float %b) {
@@ -154,13 +155,13 @@ entry:
   ret void
 }
 
-; EG-CHECK: @store_v4i32
+; EG-CHECK-LABEL: @store_v4i32
 ; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
 ; EG-CHECK-NOT: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @store_v4i32
+; CM-CHECK-LABEL: @store_v4i32
 ; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
 ; CM-CHECK-NOT: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @store_v4i32
+; SI-CHECK-LABEL: @store_v4i32
 ; SI-CHECK: BUFFER_STORE_DWORDX4
 define void @store_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %in) {
 entry:
@@ -172,41 +173,42 @@ entry:
 ; Local Address Space
 ;===------------------------------------------------------------------------===;
 
-; EG-CHECK: @store_local_i8
+; EG-CHECK-LABEL: @store_local_i8
 ; EG-CHECK: LDS_BYTE_WRITE
-; SI-CHECK: @store_local_i8
+; SI-CHECK-LABEL: @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-LABEL: @store_local_i16
 ; EG-CHECK: LDS_SHORT_WRITE
-; SI-CHECK: @store_local_i16
+; SI-CHECK-LABEL: @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-LABEL: @store_local_v2i16
 ; EG-CHECK: LDS_WRITE
-; CM-CHECK: @store_local_v2i16
+; CM-CHECK-LABEL: @store_local_v2i16
 ; CM-CHECK: LDS_WRITE
-; SI-CHECK: @store_local_v2i16
-; SI-CHECK: DS_WRITE_B32
+; SI-CHECK-LABEL: @store_local_v2i16
+; SI-CHECK: DS_WRITE_B16
+; SI-CHECK: DS_WRITE_B16
 define void @store_local_v2i16(<2 x i16> addrspace(3)* %out, <2 x i16> %in) {
 entry:
   store <2 x i16> %in, <2 x i16> addrspace(3)* %out
   ret void
 }
 
-; EG-CHECK: @store_local_v4i8
+; EG-CHECK-LABEL: @store_local_v4i8
 ; EG-CHECK: LDS_WRITE
-; CM-CHECK: @store_local_v4i8
+; CM-CHECK-LABEL: @store_local_v4i8
 ; CM-CHECK: LDS_WRITE
-; SI-CHECK: @store_local_v4i8
+; SI-CHECK-LABEL: @store_local_v4i8
 ; SI-CHECK: DS_WRITE_B8
 ; SI-CHECK: DS_WRITE_B8
 ; SI-CHECK: DS_WRITE_B8
@@ -217,13 +219,13 @@ entry:
   ret void
 }
 
-; EG-CHECK: @store_local_v2i32
+; EG-CHECK-LABEL: @store_local_v2i32
 ; EG-CHECK: LDS_WRITE
 ; EG-CHECK: LDS_WRITE
-; CM-CHECK: @store_local_v2i32
+; CM-CHECK-LABEL: @store_local_v2i32
 ; CM-CHECK: LDS_WRITE
 ; CM-CHECK: LDS_WRITE
-; SI-CHECK: @store_local_v2i32
+; SI-CHECK-LABEL: @store_local_v2i32
 ; SI-CHECK: DS_WRITE_B32
 ; SI-CHECK: DS_WRITE_B32
 define void @store_local_v2i32(<2 x i32> addrspace(3)* %out, <2 x i32> %in) {
@@ -232,17 +234,17 @@ entry:
   ret void
 }
 
-; EG-CHECK: @store_local_v4i32
+; EG-CHECK-LABEL: @store_local_v4i32
 ; EG-CHECK: LDS_WRITE
 ; EG-CHECK: LDS_WRITE
 ; EG-CHECK: LDS_WRITE
 ; EG-CHECK: LDS_WRITE
-; CM-CHECK: @store_local_v4i32
+; CM-CHECK-LABEL: @store_local_v4i32
 ; CM-CHECK: LDS_WRITE
 ; CM-CHECK: LDS_WRITE
 ; CM-CHECK: LDS_WRITE
 ; CM-CHECK: LDS_WRITE
-; SI-CHECK: @store_local_v4i32
+; SI-CHECK-LABEL: @store_local_v4i32
 ; SI-CHECK: DS_WRITE_B32
 ; SI-CHECK: DS_WRITE_B32
 ; SI-CHECK: DS_WRITE_B32
@@ -260,11 +262,11 @@ entry:
 ; Evergreen / Northern Islands don't support 64-bit stores yet, so there should
 ; be two 32-bit stores.
 
-; EG-CHECK: @vecload2
+; EG-CHECK-LABEL: @vecload2
 ; EG-CHECK: MEM_RAT_CACHELESS STORE_RAW
-; CM-CHECK: @vecload2
+; CM-CHECK-LABEL: @vecload2
 ; CM-CHECK: MEM_RAT_CACHELESS STORE_DWORD
-; SI-CHECK: @vecload2
+; SI-CHECK-LABEL: @vecload2
 ; SI-CHECK: BUFFER_STORE_DWORDX2
 define void @vecload2(i32 addrspace(1)* nocapture %out, i32 addrspace(2)* nocapture %mem) #0 {
 entry:
diff --git a/test/CodeGen/R600/trunc.ll b/test/CodeGen/R600/trunc.ll
index be7a430885..58709dffc9 100644
--- a/test/CodeGen/R600/trunc.ll
+++ b/test/CodeGen/R600/trunc.ll
@@ -21,7 +21,6 @@ define void @trunc_i64_to_i32_store(i32 addrspace(1)* %out, i64 %in) {
 ; SI: S_LOAD_DWORDX2
 ; SI: S_LOAD_DWORDX2 [[SREG:SGPR[0-9]+_SGPR[0-9]+]]
 ; SI: V_LSHL_B64 [[LO_VREG:VGPR[0-9]+]]_VGPR{{[0-9]+}}, [[SREG]], 2
-; SI-NOT: [[LO_VREG]]
 ; SI: BUFFER_STORE_DWORD [[LO_VREG]],
 define void @trunc_shl_i64(i32 addrspace(1)* %out, i64 %a) {
   %b = shl i64 %a, 2