GlobalISel: Implement bitcast action for G_INSERT_VECTOR_ELT

[arsenm]

Summary

This mirrors the support for the equivalent extracts. This also
creates a huge mess that would be greatly improved if we had any bit
operation combines.

[arsenm]

Rebase

[arsenm]

Rebase and add MIR sample

[arsenm]

ping

[aemerson]

Thanks for the helpful comments. LGTM with a few nits.

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

2502

Early exit if NewNumElts >= OldNumElts to save on indent?

2519

Unnecessary braces

2530

same

[arsenm]

e2f1b48f867d9ed349b7520dea495717bca26f31

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

2502

I considered this, but this function is only half implemented. The >= case should also be here. I think early exit would make sense for the shorter case, but I'm not sure which one that would be without implementing it

2519

This covers multiple lines

2530

Also covers multiple lines

llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h

public:
   LegalizeResult narrowScalarCTLZ(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult narrowScalarCTTZ(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult narrowScalarCTPOP(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
 
   /// Perform Bitcast legalize action on G_EXTRACT_VECTOR_ELT.
   LegalizeResult bitcastExtractVectorElt(MachineInstr &MI, unsigned TypeIdx,
                                          LLT CastTy);
 
+  /// Perform Bitcast legalize action on G_INSERT_VECTOR_ELT.
+  LegalizeResult bitcastInsertVectorElt(MachineInstr &MI, unsigned TypeIdx,
+                                        LLT CastTy);
+
   LegalizeResult lowerBitcast(MachineInstr &MI);
   LegalizeResult lowerBitCount(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
 
   LegalizeResult lowerU64ToF32BitOps(MachineInstr &MI);
   LegalizeResult lowerUITOFP(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerSITOFP(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerFPTOUI(MachineInstr &MI, unsigned TypeIdx, LLT Ty);
   LegalizeResult lowerFPTOSI(MachineInstr &MI);

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

if (DstTy.isVector()) {
     MIRBuilder.buildMerge(Dst, SrcRegs);
     MI.eraseFromParent();
     return Legalized;
   }
 
   return UnableToLegalize;
 }
 
+/// Figure out the bit offset into a register when coercing a vector index for
+/// the wide element type. This is only for the case when promoting vector to
+/// one with larger elements.
+//
+///
+/// %offset_idx = G_AND %idx, ~(-1 << Log2(DstEltSize / SrcEltSize))
+/// %offset_bits = G_SHL %offset_idx, Log2(SrcEltSize)
+static Register getBitcastWiderVectorElementOffset(MachineIRBuilder &B,
+                                                   Register Idx,
+                                                   unsigned NewEltSize,
+                                                   unsigned OldEltSize) {
+  const unsigned Log2EltRatio = Log2_32(NewEltSize / OldEltSize);
+  LLT IdxTy = B.getMRI()->getType(Idx);
+
+  // Now figure out the amount we need to shift to get the target bits.
+  auto OffsetMask = B.buildConstant(
+    IdxTy, ~(APInt::getAllOnesValue(IdxTy.getSizeInBits()) << Log2EltRatio));
+  auto OffsetIdx = B.buildAnd(IdxTy, Idx, OffsetMask);
+  return B.buildShl(IdxTy, OffsetIdx,
+                    B.buildConstant(IdxTy, Log2_32(OldEltSize))).getReg(0);
+}
+
 /// Perform a G_EXTRACT_VECTOR_ELT in a different sized vector element. If this
 /// is casting to a vector with a smaller element size, perform multiple element
 /// extracts and merge the results. If this is coercing to a vector with larger
 /// elements, index the bitcasted vector and extract the target element with bit
 /// operations. This is intended to force the indexing in the native register
 /// size for architectures that can dynamically index the register file.
 LegalizerHelper::LegalizeResult
 LegalizerHelper::bitcastExtractVectorElt(MachineInstr &MI, unsigned TypeIdx,
if (NewNumElts < OldNumElts) {
     auto ScaledIdx = MIRBuilder.buildLShr(IdxTy, Idx, Log2Ratio);
 
     Register WideElt = CastVec;
     if (CastTy.isVector()) {
       WideElt = MIRBuilder.buildExtractVectorElement(NewEltTy, CastVec,
                                                      ScaledIdx).getReg(0);
     }
 
-    // Now figure out the amount we need to shift to get the target bits.
-    auto OffsetMask = MIRBuilder.buildConstant(
-      IdxTy, ~(APInt::getAllOnesValue(IdxTy.getSizeInBits()) << Log2EltRatio));
-    auto OffsetIdx = MIRBuilder.buildAnd(IdxTy, Idx, OffsetMask);
-    auto OffsetBits = MIRBuilder.buildShl(
-      IdxTy, OffsetIdx,
-      MIRBuilder.buildConstant(IdxTy, Log2_32(OldEltSize)));
+    // Compute the bit offset into the register of the target element.
+    Register OffsetBits = getBitcastWiderVectorElementOffset(
+      MIRBuilder, Idx, NewEltSize, OldEltSize);
 
     // Shift the wide element to get the target element.
     auto ExtractedBits = MIRBuilder.buildLShr(NewEltTy, WideElt, OffsetBits);
     MIRBuilder.buildTrunc(Dst, ExtractedBits);
     MI.eraseFromParent();
     return Legalized;
   }
 
   return UnableToLegalize;
 }
 
+/// Emit code to insert \p InsertReg into \p TargetRet at \p OffsetBits in \p
+/// TargetReg, while preserving other bits in \p TargetReg.
+///
+/// (InsertReg << Offset) | (TargetReg & ~(-1 >> InsertReg.size()) << Offset)
+static Register buildBitFieldInsert(MachineIRBuilder &B,
+                                    Register TargetReg, Register InsertReg,
+                                    Register OffsetBits) {
+  LLT TargetTy = B.getMRI()->getType(TargetReg);
+  LLT InsertTy = B.getMRI()->getType(InsertReg);
+  auto ZextVal = B.buildZExt(TargetTy, InsertReg);
+  auto ShiftedInsertVal = B.buildShl(TargetTy, ZextVal, OffsetBits);
+
+  // Produce a bitmask of the value to insert
+  auto EltMask = B.buildConstant(
+    TargetTy, APInt::getLowBitsSet(TargetTy.getSizeInBits(),
+                                   InsertTy.getSizeInBits()));
+  // Shift it into position
+  auto ShiftedMask = B.buildShl(TargetTy, EltMask, OffsetBits);
+  auto InvShiftedMask = B.buildNot(TargetTy, ShiftedMask);
+
+  // Clear out the bits in the wide element
+  auto MaskedOldElt = B.buildAnd(TargetTy, TargetReg, InvShiftedMask);
+
+  // The value to insert has all zeros already, so stick it into the masked
+  // wide element.
+  return B.buildOr(TargetTy, MaskedOldElt, ShiftedInsertVal).getReg(0);
+}
+
+/// Perform a G_INSERT_VECTOR_ELT in a different sized vector element. If this
+/// is increasing the element size, perform the indexing in the target element
+/// type, and use bit operations to insert at the element position. This is
+/// intended for architectures that can dynamically index the register file and
+/// want to force indexing in the native register size.
+LegalizerHelper::LegalizeResult
+LegalizerHelper::bitcastInsertVectorElt(MachineInstr &MI, unsigned TypeIdx,
+                                        LLT CastTy) {
+  if (TypeIdx != 0)
+    return UnableToLegalize;
+
+  Register Dst = MI.getOperand(0).getReg();
+  Register SrcVec = MI.getOperand(1).getReg();
+  Register Val = MI.getOperand(2).getReg();
+  Register Idx = MI.getOperand(3).getReg();
+
+  LLT VecTy = MRI.getType(Dst);
+  LLT ValTy = MRI.getType(Val);
+  LLT IdxTy = MRI.getType(Idx);
+
+  LLT VecEltTy = VecTy.getElementType();
+  if (VecEltTy != ValTy)   // XXX - Is this valid mir?
+    return UnableToLegalize;
+
+  LLT NewEltTy = CastTy.isVector() ? CastTy.getElementType() : CastTy;
+  const unsigned NewEltSize = NewEltTy.getSizeInBits();
+  const unsigned OldEltSize = VecEltTy.getSizeInBits();
+
+  unsigned NewNumElts = CastTy.isVector() ? CastTy.getNumElements() : 1;
+  unsigned OldNumElts = VecTy.getNumElements();
+
+
+  Register CastVec = MIRBuilder.buildBitcast(CastTy, SrcVec).getReg(0);
+  if (NewNumElts < OldNumElts) {
+    if (NewEltSize % OldEltSize != 0)
+      return UnableToLegalize;
+
+    // This only depends on powers of 2 because we use bit tricks to figure out
+    // the bit offset we need to shift to get the target element. A general
+    // expansion could emit division/multiply.
+    if (!isPowerOf2_32(NewEltSize / OldEltSize))
+      return UnableToLegalize;
+
+    const unsigned Log2EltRatio = Log2_32(NewEltSize / OldEltSize);
+    auto Log2Ratio = MIRBuilder.buildConstant(IdxTy, Log2EltRatio);
+
+    // Divide to get the index in the wider element type.
+    auto ScaledIdx = MIRBuilder.buildLShr(IdxTy, Idx, Log2Ratio);
+
+    Register ExtractedElt = CastVec;
+    if (CastTy.isVector()) {
+      ExtractedElt = MIRBuilder.buildExtractVectorElement(NewEltTy, CastVec,
+                                                          ScaledIdx).getReg(0);
+    }
+
+    // Compute the bit offset into the register of the target element.
+    Register OffsetBits = getBitcastWiderVectorElementOffset(
+      MIRBuilder, Idx, NewEltSize, OldEltSize);
+
+    Register InsertedElt = buildBitFieldInsert(MIRBuilder, ExtractedElt,
+                                               Val, OffsetBits);
+    if (CastTy.isVector()) {
+      InsertedElt = MIRBuilder.buildInsertVectorElement(
+        CastTy, CastVec, InsertedElt, ScaledIdx).getReg(0);
+    }
+
+    MIRBuilder.buildBitcast(Dst, InsertedElt);
+    MI.eraseFromParent();
+    return Legalized;
+  }
+
+  return UnableToLegalize;
+}
+
 LegalizerHelper::LegalizeResult
 LegalizerHelper::bitcast(MachineInstr &MI, unsigned TypeIdx, LLT CastTy) {
   switch (MI.getOpcode()) {
   case TargetOpcode::G_LOAD: {
     if (TypeIdx != 0)
       return UnableToLegalize;
 
     Observer.changingInstr(MI);
case TargetOpcode::G_XOR: {
     bitcastSrc(MI, CastTy, 1);
     bitcastSrc(MI, CastTy, 2);
     bitcastDst(MI, CastTy, 0);
     Observer.changedInstr(MI);
     return Legalized;
   }
   case TargetOpcode::G_EXTRACT_VECTOR_ELT:
     return bitcastExtractVectorElt(MI, TypeIdx, CastTy);
+  case TargetOpcode::G_INSERT_VECTOR_ELT:
+    return bitcastInsertVectorElt(MI, TypeIdx, CastTy);
   default:
     return UnableToLegalize;
   }
 }
 
 LegalizerHelper::LegalizeResult
 LegalizerHelper::lower(MachineInstr &MI, unsigned TypeIdx, LLT Ty) {
   using namespace TargetOpcode;

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

getActionDefinitionsBuilder(Op)
           const LLT VecTy = Query.Types[VecTypeIdx];
           const LLT IdxTy = Query.Types[IdxTypeIdx];
           const unsigned EltSize = EltTy.getSizeInBits();
           return (EltSize == 32 || EltSize == 64) &&
                   VecTy.getSizeInBits() % 32 == 0 &&
                   VecTy.getSizeInBits() <= MaxRegisterSize &&
                   IdxTy.getSizeInBits() == 32;
         })
-      .bitcastIf(all(sizeIsMultiple32(1), scalarOrEltNarrowerThan(1, 32)),
-                 bitcastToVectorElement32(1))
+      .bitcastIf(all(sizeIsMultiple32(VecTypeIdx), scalarOrEltNarrowerThan(VecTypeIdx, 32)),
+                 bitcastToVectorElement32(VecTypeIdx))
       //.bitcastIf(vectorSmallerThan(1, 32), bitcastToScalar(1))
       .bitcastIf(
-        all(sizeIsMultiple32(1), scalarOrEltWiderThan(1, 64)),
+        all(sizeIsMultiple32(VecTypeIdx), scalarOrEltWiderThan(VecTypeIdx, 64)),
         [=](const LegalityQuery &Query) {
           // For > 64-bit element types, try to turn this into a 64-bit
           // element vector since we may be able to do better indexing
           // if this is scalar. If not, fall back to 32.
           const LLT EltTy = Query.Types[EltTypeIdx];
           const LLT VecTy = Query.Types[VecTypeIdx];
           const unsigned DstEltSize = EltTy.getSizeInBits();
           const unsigned VecSize = VecTy.getSizeInBits();

llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.i16.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9 %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX8 %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX7 %
+
+define amdgpu_ps void @insertelement_s_v2i16_s_s(<2 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v2i16_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX9-NEXT:    s_and_b32 s1, s5, 1
+; GFX9-NEXT:    s_mov_b32 s2, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_and_b32 s3, s4, s2
+; GFX9-NEXT:    s_lshl_b32 s3, s3, s1
+; GFX9-NEXT:    s_lshl_b32 s1, s2, s1
+; GFX9-NEXT:    s_not_b32 s1, s1
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s0, s0, s1
+; GFX9-NEXT:    s_or_b32 s0, s0, s3
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v2, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v2i16_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX8-NEXT:    s_and_b32 s1, s5, 1
+; GFX8-NEXT:    s_mov_b32 s2, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_and_b32 s3, s4, s2
+; GFX8-NEXT:    s_lshl_b32 s3, s3, s1
+; GFX8-NEXT:    s_lshl_b32 s1, s2, s1
+; GFX8-NEXT:    s_not_b32 s1, s1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s0, s0, s1
+; GFX8-NEXT:    s_or_b32 s0, s0, s3
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v2, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v2i16_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX7-NEXT:    s_and_b32 s1, s5, 1
+; GFX7-NEXT:    s_mov_b32 s2, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    s_and_b32 s3, s4, s2
+; GFX7-NEXT:    s_lshl_b32 s3, s3, s1
+; GFX7-NEXT:    s_lshl_b32 s1, s2, s1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s0, s0, s1
+; GFX7-NEXT:    s_or_b32 s0, s0, s3
+; GFX7-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7-NEXT:    v_mov_b32_e32 v2, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i16>, <2 x i16> addrspace(4)* %ptr
+  %insert = insertelement <2 x i16> %vec, i16 %val, i32 %idx
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v2i16_s_s(<2 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v2i16_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    s_and_b32 s1, s3, 1
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_and_b32 s2, s2, s0
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX9-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX9-NEXT:    s_not_b32 s0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_or_b32 v2, v0, s0, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v2i16_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dword v0, v[0:1]
+; GFX8-NEXT:    s_and_b32 s1, s3, 1
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_and_b32 s2, s2, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX8-NEXT:    v_or_b32_e32 v2, s2, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v2i16_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    flat_load_dword v0, v[0:1]
+; GFX7-NEXT:    s_and_b32 s1, s3, 1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    s_and_b32 s2, s2, s0
+; GFX7-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX7-NEXT:    s_not_b32 s0, s0
+; GFX7-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_or_b32_e32 v2, s2, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i16>, <2 x i16> addrspace(1 )* %ptr
+  %insert = insertelement <2 x i16> %vec, i16 %val, i32 %idx
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v2i16_v_s(<2 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v2i16_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX9-NEXT:    s_and_b32 s1, s4, 1
+; GFX9-NEXT:    s_mov_b32 s2, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    v_and_b32_e32 v0, s2, v0
+; GFX9-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX9-NEXT:    s_not_b32 s2, s2
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_and_b32 s0, s0, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, s0
+; GFX9-NEXT:    v_lshl_or_b32 v2, v0, s1, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v2i16_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX8-NEXT:    s_and_b32 s1, s4, 1
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_mov_b32 s2, 0xffff
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_lshl_b32 s1, s2, s1
+; GFX8-NEXT:    s_not_b32 s1, s1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_and_b32 s0, s0, s1
+; GFX8-NEXT:    v_or_b32_e32 v2, s0, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v2i16_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX7-NEXT:    s_and_b32 s1, s4, 1
+; GFX7-NEXT:    s_mov_b32 s2, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    v_and_b32_e32 v0, s2, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s1, v0
+; GFX7-NEXT:    s_lshl_b32 s1, s2, s1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_and_b32 s0, s0, s1
+; GFX7-NEXT:    v_or_b32_e32 v2, s0, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i16>, <2 x i16> addrspace(4)* %ptr
+  %insert = insertelement <2 x i16> %vec, i16 %val, i32 %idx
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v2i16_s_v(<2 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v2i16_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX9-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT:    s_mov_b32 s1, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX9-NEXT:    s_and_b32 s2, s4, s1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v0, s2
+; GFX9-NEXT:    v_lshlrev_b32_e64 v0, v0, s1
+; GFX9-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_and_or_b32 v2, s0, v0, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v2i16_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX8-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX8-NEXT:    s_mov_b32 s1, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX8-NEXT:    s_and_b32 s2, s4, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v0, s2
+; GFX8-NEXT:    v_lshlrev_b32_e64 v0, v0, s1
+; GFX8-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX8-NEXT:    v_or_b32_e32 v2, v0, v1
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v2i16_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX7-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX7-NEXT:    s_mov_b32 s1, 0xffff
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX7-NEXT:    s_and_b32 s2, s4, s1
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s2, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v0, s1, v0
+; GFX7-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_or_b32_e32 v2, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i16>, <2 x i16> addrspace(4)* %ptr
+  %insert = insertelement <2 x i16> %vec, i16 %val, i32 %idx
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v2i16_v_v(<2 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v2i16_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    s_mov_b32 s1, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s1
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_and_or_b32 v2, s0, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v2i16_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    s_mov_b32 s1, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s1
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX8-NEXT:    v_or_b32_e32 v2, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v2i16_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX7-NEXT:    s_mov_b32 s1, 0xffff
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, s1, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s1, v1
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i16>, <2 x i16> addrspace(4)* %ptr
+  %insert = insertelement <2 x i16> %vec, i16 %val, i32 %idx
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v2i16_s_v(<2 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v2i16_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_and_b32 s1, s2, s0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v1, s1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_or_b32 v2, v0, v1, v2
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v2i16_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dword v0, v[0:1]
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_and_b32 s1, s2, s0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v1, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX8-NEXT:    v_or_b32_e32 v2, v0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v2i16_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    flat_load_dword v0, v[0:1]
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_and_b32 s1, s2, s0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s1, v1
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, v0, v2
+; GFX7-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i16>, <2 x i16> addrspace(1)* %ptr
+  %insert = insertelement <2 x i16> %vec, i16 %val, i32 %idx
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v2i16_v_s(<2 x i16> addrspace(1)* %ptr, i16 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v2i16_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    s_and_b32 s1, s2, 1
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, s1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    s_not_b32 s0, s0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_or_b32 v2, v0, s0, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v2i16_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dword v0, v[0:1]
+; GFX8-NEXT:    s_and_b32 s1, s2, 1
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX8-NEXT:    v_or_b32_e32 v2, v0, v1
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v2i16_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    flat_load_dword v0, v[0:1]
+; GFX7-NEXT:    s_and_b32 s1, s2, 1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v2
+; GFX7-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX7-NEXT:    s_not_b32 s0, s0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, s1, v1
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_or_b32_e32 v2, v0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i16>, <2 x i16> addrspace(1)* %ptr
+  %insert = insertelement <2 x i16> %vec, i16 %val, i32 %idx
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v2i16_v_v(<2 x i16> addrspace(1)* %ptr, i16 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v2i16_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_or_b32 v2, v0, v1, v2
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v2i16_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dword v0, v[0:1]
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX8-NEXT:    v_or_b32_e32 v2, v0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v2i16_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    flat_load_dword v0, v[0:1]
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, v1, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_or_b32_e32 v2, v0, v2
+; GFX7-NEXT:    v_mov_b32_e32 v0, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0
+; GFX7-NEXT:    flat_store_dword v[0:1], v2
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i16>, <2 x i16> addrspace(1)* %ptr
+  %insert = insertelement <2 x i16> %vec, i16 %val, i32 %idx
+  store <2 x i16> %insert, <2 x i16> addrspace(1)* null
+  ret void
+}
+
+; FIXME: 3 element load/store legalization
+; define amdgpu_ps void @insertelement_s_v3i16_s_s(<3 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 inreg %idx) {
+;   %vec = load <3 x i16>, <3 x i16> addrspace(4)* %ptr
+;   %insert = insertelement <3 x i16> %vec, i16 %val, i32 %idx
+;   store <3 x i16> %insert, <3 x i16> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_v_v3i16_s_s(<3 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 inreg %idx) {
+;   %vec = load <3 x i16>, <3 x i16> addrspace(1 )* %ptr
+;   %insert = insertelement <3 x i16> %vec, i16 %val, i32 %idx
+;   store <3 x i16> %insert, <3 x i16> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_s_v3i16_v_s(<3 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 inreg %idx) {
+;   %vec = load <3 x i16>, <3 x i16> addrspace(4)* %ptr
+;   %insert = insertelement <3 x i16> %vec, i16 %val, i32 %idx
+;   store <3 x i16> %insert, <3 x i16> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_s_v3i16_s_v(<3 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 %idx) {
+;   %vec = load <3 x i16>, <3 x i16> addrspace(4)* %ptr
+;   %insert = insertelement <3 x i16> %vec, i16 %val, i32 %idx
+;   store <3 x i16> %insert, <3 x i16> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_s_v3i16_v_v(<3 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 %idx) {
+;   %vec = load <3 x i16>, <3 x i16> addrspace(4)* %ptr
+;   %insert = insertelement <3 x i16> %vec, i16 %val, i32 %idx
+;   store <3 x i16> %insert, <3 x i16> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_v_v3i16_s_v(<3 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 %idx) {
+;   %vec = load <3 x i16>, <3 x i16> addrspace(1)* %ptr
+;   %insert = insertelement <3 x i16> %vec, i16 %val, i32 %idx
+;   store <3 x i16> %insert, <3 x i16> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_v_v3i16_v_s(<3 x i16> addrspace(1)* %ptr, i16 %val, i32 inreg %idx) {
+;   %vec = load <3 x i16>, <3 x i16> addrspace(1)* %ptr
+;   %insert = insertelement <3 x i16> %vec, i16 %val, i32 %idx
+;   store <3 x i16> %insert, <3 x i16> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_v_v3i16_v_v(<3 x i16> addrspace(1)* %ptr, i16 %val, i32 %idx) {
+;   %vec = load <3 x i16>, <3 x i16> addrspace(1)* %ptr
+;   %insert = insertelement <3 x i16> %vec, i16 %val, i32 %idx
+;   store <3 x i16> %insert, <3 x i16> addrspace(1)* null
+;   ret void
+; }
+
+define amdgpu_ps void @insertelement_v_v4i16_s_s(<4 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v4i16_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_lshr_b32 s1, s3, 1
+; GFX9-NEXT:    s_and_b32 s3, s3, 1
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_and_b32 s2, s2, s0
+; GFX9-NEXT:    s_lshl_b32 s3, s3, 4
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s3
+; GFX9-NEXT:    s_lshl_b32 s2, s2, s3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX9-NEXT:    s_not_b32 s0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
+; GFX9-NEXT:    v_and_or_b32 v2, v3, s0, v2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v4i16_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    s_lshr_b32 s1, s3, 1
+; GFX8-NEXT:    s_and_b32 s3, s3, 1
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s3, s3, 4
+; GFX8-NEXT:    s_and_b32 s2, s2, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    s_lshl_b32 s2, s2, s3
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
+; GFX8-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, s2, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v4i16_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX7-NEXT:    s_lshr_b32 s1, s3, 1
+; GFX7-NEXT:    s_and_b32 s3, s3, 1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s3, s3, 4
+; GFX7-NEXT:    s_and_b32 s2, s2, s0
+; GFX7-NEXT:    s_lshl_b32 s0, s0, s3
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX7-NEXT:    s_not_b32 s0, s0
+; GFX7-NEXT:    s_lshl_b32 s2, s2, s3
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_or_b32_e32 v2, s2, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0
+; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i16>, <4 x i16> addrspace(1 )* %ptr
+  %insert = insertelement <4 x i16> %vec, i16 %val, i32 %idx
+  store <4 x i16> %insert, <4 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v4i16_v_s(<4 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v4i16_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX9-NEXT:    s_lshr_b32 s2, s4, 1
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX9-NEXT:    s_mov_b32 s5, 0xffff
+; GFX9-NEXT:    v_and_b32_e32 v0, s5, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX9-NEXT:    s_and_b32 s4, s4, 1
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX9-NEXT:    s_lshl_b32 s5, s5, s4
+; GFX9-NEXT:    s_not_b32 s5, s5
+; GFX9-NEXT:    s_and_b32 s3, s3, s5
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    v_lshl_or_b32 v2, v0, s4, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v4i16_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX8-NEXT:    s_lshr_b32 s2, s4, 1
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX8-NEXT:    s_mov_b32 s5, 0xffff
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX8-NEXT:    s_and_b32 s4, s4, 1
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    s_lshl_b32 s4, s5, s4
+; GFX8-NEXT:    s_not_b32 s4, s4
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    s_and_b32 s3, s3, s4
+; GFX8-NEXT:    v_or_b32_e32 v2, s3, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v4i16_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX7-NEXT:    s_lshr_b32 s2, s4, 1
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX7-NEXT:    s_mov_b32 s5, 0xffff
+; GFX7-NEXT:    v_and_b32_e32 v0, s5, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX7-NEXT:    s_and_b32 s4, s4, 1
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
+; GFX7-NEXT:    s_lshl_b32 s4, s5, s4
+; GFX7-NEXT:    s_not_b32 s4, s4
+; GFX7-NEXT:    s_and_b32 s3, s3, s4
+; GFX7-NEXT:    v_or_b32_e32 v2, s3, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0
+; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i16>, <4 x i16> addrspace(4)* %ptr
+  %insert = insertelement <4 x i16> %vec, i16 %val, i32 %idx
+  store <4 x i16> %insert, <4 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v4i16_s_v(<4 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v4i16_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 1, v0
+; GFX9-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT:    s_mov_b32 s2, 0xffff
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s0
+; GFX9-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX9-NEXT:    s_and_b32 s3, s4, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_lshlrev_b32_e64 v3, v0, s3
+; GFX9-NEXT:    v_lshlrev_b32_e64 v0, v0, s2
+; GFX9-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX9-NEXT:    v_and_or_b32 v3, v1, v0, v3
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v4i16_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 1, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX8-NEXT:    s_mov_b32 s2, 0xffff
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s0
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX8-NEXT:    s_and_b32 s3, s4, s2
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX8-NEXT:    v_lshlrev_b32_e64 v3, v0, s3
+; GFX8-NEXT:    v_lshlrev_b32_e64 v0, v0, s2
+; GFX8-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX8-NEXT:    v_or_b32_e32 v3, v0, v3
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v4i16_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 1, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX7-NEXT:    s_mov_b32 s2, 0xffff
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    v_mov_b32_e32 v3, s1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX7-NEXT:    s_and_b32 s3, s4, s2
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX7-NEXT:    v_lshl_b32_e32 v3, s3, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v0, s2, v0
+; GFX7-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_or_b32_e32 v3, v0, v3
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0
+; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i16>, <4 x i16> addrspace(4)* %ptr
+  %insert = insertelement <4 x i16> %vec, i16 %val, i32 %idx
+  store <4 x i16> %insert, <4 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v4i16_v_v(<4 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v4i16_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 1, v1
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    s_mov_b32 s2, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s2
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v3, s0
+; GFX9-NEXT:    v_mov_b32_e32 v4, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_and_or_b32 v3, v3, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v4i16_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 1, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    s_mov_b32 s2, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s2
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v3, s0
+; GFX8-NEXT:    v_mov_b32_e32 v4, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, v3, v1
+; GFX8-NEXT:    v_or_b32_e32 v3, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v4i16_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 1, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX7-NEXT:    s_mov_b32 s2, 0xffff
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, s2, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s2, v1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v3, s0
+; GFX7-NEXT:    v_mov_b32_e32 v4, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v3, v1
+; GFX7-NEXT:    v_or_b32_e32 v3, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0
+; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i16>, <4 x i16> addrspace(4)* %ptr
+  %insert = insertelement <4 x i16> %vec, i16 %val, i32 %idx
+  store <4 x i16> %insert, <4 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v4i16_s_v(<4 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v4i16_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 1, v2
+; GFX9-NEXT:    v_and_b32_e32 v2, 1, v2
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_and_b32 s1, s2, s0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 4, v2
+; GFX9-NEXT:    v_lshlrev_b32_e64 v4, v2, s1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v2, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v3
+; GFX9-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v3
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX9-NEXT:    v_and_or_b32 v2, v5, v2, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v4i16_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 1, v2
+; GFX8-NEXT:    v_and_b32_e32 v2, 1, v2
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_and_b32 s1, s2, s0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 4, v2
+; GFX8-NEXT:    v_lshlrev_b32_e64 v4, v2, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v2, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v3
+; GFX8-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v3
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX8-NEXT:    v_and_b32_e32 v2, v5, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v4i16_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, 1, v2
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_and_b32 s1, s2, s0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 4, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v4, s1, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v3
+; GFX7-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v3
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v2, v5, v2
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0
+; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i16>, <4 x i16> addrspace(1)* %ptr
+  %insert = insertelement <4 x i16> %vec, i16 %val, i32 %idx
+  store <4 x i16> %insert, <4 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v4i16_v_s(<4 x i16> addrspace(1)* %ptr, i16 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v4i16_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_lshr_b32 s1, s2, 1
+; GFX9-NEXT:    s_and_b32 s2, s2, 1
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 4
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s2, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    s_not_b32 s0, s0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
+; GFX9-NEXT:    v_and_or_b32 v2, v3, s0, v2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v4i16_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    s_lshr_b32 s1, s2, 1
+; GFX8-NEXT:    s_and_b32 s2, s2, 1
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 4
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
+; GFX8-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v4i16_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX7-NEXT:    s_lshr_b32 s1, s2, 1
+; GFX7-NEXT:    s_and_b32 s2, s2, 1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 4
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    s_lshl_b32 s0, s0, s2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX7-NEXT:    s_not_b32 s0, s0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, s2, v2
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0
+; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i16>, <4 x i16> addrspace(1)* %ptr
+  %insert = insertelement <4 x i16> %vec, i16 %val, i32 %idx
+  store <4 x i16> %insert, <4 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v4i16_v_v(<4 x i16> addrspace(1)* %ptr, i16 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v4i16_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 1, v3
+; GFX9-NEXT:    v_and_b32_e32 v3, 1, v3
+; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 4, v3
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v3, v3, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v4
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX9-NEXT:    v_and_or_b32 v2, v5, v3, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v4i16_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 1, v3
+; GFX8-NEXT:    v_and_b32_e32 v3, 1, v3
+; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 4, v3
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v3, v3, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX8-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v4
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX8-NEXT:    v_and_b32_e32 v3, v5, v3
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v4i16_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 1, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, 1, v3
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 4, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v3, s0, v3
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX7-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v4
+; GFX7-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v3, v5, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0
+; GFX7-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i16>, <4 x i16> addrspace(1)* %ptr
+  %insert = insertelement <4 x i16> %vec, i16 %val, i32 %idx
+  store <4 x i16> %insert, <4 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v8i16_s_s(<8 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v8i16_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX9-NEXT:    s_lshr_b32 s6, s5, 1
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX9-NEXT:    s_mov_b32 s8, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_cselect_b32 s7, s1, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX9-NEXT:    s_cselect_b32 s7, s2, s7
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX9-NEXT:    s_cselect_b32 s7, s3, s7
+; GFX9-NEXT:    s_and_b32 s5, s5, 1
+; GFX9-NEXT:    s_lshl_b32 s5, s5, 4
+; GFX9-NEXT:    s_and_b32 s4, s4, s8
+; GFX9-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX9-NEXT:    s_lshl_b32 s5, s8, s5
+; GFX9-NEXT:    s_not_b32 s5, s5
+; GFX9-NEXT:    s_and_b32 s5, s7, s5
+; GFX9-NEXT:    s_or_b32 s4, s5, s4
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 0
+; GFX9-NEXT:    s_cselect_b32 s0, s4, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX9-NEXT:    s_cselect_b32 s1, s4, s1
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX9-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX9-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v8i16_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX8-NEXT:    s_lshr_b32 s6, s5, 1
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX8-NEXT:    s_mov_b32 s8, 0xffff
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_cselect_b32 s7, s1, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX8-NEXT:    s_cselect_b32 s7, s2, s7
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX8-NEXT:    s_cselect_b32 s7, s3, s7
+; GFX8-NEXT:    s_and_b32 s5, s5, 1
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 4
+; GFX8-NEXT:    s_and_b32 s4, s4, s8
+; GFX8-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX8-NEXT:    s_lshl_b32 s5, s8, s5
+; GFX8-NEXT:    s_not_b32 s5, s5
+; GFX8-NEXT:    s_and_b32 s5, s7, s5
+; GFX8-NEXT:    s_or_b32 s4, s5, s4
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 0
+; GFX8-NEXT:    s_cselect_b32 s0, s4, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX8-NEXT:    s_cselect_b32 s1, s4, s1
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX8-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX8-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v8i16_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX7-NEXT:    s_lshr_b32 s6, s5, 1
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX7-NEXT:    s_mov_b32 s8, 0xffff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_cselect_b32 s7, s1, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX7-NEXT:    s_cselect_b32 s7, s2, s7
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX7-NEXT:    s_cselect_b32 s7, s3, s7
+; GFX7-NEXT:    s_and_b32 s5, s5, 1
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 4
+; GFX7-NEXT:    s_and_b32 s4, s4, s8
+; GFX7-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX7-NEXT:    s_lshl_b32 s5, s8, s5
+; GFX7-NEXT:    s_not_b32 s5, s5
+; GFX7-NEXT:    s_and_b32 s5, s7, s5
+; GFX7-NEXT:    s_or_b32 s4, s5, s4
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 0
+; GFX7-NEXT:    s_cselect_b32 s0, s4, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX7-NEXT:    s_cselect_b32 s1, s4, s1
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX7-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX7-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_mov_b32_e32 v2, s2
+; GFX7-NEXT:    v_mov_b32_e32 v3, s3
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i16>, <8 x i16> addrspace(4)* %ptr
+  %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx
+  store <8 x i16> %insert, <8 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v8i16_s_s(<8 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v8i16_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX9-NEXT:    s_and_b32 s1, s3, 1
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_lshr_b32 s4, s3, 1
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_and_b32 s2, s2, s0
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX9-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX9-NEXT:    s_not_b32 s5, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX9-NEXT:    v_mov_b32_e32 v4, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v5, v2, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v5, v3, s[2:3]
+; GFX9-NEXT:    v_and_or_b32 v4, v5, s5, v4
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v8i16_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
+; GFX8-NEXT:    s_and_b32 s1, s3, 1
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_lshr_b32 s4, s3, 1
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_and_b32 s2, s2, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX8-NEXT:    s_lshl_b32 s5, s2, s1
+; GFX8-NEXT:    s_not_b32 s6, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e32 v4, v0, v1, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v3, s[2:3]
+; GFX8-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX8-NEXT:    v_or_b32_e32 v4, s5, v4
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v8i16_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s10, 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
+; GFX7-NEXT:    s_and_b32 s1, s3, 1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_lshr_b32 s4, s3, 1
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    s_and_b32 s2, s2, s0
+; GFX7-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX7-NEXT:    s_lshl_b32 s5, s2, s1
+; GFX7-NEXT:    s_not_b32 s6, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e32 v4, v0, v1, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v2, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v3, s[2:3]
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX7-NEXT:    v_or_b32_e32 v4, s5, v4
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[2:3]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i16>, <8 x i16> addrspace(1 )* %ptr
+  %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx
+  store <8 x i16> %insert, <8 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v8i16_v_s(<8 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v8i16_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX9-NEXT:    s_lshr_b32 s5, s4, 1
+; GFX9-NEXT:    s_cmp_eq_u32 s5, 1
+; GFX9-NEXT:    s_mov_b32 s7, 0xffff
+; GFX9-NEXT:    v_and_b32_e32 v0, s7, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_cselect_b32 s6, s1, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s5, 2
+; GFX9-NEXT:    s_cselect_b32 s6, s2, s6
+; GFX9-NEXT:    s_cmp_eq_u32 s5, 3
+; GFX9-NEXT:    s_cselect_b32 s6, s3, s6
+; GFX9-NEXT:    s_and_b32 s4, s4, 1
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX9-NEXT:    s_lshl_b32 s7, s7, s4
+; GFX9-NEXT:    s_not_b32 s7, s7
+; GFX9-NEXT:    s_and_b32 s6, s6, s7
+; GFX9-NEXT:    v_mov_b32_e32 v1, s6
+; GFX9-NEXT:    v_lshl_or_b32 v4, v0, s4, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 3
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v8i16_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX8-NEXT:    s_lshr_b32 s5, s4, 1
+; GFX8-NEXT:    s_cmp_eq_u32 s5, 1
+; GFX8-NEXT:    s_mov_b32 s7, 0xffff
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_cselect_b32 s6, s1, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s5, 2
+; GFX8-NEXT:    s_cselect_b32 s6, s2, s6
+; GFX8-NEXT:    s_cmp_eq_u32 s5, 3
+; GFX8-NEXT:    s_cselect_b32 s6, s3, s6
+; GFX8-NEXT:    s_and_b32 s4, s4, 1
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    s_lshl_b32 s4, s7, s4
+; GFX8-NEXT:    s_not_b32 s4, s4
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    s_and_b32 s4, s6, s4
+; GFX8-NEXT:    v_or_b32_e32 v4, s4, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 3
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v8i16_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX7-NEXT:    s_lshr_b32 s5, s4, 1
+; GFX7-NEXT:    s_cmp_eq_u32 s5, 1
+; GFX7-NEXT:    s_mov_b32 s7, 0xffff
+; GFX7-NEXT:    v_and_b32_e32 v0, s7, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_cselect_b32 s6, s1, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s5, 2
+; GFX7-NEXT:    s_cselect_b32 s6, s2, s6
+; GFX7-NEXT:    s_cmp_eq_u32 s5, 3
+; GFX7-NEXT:    s_cselect_b32 s6, s3, s6
+; GFX7-NEXT:    s_and_b32 s4, s4, 1
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
+; GFX7-NEXT:    s_lshl_b32 s4, s7, s4
+; GFX7-NEXT:    s_not_b32 s4, s4
+; GFX7-NEXT:    s_and_b32 s4, s6, s4
+; GFX7-NEXT:    v_or_b32_e32 v4, s4, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, s2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2
+; GFX7-NEXT:    v_mov_b32_e32 v3, s3
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 3
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i16>, <8 x i16> addrspace(4)* %ptr
+  %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx
+  store <8 x i16> %insert, <8 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v8i16_s_v(<8 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v8i16_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[2:3], 0x0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 1, v0
+; GFX9-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT:    s_mov_b32 s5, 0xffff
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s8
+; GFX9-NEXT:    v_mov_b32_e32 v2, s9
+; GFX9-NEXT:    v_mov_b32_e32 v3, s10
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX9-NEXT:    s_and_b32 s4, s4, s5
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v0, s4
+; GFX9-NEXT:    v_lshlrev_b32_e64 v0, v0, s5
+; GFX9-NEXT:    v_mov_b32_e32 v5, s11
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[2:3]
+; GFX9-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX9-NEXT:    v_and_or_b32 v5, v1, v0, v2
+; GFX9-NEXT:    v_mov_b32_e32 v0, s8
+; GFX9-NEXT:    v_mov_b32_e32 v1, s9
+; GFX9-NEXT:    v_mov_b32_e32 v2, s10
+; GFX9-NEXT:    v_mov_b32_e32 v3, s11
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v8i16_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[8:11], s[2:3], 0x0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 1, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX8-NEXT:    s_mov_b32 s5, 0xffff
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s8
+; GFX8-NEXT:    v_mov_b32_e32 v2, s9
+; GFX8-NEXT:    v_mov_b32_e32 v3, s10
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX8-NEXT:    s_and_b32 s4, s4, s5
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v0, s4
+; GFX8-NEXT:    v_lshlrev_b32_e64 v0, v0, s5
+; GFX8-NEXT:    v_mov_b32_e32 v5, s11
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[2:3]
+; GFX8-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX8-NEXT:    v_or_b32_e32 v5, v0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v0, s8
+; GFX8-NEXT:    v_mov_b32_e32 v1, s9
+; GFX8-NEXT:    v_mov_b32_e32 v2, s10
+; GFX8-NEXT:    v_mov_b32_e32 v3, s11
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v8i16_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[2:3], 0x0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 1, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX7-NEXT:    s_mov_b32 s5, 0xffff
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v1, s8
+; GFX7-NEXT:    v_mov_b32_e32 v2, s9
+; GFX7-NEXT:    v_mov_b32_e32 v3, s10
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX7-NEXT:    s_and_b32 s4, s4, s5
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s4, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v0, s5, v0
+; GFX7-NEXT:    v_mov_b32_e32 v5, s11
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[2:3]
+; GFX7-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_or_b32_e32 v5, v0, v2
+; GFX7-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-NEXT:    v_mov_b32_e32 v1, s9
+; GFX7-NEXT:    v_mov_b32_e32 v2, s10
+; GFX7-NEXT:    v_mov_b32_e32 v3, s11
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i16>, <8 x i16> addrspace(4)* %ptr
+  %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx
+  store <8 x i16> %insert, <8 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v8i16_v_v(<8 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v8i16_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 1, v1
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX9-NEXT:    s_mov_b32 s8, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s8
+; GFX9-NEXT:    v_mov_b32_e32 v6, s7
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[2:3]
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_and_or_b32 v5, v2, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NEXT:    v_mov_b32_e32 v2, s6
+; GFX9-NEXT:    v_mov_b32_e32 v3, s7
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v8i16_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 1, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v2, s4
+; GFX8-NEXT:    v_mov_b32_e32 v3, s5
+; GFX8-NEXT:    v_mov_b32_e32 v5, s6
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX8-NEXT:    s_mov_b32 s8, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s8
+; GFX8-NEXT:    v_mov_b32_e32 v6, s7
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[2:3]
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX8-NEXT:    v_or_b32_e32 v5, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s4
+; GFX8-NEXT:    v_mov_b32_e32 v1, s5
+; GFX8-NEXT:    v_mov_b32_e32 v2, s6
+; GFX8-NEXT:    v_mov_b32_e32 v3, s7
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v8i16_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 1, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX7-NEXT:    s_mov_b32 s8, 0xffff
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v2, s4
+; GFX7-NEXT:    v_mov_b32_e32 v3, s5
+; GFX7-NEXT:    v_mov_b32_e32 v5, s6
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, s8, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s8, v1
+; GFX7-NEXT:    v_mov_b32_e32 v6, s7
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[2:3]
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_or_b32_e32 v5, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s4
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mov_b32_e32 v2, s6
+; GFX7-NEXT:    v_mov_b32_e32 v3, s7
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i16>, <8 x i16> addrspace(4)* %ptr
+  %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx
+  store <8 x i16> %insert, <8 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v8i16_s_v(<8 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v8i16_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[3:6], v[0:1], off
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 1, v2
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_and_b32 s1, s2, s0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v1, s1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e32 v7, v3, v4, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v7, v7, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v7, v7, v6, s[2:3]
+; GFX9-NEXT:    v_and_or_b32 v7, v7, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v4, v7, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v5, v7, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v3, v7, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v6, v7, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v8i16_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[3:6], v[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, 1, v2
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_and_b32 s1, s2, s0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v1, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e32 v7, v3, v4, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v7, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v7, v6, s[2:3]
+; GFX8-NEXT:    v_and_b32_e32 v1, v7, v1
+; GFX8-NEXT:    v_or_b32_e32 v7, v1, v2
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v4, v7, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v5, v7, s[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v3, v7, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v6, v7, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v8i16_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s10, 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[3:6], v[0:1], s[8:11], 0 addr64
+; GFX7-NEXT:    v_lshrrev_b32_e32 v0, 1, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_and_b32 s1, s2, s0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s1, v1
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v0
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e32 v7, v3, v4, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v7, v7, v5, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v7, v7, v6, s[2:3]
+; GFX7-NEXT:    v_and_b32_e32 v1, v7, v1
+; GFX7-NEXT:    v_or_b32_e32 v7, v1, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v3, v7, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v4, v7, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v5, v7, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v6, v7, s[2:3]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i16>, <8 x i16> addrspace(1)* %ptr
+  %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx
+  store <8 x i16> %insert, <8 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v8i16_v_s(<8 x i16> addrspace(1)* %ptr, i16 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v8i16_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[3:6], v[0:1], off
+; GFX9-NEXT:    s_and_b32 s1, s2, 1
+; GFX9-NEXT:    s_lshr_b32 s4, s2, 1
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    s_not_b32 s5, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[2:3]
+; GFX9-NEXT:    v_and_or_b32 v7, v1, s5, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v4, v7, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v5, v7, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v3, v7, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v6, v7, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v8i16_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[3:6], v[0:1]
+; GFX8-NEXT:    s_and_b32 s1, s2, 1
+; GFX8-NEXT:    s_lshr_b32 s4, s2, 1
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX8-NEXT:    v_mov_b32_e32 v0, s1
+; GFX8-NEXT:    s_not_b32 s5, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[2:3]
+; GFX8-NEXT:    v_and_b32_e32 v1, s5, v1
+; GFX8-NEXT:    v_or_b32_e32 v7, v1, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v4, v7, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v5, v7, s[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v3, v7, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v6, v7, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v8i16_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s10, 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[3:6], v[0:1], s[8:11], 0 addr64
+; GFX7-NEXT:    s_and_b32 s1, s2, 1
+; GFX7-NEXT:    s_lshr_b32 s4, s2, 1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v2
+; GFX7-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s1, v0
+; GFX7-NEXT:    s_not_b32 s5, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[2:3]
+; GFX7-NEXT:    v_and_b32_e32 v1, s5, v1
+; GFX7-NEXT:    v_or_b32_e32 v7, v1, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v3, v7, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v4, v7, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v5, v7, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v6, v7, s[2:3]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i16>, <8 x i16> addrspace(1)* %ptr
+  %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx
+  store <8 x i16> %insert, <8 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v8i16_v_v(<8 x i16> addrspace(1)* %ptr, i16 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v8i16_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[0:1], off
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 1, v3
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v7, s[2:3]
+; GFX9-NEXT:    v_and_or_b32 v3, v3, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v4, v3, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v5, v3, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v6, v3, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v7, v3, s[2:3]
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v8i16_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[4:7], v[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, 1, v3
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v7, s[2:3]
+; GFX8-NEXT:    v_and_b32_e32 v1, v3, v1
+; GFX8-NEXT:    v_or_b32_e32 v3, v1, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v4, v3, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v5, v3, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v6, v3, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v7, v3, s[2:3]
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v8i16_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s10, 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[4:7], v[0:1], s[8:11], 0 addr64
+; GFX7-NEXT:    v_lshrrev_b32_e32 v0, 1, v3
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, v1, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v0
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v7, s[2:3]
+; GFX7-NEXT:    v_and_b32_e32 v1, v3, v1
+; GFX7-NEXT:    v_or_b32_e32 v3, v1, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v4, v3, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v5, v3, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v6, v3, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v7, v3, s[2:3]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i16>, <8 x i16> addrspace(1)* %ptr
+  %insert = insertelement <8 x i16> %vec, i16 %val, i32 %idx
+  store <8 x i16> %insert, <8 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v16i16_s_s(<16 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v16i16_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
+; GFX9-NEXT:    s_lshr_b32 s7, s5, 1
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 1
+; GFX9-NEXT:    s_mov_b32 s2, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_cselect_b32 s0, s9, s8
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 2
+; GFX9-NEXT:    s_cselect_b32 s0, s10, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 3
+; GFX9-NEXT:    s_cselect_b32 s0, s11, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 4
+; GFX9-NEXT:    s_cselect_b32 s0, s12, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 5
+; GFX9-NEXT:    s_cselect_b32 s0, s13, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 6
+; GFX9-NEXT:    s_cselect_b32 s0, s14, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX9-NEXT:    s_cselect_b32 s0, s15, s0
+; GFX9-NEXT:    s_and_b32 s1, s5, 1
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_and_b32 s3, s4, s2
+; GFX9-NEXT:    s_lshl_b32 s3, s3, s1
+; GFX9-NEXT:    s_lshl_b32 s1, s2, s1
+; GFX9-NEXT:    s_not_b32 s1, s1
+; GFX9-NEXT:    s_and_b32 s0, s0, s1
+; GFX9-NEXT:    s_or_b32 s16, s0, s3
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 0
+; GFX9-NEXT:    s_cselect_b32 s0, s16, s8
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 1
+; GFX9-NEXT:    s_cselect_b32 s1, s16, s9
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 2
+; GFX9-NEXT:    s_cselect_b32 s2, s16, s10
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 3
+; GFX9-NEXT:    s_cselect_b32 s3, s16, s11
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 4
+; GFX9-NEXT:    s_cselect_b32 s4, s16, s12
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 5
+; GFX9-NEXT:    s_cselect_b32 s5, s16, s13
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 6
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    s_cselect_b32 s6, s16, s14
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX9-NEXT:    s_cselect_b32 s7, s16, s15
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NEXT:    s_add_u32 s0, 0, 16
+; GFX9-NEXT:    s_addc_u32 s1, 0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v4, s4
+; GFX9-NEXT:    v_mov_b32_e32 v11, s1
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, s5
+; GFX9-NEXT:    v_mov_b32_e32 v6, s6
+; GFX9-NEXT:    v_mov_b32_e32 v7, s7
+; GFX9-NEXT:    v_mov_b32_e32 v10, s0
+; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
+; GFX9-NEXT:    global_store_dwordx4 v[10:11], v[4:7], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v16i16_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
+; GFX8-NEXT:    s_lshr_b32 s7, s5, 1
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 1
+; GFX8-NEXT:    s_mov_b32 s2, 0xffff
+; GFX8-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_cselect_b32 s0, s9, s8
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 2
+; GFX8-NEXT:    s_cselect_b32 s0, s10, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 3
+; GFX8-NEXT:    s_cselect_b32 s0, s11, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 4
+; GFX8-NEXT:    s_cselect_b32 s0, s12, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 5
+; GFX8-NEXT:    s_cselect_b32 s0, s13, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 6
+; GFX8-NEXT:    s_cselect_b32 s0, s14, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX8-NEXT:    s_cselect_b32 s0, s15, s0
+; GFX8-NEXT:    s_and_b32 s1, s5, 1
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_and_b32 s3, s4, s2
+; GFX8-NEXT:    s_lshl_b32 s3, s3, s1
+; GFX8-NEXT:    s_lshl_b32 s1, s2, s1
+; GFX8-NEXT:    s_not_b32 s1, s1
+; GFX8-NEXT:    s_and_b32 s0, s0, s1
+; GFX8-NEXT:    s_or_b32 s16, s0, s3
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 0
+; GFX8-NEXT:    s_cselect_b32 s0, s16, s8
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 1
+; GFX8-NEXT:    s_cselect_b32 s1, s16, s9
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 2
+; GFX8-NEXT:    s_cselect_b32 s2, s16, s10
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 3
+; GFX8-NEXT:    s_cselect_b32 s3, s16, s11
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 4
+; GFX8-NEXT:    s_cselect_b32 s4, s16, s12
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 5
+; GFX8-NEXT:    s_cselect_b32 s5, s16, s13
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 6
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    s_cselect_b32 s6, s16, s14
+; GFX8-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX8-NEXT:    s_cselect_b32 s7, s16, s15
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    s_add_u32 s0, 0, 16
+; GFX8-NEXT:    s_addc_u32 s1, 0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v4, s4
+; GFX8-NEXT:    v_mov_b32_e32 v11, s1
+; GFX8-NEXT:    v_mov_b32_e32 v9, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, s5
+; GFX8-NEXT:    v_mov_b32_e32 v6, s6
+; GFX8-NEXT:    v_mov_b32_e32 v7, s7
+; GFX8-NEXT:    v_mov_b32_e32 v10, s0
+; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GFX8-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v16i16_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
+; GFX7-NEXT:    s_lshr_b32 s7, s5, 1
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 1
+; GFX7-NEXT:    s_mov_b32 s2, 0xffff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_cselect_b32 s0, s9, s8
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 2
+; GFX7-NEXT:    s_cselect_b32 s0, s10, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 3
+; GFX7-NEXT:    s_cselect_b32 s0, s11, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 4
+; GFX7-NEXT:    s_cselect_b32 s0, s12, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 5
+; GFX7-NEXT:    s_cselect_b32 s0, s13, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 6
+; GFX7-NEXT:    s_cselect_b32 s0, s14, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX7-NEXT:    s_cselect_b32 s0, s15, s0
+; GFX7-NEXT:    s_and_b32 s1, s5, 1
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    s_and_b32 s3, s4, s2
+; GFX7-NEXT:    s_lshl_b32 s3, s3, s1
+; GFX7-NEXT:    s_lshl_b32 s1, s2, s1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_and_b32 s0, s0, s1
+; GFX7-NEXT:    s_or_b32 s16, s0, s3
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 0
+; GFX7-NEXT:    s_cselect_b32 s0, s16, s8
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 1
+; GFX7-NEXT:    s_cselect_b32 s1, s16, s9
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 2
+; GFX7-NEXT:    s_cselect_b32 s2, s16, s10
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 3
+; GFX7-NEXT:    s_cselect_b32 s3, s16, s11
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 4
+; GFX7-NEXT:    s_cselect_b32 s4, s16, s12
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 5
+; GFX7-NEXT:    s_cselect_b32 s5, s16, s13
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 6
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    s_cselect_b32 s6, s16, s14
+; GFX7-NEXT:    s_cmp_eq_u32 s7, 7
+; GFX7-NEXT:    s_cselect_b32 s7, s16, s15
+; GFX7-NEXT:    v_mov_b32_e32 v4, s4
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_mov_b32_e32 v2, s2
+; GFX7-NEXT:    v_mov_b32_e32 v3, s3
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    v_mov_b32_e32 v5, s5
+; GFX7-NEXT:    v_mov_b32_e32 v6, s6
+; GFX7-NEXT:    v_mov_b32_e32 v7, s7
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[8:11], 0 offset:16
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i16>, <16 x i16> addrspace(4)* %ptr
+  %insert = insertelement <16 x i16> %vec, i16 %val, i32 %idx
+  store <16 x i16> %insert, <16 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v16i16_s_s(<16 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v16i16_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[2:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx4 v[6:9], v[0:1], off offset:16
+; GFX9-NEXT:    s_and_b32 s1, s3, 1
+; GFX9-NEXT:    s_lshr_b32 s12, s3, 1
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_and_b32 s2, s2, s0
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s12, 1
+; GFX9-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX9-NEXT:    s_not_b32 s13, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s12, 2
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s12, 3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], s12, 4
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[6:7], s12, 5
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[8:9], s12, 6
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[10:11], s12, 7
+; GFX9-NEXT:    s_waitcnt vmcnt(1)
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v2, v3, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v8, s[8:9]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[10:11]
+; GFX9-NEXT:    v_and_or_b32 v10, v1, s13, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[12:13], s12, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v2, v10, s[12:13]
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v4, v10, s[0:1]
+; GFX9-NEXT:    s_add_u32 s0, 0, 16
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v10, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v5, v10, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v6, v10, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v7, v10, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v8, v10, s[8:9]
+; GFX9-NEXT:    v_cndmask_b32_e64 v7, v9, v10, s[10:11]
+; GFX9-NEXT:    s_addc_u32 s1, 0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-NEXT:    v_mov_b32_e32 v11, s1
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0
+; GFX9-NEXT:    v_mov_b32_e32 v10, s0
+; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
+; GFX9-NEXT:    global_store_dwordx4 v[10:11], v[4:7], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v16i16_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[2:5], v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, 16, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dwordx4 v[6:9], v[0:1]
+; GFX8-NEXT:    s_and_b32 s1, s3, 1
+; GFX8-NEXT:    s_lshr_b32 s12, s3, 1
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_and_b32 s2, s2, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s12, 1
+; GFX8-NEXT:    s_lshl_b32 s13, s2, s1
+; GFX8-NEXT:    s_not_b32 s14, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s12, 2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], s12, 3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], s12, 4
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[6:7], s12, 5
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[8:9], s12, 6
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[10:11], s12, 7
+; GFX8-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[2:3]
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v6, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v7, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v8, s[8:9]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[10:11]
+; GFX8-NEXT:    v_and_b32_e32 v0, s14, v0
+; GFX8-NEXT:    v_or_b32_e32 v10, s13, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[12:13], s12, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v2, v10, s[12:13]
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v4, v10, s[0:1]
+; GFX8-NEXT:    s_add_u32 s0, 0, 16
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v3, v10, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v5, v10, s[2:3]
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v6, v10, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v5, v7, v10, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v8, v10, s[8:9]
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v9, v10, s[10:11]
+; GFX8-NEXT:    s_addc_u32 s1, 0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8-NEXT:    v_mov_b32_e32 v11, s1
+; GFX8-NEXT:    v_mov_b32_e32 v9, 0
+; GFX8-NEXT:    v_mov_b32_e32 v10, s0
+; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GFX8-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v16i16_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s18, 0
+; GFX7-NEXT:    s_mov_b32 s19, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[16:17], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[2:5], v[0:1], s[16:19], 0 addr64
+; GFX7-NEXT:    buffer_load_dwordx4 v[6:9], v[0:1], s[16:19], 0 addr64 offset:16
+; GFX7-NEXT:    s_and_b32 s1, s3, 1
+; GFX7-NEXT:    s_lshr_b32 s12, s3, 1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    s_and_b32 s2, s2, s0
+; GFX7-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s12, 1
+; GFX7-NEXT:    s_lshl_b32 s13, s2, s1
+; GFX7-NEXT:    s_not_b32 s14, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s12, 2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], s12, 3
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], s12, 4
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[6:7], s12, 5
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[8:9], s12, 6
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[10:11], s12, 7
+; GFX7-NEXT:    s_mov_b32 s18, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(1)
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[2:3]
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v6, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v7, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v8, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[10:11]
+; GFX7-NEXT:    v_and_b32_e32 v0, s14, v0
+; GFX7-NEXT:    v_or_b32_e32 v10, s13, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[12:13], s12, 0
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v2, v10, s[12:13]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v3, v10, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v4, v10, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v5, v10, s[2:3]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v6, v10, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v7, v10, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v6, v8, v10, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v7, v9, v10, s[10:11]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[16:19], 0
+; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[16:19], 0 offset:16
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i16>, <16 x i16> addrspace(1 )* %ptr
+  %insert = insertelement <16 x i16> %vec, i16 %val, i32 %idx
+  store <16 x i16> %insert, <16 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v16i16_v_s(<16 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v16i16_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
+; GFX9-NEXT:    s_lshr_b32 s2, s4, 1
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX9-NEXT:    s_mov_b32 s3, 0xffff
+; GFX9-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_cselect_b32 s0, s9, s8
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 2
+; GFX9-NEXT:    s_cselect_b32 s0, s10, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 3
+; GFX9-NEXT:    s_cselect_b32 s0, s11, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 4
+; GFX9-NEXT:    s_cselect_b32 s0, s12, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 5
+; GFX9-NEXT:    s_cselect_b32 s0, s13, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 6
+; GFX9-NEXT:    s_cselect_b32 s0, s14, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 7
+; GFX9-NEXT:    s_cselect_b32 s0, s15, s0
+; GFX9-NEXT:    s_and_b32 s1, s4, 1
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_lshl_b32 s3, s3, s1
+; GFX9-NEXT:    s_not_b32 s3, s3
+; GFX9-NEXT:    s_and_b32 s0, s0, s3
+; GFX9-NEXT:    v_mov_b32_e32 v1, s0
+; GFX9-NEXT:    v_lshl_or_b32 v8, v0, s1, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s8
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v1, s9
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, s10
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 2
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s11
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 3
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v8, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v5, s13
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 5
+; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v8, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v6, s14
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 6
+; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v8, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v4, s12
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s2, 4
+; GFX9-NEXT:    v_mov_b32_e32 v7, s15
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 7
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v4, v8, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v7, v7, v8, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-NEXT:    s_add_u32 s0, 0, 16
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0
+; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
+; GFX9-NEXT:    s_addc_u32 s1, 0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_store_dwordx4 v[0:1], v[4:7], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v16i16_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
+; GFX8-NEXT:    s_lshr_b32 s2, s4, 1
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX8-NEXT:    s_mov_b32 s3, 0xffff
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_cselect_b32 s0, s9, s8
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 2
+; GFX8-NEXT:    s_cselect_b32 s0, s10, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 3
+; GFX8-NEXT:    s_cselect_b32 s0, s11, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 4
+; GFX8-NEXT:    s_cselect_b32 s0, s12, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 5
+; GFX8-NEXT:    s_cselect_b32 s0, s13, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 6
+; GFX8-NEXT:    s_cselect_b32 s0, s14, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 7
+; GFX8-NEXT:    s_cselect_b32 s0, s15, s0
+; GFX8-NEXT:    s_and_b32 s1, s4, 1
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_lshl_b32 s1, s3, s1
+; GFX8-NEXT:    s_not_b32 s1, s1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    s_and_b32 s0, s0, s1
+; GFX8-NEXT:    v_or_b32_e32 v8, s0, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s8
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v1, s9
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, s10
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 2
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v3, s11
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 3
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v8, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v5, s13
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 5
+; GFX8-NEXT:    v_cndmask_b32_e32 v5, v5, v8, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v6, s14
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 6
+; GFX8-NEXT:    v_cndmask_b32_e32 v6, v6, v8, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v4, s12
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s2, 4
+; GFX8-NEXT:    v_mov_b32_e32 v7, s15
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 7
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v8, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v7, v7, v8, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8-NEXT:    s_add_u32 s0, 0, 16
+; GFX8-NEXT:    v_mov_b32_e32 v9, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GFX8-NEXT:    s_addc_u32 s1, 0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dwordx4 v[0:1], v[4:7]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v16i16_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx8 s[8:15], s[2:3], 0x0
+; GFX7-NEXT:    s_lshr_b32 s2, s4, 1
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX7-NEXT:    s_mov_b32 s3, 0xffff
+; GFX7-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_cselect_b32 s0, s9, s8
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 2
+; GFX7-NEXT:    s_cselect_b32 s0, s10, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 3
+; GFX7-NEXT:    s_cselect_b32 s0, s11, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 4
+; GFX7-NEXT:    s_cselect_b32 s0, s12, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 5
+; GFX7-NEXT:    s_cselect_b32 s0, s13, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 6
+; GFX7-NEXT:    s_cselect_b32 s0, s14, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 7
+; GFX7-NEXT:    s_cselect_b32 s0, s15, s0
+; GFX7-NEXT:    s_and_b32 s1, s4, 1
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s1, v0
+; GFX7-NEXT:    s_lshl_b32 s1, s3, s1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_and_b32 s0, s0, s1
+; GFX7-NEXT:    v_or_b32_e32 v8, s0, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s9
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, s10
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 2
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v8, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v3, s11
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 3
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v3, v8, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v5, s13
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 5
+; GFX7-NEXT:    v_cndmask_b32_e32 v5, v5, v8, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v4, s12
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s2, 4
+; GFX7-NEXT:    v_mov_b32_e32 v6, s14
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 6
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v8, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e32 v6, v6, v8, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 7
+; GFX7-NEXT:    v_mov_b32_e32 v7, s15
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    v_cndmask_b32_e32 v7, v7, v8, vcc
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i16>, <16 x i16> addrspace(4)* %ptr
+  %insert = insertelement <16 x i16> %vec, i16 %val, i32 %idx
+  store <16 x i16> %insert, <16 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v16i16_s_v(<16 x i16> addrspace(4)* inreg %ptr, i16 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v16i16_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx8 s[16:23], s[2:3], 0x0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 1, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v8
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s16
+; GFX9-NEXT:    v_mov_b32_e32 v2, s17
+; GFX9-NEXT:    v_mov_b32_e32 v3, s18
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v4, s19
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v5, s20
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[2:3]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[14:15], 4, v8
+; GFX9-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX9-NEXT:    s_mov_b32 s5, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v6, s21
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[14:15]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v8
+; GFX9-NEXT:    v_mov_b32_e32 v7, s22
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[6:7]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v8
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX9-NEXT:    s_and_b32 s4, s4, s5
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v0, s4
+; GFX9-NEXT:    v_lshlrev_b32_e64 v0, v0, s5
+; GFX9-NEXT:    v_mov_b32_e32 v9, s23
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[8:9]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v8
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[10:11]
+; GFX9-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX9-NEXT:    v_and_or_b32 v9, v1, v0, v2
+; GFX9-NEXT:    v_mov_b32_e32 v0, s16
+; GFX9-NEXT:    v_mov_b32_e32 v1, s17
+; GFX9-NEXT:    v_mov_b32_e32 v2, s18
+; GFX9-NEXT:    v_mov_b32_e32 v3, s19
+; GFX9-NEXT:    v_mov_b32_e32 v4, s20
+; GFX9-NEXT:    v_mov_b32_e32 v5, s21
+; GFX9-NEXT:    v_mov_b32_e32 v6, s22
+; GFX9-NEXT:    v_mov_b32_e32 v7, s23
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v8
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[12:13]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v4, v9, s[14:15]
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v6, v9, s[8:9]
+; GFX9-NEXT:    v_cndmask_b32_e64 v7, v7, v9, s[10:11]
+; GFX9-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-NEXT:    s_add_u32 s0, 0, 16
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0
+; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
+; GFX9-NEXT:    s_addc_u32 s1, 0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_store_dwordx4 v[0:1], v[4:7], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v16i16_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx8 s[16:23], s[2:3], 0x0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 1, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v8
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v8
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v1, s16
+; GFX8-NEXT:    v_mov_b32_e32 v2, s17
+; GFX8-NEXT:    v_mov_b32_e32 v3, s18
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v4, s19
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v5, s20
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[2:3]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[14:15], 4, v8
+; GFX8-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX8-NEXT:    s_mov_b32 s5, 0xffff
+; GFX8-NEXT:    v_mov_b32_e32 v6, s21
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[14:15]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v8
+; GFX8-NEXT:    v_mov_b32_e32 v7, s22
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[6:7]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v8
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX8-NEXT:    s_and_b32 s4, s4, s5
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v0, s4
+; GFX8-NEXT:    v_lshlrev_b32_e64 v0, v0, s5
+; GFX8-NEXT:    v_mov_b32_e32 v9, s23
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[8:9]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v8
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[10:11]
+; GFX8-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX8-NEXT:    v_or_b32_e32 v9, v0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v0, s16
+; GFX8-NEXT:    v_mov_b32_e32 v1, s17
+; GFX8-NEXT:    v_mov_b32_e32 v2, s18
+; GFX8-NEXT:    v_mov_b32_e32 v3, s19
+; GFX8-NEXT:    v_mov_b32_e32 v4, s20
+; GFX8-NEXT:    v_mov_b32_e32 v5, s21
+; GFX8-NEXT:    v_mov_b32_e32 v6, s22
+; GFX8-NEXT:    v_mov_b32_e32 v7, s23
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v8
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[12:13]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[2:3]
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v9, s[14:15]
+; GFX8-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v6, v9, s[8:9]
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v7, v9, s[10:11]
+; GFX8-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8-NEXT:    s_add_u32 s0, 0, 16
+; GFX8-NEXT:    v_mov_b32_e32 v9, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GFX8-NEXT:    s_addc_u32 s1, 0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dwordx4 v[0:1], v[4:7]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v16i16_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx8 s[16:23], s[2:3], 0x0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 1, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v8
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v8
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v1, s16
+; GFX7-NEXT:    v_mov_b32_e32 v2, s17
+; GFX7-NEXT:    v_mov_b32_e32 v3, s18
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v4, s19
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX7-NEXT:    v_mov_b32_e32 v5, s20
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[2:3]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[14:15], 4, v8
+; GFX7-NEXT:    v_and_b32_e32 v0, 1, v0
+; GFX7-NEXT:    s_mov_b32 s5, 0xffff
+; GFX7-NEXT:    v_mov_b32_e32 v6, s21
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[14:15]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v8
+; GFX7-NEXT:    v_mov_b32_e32 v7, s22
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[6:7]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
+; GFX7-NEXT:    s_and_b32 s4, s4, s5
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s4, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v0, s5, v0
+; GFX7-NEXT:    v_mov_b32_e32 v9, s23
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[8:9]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v8
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[10:11]
+; GFX7-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_or_b32_e32 v9, v0, v2
+; GFX7-NEXT:    v_mov_b32_e32 v0, s16
+; GFX7-NEXT:    v_mov_b32_e32 v1, s17
+; GFX7-NEXT:    v_mov_b32_e32 v2, s18
+; GFX7-NEXT:    v_mov_b32_e32 v3, s19
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v8
+; GFX7-NEXT:    v_mov_b32_e32 v4, s20
+; GFX7-NEXT:    v_mov_b32_e32 v5, s21
+; GFX7-NEXT:    v_mov_b32_e32 v6, s22
+; GFX7-NEXT:    v_mov_b32_e32 v7, s23
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[2:3]
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[12:13]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v9, s[14:15]
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v6, v6, v9, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v7, v7, v9, s[10:11]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i16>, <16 x i16> addrspace(4)* %ptr
+  %insert = insertelement <16 x i16> %vec, i16 %val, i32 %idx
+  store <16 x i16> %insert, <16 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v16i16_v_v(<16 x i16> addrspace(4)* inreg %ptr, i16 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v16i16_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx8 s[12:19], s[2:3], 0x0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 1, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v8
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v2, s12
+; GFX9-NEXT:    v_mov_b32_e32 v3, s13
+; GFX9-NEXT:    v_mov_b32_e32 v4, s14
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v5, s15
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v6, s16
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[2:3]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v8
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX9-NEXT:    v_mov_b32_e32 v7, s17
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[4:5]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v8
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    v_mov_b32_e32 v9, s18
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v7, s[6:7]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v8
+; GFX9-NEXT:    s_mov_b32 s20, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s20
+; GFX9-NEXT:    v_mov_b32_e32 v10, s19
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[8:9]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v8
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v10, s[10:11]
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_and_or_b32 v9, v2, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s12
+; GFX9-NEXT:    v_mov_b32_e32 v1, s13
+; GFX9-NEXT:    v_mov_b32_e32 v2, s14
+; GFX9-NEXT:    v_mov_b32_e32 v3, s15
+; GFX9-NEXT:    v_mov_b32_e32 v4, s16
+; GFX9-NEXT:    v_mov_b32_e32 v5, s17
+; GFX9-NEXT:    v_mov_b32_e32 v6, s18
+; GFX9-NEXT:    v_mov_b32_e32 v7, s19
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v8
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[12:13]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v4, v9, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v6, v9, s[8:9]
+; GFX9-NEXT:    v_cndmask_b32_e64 v7, v7, v9, s[10:11]
+; GFX9-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-NEXT:    s_add_u32 s0, 0, 16
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0
+; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
+; GFX9-NEXT:    s_addc_u32 s1, 0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_store_dwordx4 v[0:1], v[4:7], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v16i16_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx8 s[12:19], s[2:3], 0x0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 1, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v8
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v8
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v2, s12
+; GFX8-NEXT:    v_mov_b32_e32 v3, s13
+; GFX8-NEXT:    v_mov_b32_e32 v4, s14
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v5, s15
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v6, s16
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[2:3]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v8
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX8-NEXT:    v_mov_b32_e32 v7, s17
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[4:5]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v8
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    v_mov_b32_e32 v9, s18
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v7, s[6:7]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v8
+; GFX8-NEXT:    s_mov_b32 s20, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s20
+; GFX8-NEXT:    v_mov_b32_e32 v10, s19
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[8:9]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v8
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v10, s[10:11]
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX8-NEXT:    v_or_b32_e32 v9, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s12
+; GFX8-NEXT:    v_mov_b32_e32 v1, s13
+; GFX8-NEXT:    v_mov_b32_e32 v2, s14
+; GFX8-NEXT:    v_mov_b32_e32 v3, s15
+; GFX8-NEXT:    v_mov_b32_e32 v4, s16
+; GFX8-NEXT:    v_mov_b32_e32 v5, s17
+; GFX8-NEXT:    v_mov_b32_e32 v6, s18
+; GFX8-NEXT:    v_mov_b32_e32 v7, s19
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v8
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[12:13]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[2:3]
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v9, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v6, v9, s[8:9]
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v7, v9, s[10:11]
+; GFX8-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8-NEXT:    s_add_u32 s0, 0, 16
+; GFX8-NEXT:    v_mov_b32_e32 v9, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GFX8-NEXT:    s_addc_u32 s1, 0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dwordx4 v[0:1], v[4:7]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v16i16_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx8 s[12:19], s[2:3], 0x0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 1, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v8
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v8
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    v_mov_b32_e32 v2, s12
+; GFX7-NEXT:    v_mov_b32_e32 v3, s13
+; GFX7-NEXT:    v_mov_b32_e32 v4, s14
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v5, s15
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX7-NEXT:    v_mov_b32_e32 v6, s16
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[2:3]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v8
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v1
+; GFX7-NEXT:    s_mov_b32 s20, 0xffff
+; GFX7-NEXT:    v_mov_b32_e32 v7, s17
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[4:5]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v8
+; GFX7-NEXT:    v_mov_b32_e32 v9, s18
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v7, s[6:7]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, s20, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s20, v1
+; GFX7-NEXT:    v_mov_b32_e32 v10, s19
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[8:9]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v8
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v10, s[10:11]
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_or_b32_e32 v9, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s12
+; GFX7-NEXT:    v_mov_b32_e32 v1, s13
+; GFX7-NEXT:    v_mov_b32_e32 v2, s14
+; GFX7-NEXT:    v_mov_b32_e32 v3, s15
+; GFX7-NEXT:    v_mov_b32_e32 v4, s16
+; GFX7-NEXT:    v_mov_b32_e32 v5, s17
+; GFX7-NEXT:    v_mov_b32_e32 v6, s18
+; GFX7-NEXT:    v_mov_b32_e32 v7, s19
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v8
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[2:3]
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v9, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v9, s[12:13]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v9, vcc
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v9, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v6, v6, v9, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v7, v7, v9, s[10:11]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i16>, <16 x i16> addrspace(4)* %ptr
+  %insert = insertelement <16 x i16> %vec, i16 %val, i32 %idx
+  store <16 x i16> %insert, <16 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v16i16_s_v(<16 x i16> addrspace(1)* %ptr, i16 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v16i16_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[3:6], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx4 v[7:10], v[0:1], off offset:16
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 1, v2
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_and_b32 s1, s2, s0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v1, s1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v0
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v0
+; GFX9-NEXT:    s_waitcnt vmcnt(1)
+; GFX9-NEXT:    v_cndmask_b32_e32 v11, v3, v4, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v11, v11, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v11, v11, v6, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e64 v11, v11, v7, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v11, v11, v8, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v11, v11, v9, s[8:9]
+; GFX9-NEXT:    v_cndmask_b32_e64 v11, v11, v10, s[10:11]
+; GFX9-NEXT:    v_and_or_b32 v11, v11, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v5, v11, s[0:1]
+; GFX9-NEXT:    s_add_u32 s0, 0, 16
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v3, v11, s[12:13]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v6, v11, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v8, v11, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v9, v11, s[8:9]
+; GFX9-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-NEXT:    s_addc_u32 s1, 0, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v4, v11, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v7, v11, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v7, v10, v11, s[10:11]
+; GFX9-NEXT:    v_mov_b32_e32 v11, s1
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0
+; GFX9-NEXT:    v_mov_b32_e32 v10, s0
+; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
+; GFX9-NEXT:    global_store_dwordx4 v[10:11], v[4:7], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v16i16_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[3:6], v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, 16, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dwordx4 v[7:10], v[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, 1, v2
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_and_b32 s1, s2, s0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v1, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v0
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v0
+; GFX8-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
+; GFX8-NEXT:    v_cndmask_b32_e32 v11, v3, v4, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v11, v11, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v11, v11, v6, s[2:3]
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e64 v11, v11, v7, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v11, v11, v8, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v11, v11, v9, s[8:9]
+; GFX8-NEXT:    v_cndmask_b32_e64 v11, v11, v10, s[10:11]
+; GFX8-NEXT:    v_and_b32_e32 v1, v11, v1
+; GFX8-NEXT:    v_or_b32_e32 v11, v1, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v5, v11, s[0:1]
+; GFX8-NEXT:    s_add_u32 s0, 0, 16
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v3, v11, s[12:13]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v6, v11, s[2:3]
+; GFX8-NEXT:    v_cndmask_b32_e64 v5, v8, v11, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v9, v11, s[8:9]
+; GFX8-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8-NEXT:    s_addc_u32 s1, 0, 0
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v4, v11, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v7, v11, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v10, v11, s[10:11]
+; GFX8-NEXT:    v_mov_b32_e32 v11, s1
+; GFX8-NEXT:    v_mov_b32_e32 v9, 0
+; GFX8-NEXT:    v_mov_b32_e32 v10, s0
+; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GFX8-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v16i16_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s18, 0
+; GFX7-NEXT:    s_mov_b32 s19, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[16:17], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[3:6], v[0:1], s[16:19], 0 addr64
+; GFX7-NEXT:    buffer_load_dwordx4 v[7:10], v[0:1], s[16:19], 0 addr64 offset:16
+; GFX7-NEXT:    v_lshrrev_b32_e32 v0, 1, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v2
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_and_b32 s1, s2, s0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s1, v1
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v0
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v0
+; GFX7-NEXT:    s_mov_b32 s18, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(1)
+; GFX7-NEXT:    v_cndmask_b32_e32 v11, v3, v4, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v11, v11, v5, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v11, v11, v6, s[2:3]
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e64 v11, v11, v7, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v11, v11, v8, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v11, v11, v9, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v11, v11, v10, s[10:11]
+; GFX7-NEXT:    v_and_b32_e32 v1, v11, v1
+; GFX7-NEXT:    v_or_b32_e32 v11, v1, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v3, v11, s[12:13]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v4, v11, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v5, v11, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v6, v11, s[2:3]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v7, v11, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v8, v11, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v6, v9, v11, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v7, v10, v11, s[10:11]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[16:19], 0
+; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[16:19], 0 offset:16
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i16>, <16 x i16> addrspace(1)* %ptr
+  %insert = insertelement <16 x i16> %vec, i16 %val, i32 %idx
+  store <16 x i16> %insert, <16 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v16i16_v_s(<16 x i16> addrspace(1)* %ptr, i16 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v16i16_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[3:6], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx4 v[7:10], v[0:1], off offset:16
+; GFX9-NEXT:    s_and_b32 s1, s2, 1
+; GFX9-NEXT:    s_lshr_b32 s12, s2, 1
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s12, 1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, s1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    s_not_b32 s13, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s12, 2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s12, 3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], s12, 4
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[6:7], s12, 5
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[8:9], s12, 6
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[10:11], s12, 7
+; GFX9-NEXT:    s_waitcnt vmcnt(1)
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v8, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[8:9]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v10, s[10:11]
+; GFX9-NEXT:    v_and_or_b32 v11, v1, s13, v0
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v5, v11, s[0:1]
+; GFX9-NEXT:    s_add_u32 s0, 0, 16
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[12:13], s12, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v3, v11, s[12:13]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v6, v11, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v8, v11, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v9, v11, s[8:9]
+; GFX9-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-NEXT:    s_addc_u32 s1, 0, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v4, v11, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v7, v11, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v7, v10, v11, s[10:11]
+; GFX9-NEXT:    v_mov_b32_e32 v11, s1
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0
+; GFX9-NEXT:    v_mov_b32_e32 v10, s0
+; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
+; GFX9-NEXT:    global_store_dwordx4 v[10:11], v[4:7], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v16i16_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[3:6], v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, 16, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dwordx4 v[7:10], v[0:1]
+; GFX8-NEXT:    s_and_b32 s1, s2, 1
+; GFX8-NEXT:    s_lshr_b32 s12, s2, 1
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s12, 1
+; GFX8-NEXT:    v_mov_b32_e32 v0, s1
+; GFX8-NEXT:    s_not_b32 s13, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s12, 2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], s12, 3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], s12, 4
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[6:7], s12, 5
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[8:9], s12, 6
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[10:11], s12, 7
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[2:3]
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v8, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[8:9]
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v10, s[10:11]
+; GFX8-NEXT:    v_and_b32_e32 v1, s13, v1
+; GFX8-NEXT:    v_or_b32_e32 v11, v1, v0
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v5, v11, s[0:1]
+; GFX8-NEXT:    s_add_u32 s0, 0, 16
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[12:13], s12, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v3, v11, s[12:13]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v6, v11, s[2:3]
+; GFX8-NEXT:    v_cndmask_b32_e64 v5, v8, v11, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v9, v11, s[8:9]
+; GFX8-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8-NEXT:    s_addc_u32 s1, 0, 0
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v4, v11, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v7, v11, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v10, v11, s[10:11]
+; GFX8-NEXT:    v_mov_b32_e32 v11, s1
+; GFX8-NEXT:    v_mov_b32_e32 v9, 0
+; GFX8-NEXT:    v_mov_b32_e32 v10, s0
+; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GFX8-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v16i16_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s18, 0
+; GFX7-NEXT:    s_mov_b32 s19, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[16:17], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[3:6], v[0:1], s[16:19], 0 addr64
+; GFX7-NEXT:    buffer_load_dwordx4 v[7:10], v[0:1], s[16:19], 0 addr64 offset:16
+; GFX7-NEXT:    s_and_b32 s1, s2, 1
+; GFX7-NEXT:    s_lshr_b32 s12, s2, 1
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 4
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v2
+; GFX7-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s12, 1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s1, v0
+; GFX7-NEXT:    s_not_b32 s13, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s12, 2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], s12, 3
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], s12, 4
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[6:7], s12, 5
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[8:9], s12, 6
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[10:11], s12, 7
+; GFX7-NEXT:    s_mov_b32 s18, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(1)
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[2:3]
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v8, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v10, s[10:11]
+; GFX7-NEXT:    v_and_b32_e32 v1, s13, v1
+; GFX7-NEXT:    v_or_b32_e32 v11, v1, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[12:13], s12, 0
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v3, v11, s[12:13]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v4, v11, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v5, v11, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v6, v11, s[2:3]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v7, v11, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v8, v11, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v6, v9, v11, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v7, v10, v11, s[10:11]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[16:19], 0
+; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[16:19], 0 offset:16
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i16>, <16 x i16> addrspace(1)* %ptr
+  %insert = insertelement <16 x i16> %vec, i16 %val, i32 %idx
+  store <16 x i16> %insert, <16 x i16> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v16i16_v_v(<16 x i16> addrspace(1)* %ptr, i16 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v16i16_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx4 v[8:11], v[0:1], off offset:16
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 1, v3
+; GFX9-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT:    s_mov_b32 s0, 0xffff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v0
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v0
+; GFX9-NEXT:    s_waitcnt vmcnt(1)
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v7, s[2:3]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v8, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[6:7]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v10, s[8:9]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v11, s[10:11]
+; GFX9-NEXT:    v_and_or_b32 v12, v3, v1, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v6, v12, s[0:1]
+; GFX9-NEXT:    s_add_u32 s0, 0, 16
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v4, v12, s[12:13]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v5, v12, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v8, v12, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v9, v12, s[6:7]
+; GFX9-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-NEXT:    s_addc_u32 s1, 0, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v7, v12, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v10, v12, s[8:9]
+; GFX9-NEXT:    v_cndmask_b32_e64 v7, v11, v12, s[10:11]
+; GFX9-NEXT:    v_mov_b32_e32 v11, s1
+; GFX9-NEXT:    v_mov_b32_e32 v9, 0
+; GFX9-NEXT:    v_mov_b32_e32 v10, s0
+; GFX9-NEXT:    global_store_dwordx4 v[8:9], v[0:3], off
+; GFX9-NEXT:    global_store_dwordx4 v[10:11], v[4:7], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v16i16_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[4:7], v[0:1]
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, 16, v0
+; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX8-NEXT:    flat_load_dwordx4 v[8:11], v[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, 1, v3
+; GFX8-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX8-NEXT:    s_mov_b32 s0, 0xffff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v0
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v0
+; GFX8-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v7, s[2:3]
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v8, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[6:7]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v10, s[8:9]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v11, s[10:11]
+; GFX8-NEXT:    v_and_b32_e32 v1, v3, v1
+; GFX8-NEXT:    v_or_b32_e32 v12, v1, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v6, v12, s[0:1]
+; GFX8-NEXT:    s_add_u32 s0, 0, 16
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v4, v12, s[12:13]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v5, v12, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v8, v12, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v5, v9, v12, s[6:7]
+; GFX8-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8-NEXT:    s_addc_u32 s1, 0, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v7, v12, s[2:3]
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v10, v12, s[8:9]
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v11, v12, s[10:11]
+; GFX8-NEXT:    v_mov_b32_e32 v11, s1
+; GFX8-NEXT:    v_mov_b32_e32 v9, 0
+; GFX8-NEXT:    v_mov_b32_e32 v10, s0
+; GFX8-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GFX8-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v16i16_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s18, 0
+; GFX7-NEXT:    s_mov_b32 s19, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[16:17], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[4:7], v[0:1], s[16:19], 0 addr64
+; GFX7-NEXT:    buffer_load_dwordx4 v[8:11], v[0:1], s[16:19], 0 addr64 offset:16
+; GFX7-NEXT:    v_lshrrev_b32_e32 v0, 1, v3
+; GFX7-NEXT:    v_and_b32_e32 v1, 1, v3
+; GFX7-NEXT:    s_mov_b32 s0, 0xffff
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, v1, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 4, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[6:7], 5, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[8:9], 6, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[10:11], 7, v0
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[12:13], 0, v0
+; GFX7-NEXT:    s_mov_b32 s18, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(1)
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v7, s[2:3]
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v8, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v9, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v10, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v11, s[10:11]
+; GFX7-NEXT:    v_and_b32_e32 v1, v3, v1
+; GFX7-NEXT:    v_or_b32_e32 v12, v1, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v4, v12, s[12:13]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v5, v12, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v6, v12, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v7, v12, s[2:3]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v8, v12, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v9, v12, s[6:7]
+; GFX7-NEXT:    v_cndmask_b32_e64 v6, v10, v12, s[8:9]
+; GFX7-NEXT:    v_cndmask_b32_e64 v7, v11, v12, s[10:11]
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[16:19], 0
+; GFX7-NEXT:    buffer_store_dwordx4 v[4:7], off, s[16:19], 0 offset:16
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i16>, <16 x i16> addrspace(1)* %ptr
+  %insert = insertelement <16 x i16> %vec, i16 %val, i32 %idx
+  store <16 x i16> %insert, <16 x i16> addrspace(1)* null
+  ret void
+}

llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.i8.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9 %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX8 %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX7 %
+
+define amdgpu_ps void @insertelement_s_v2i8_s_s(<2 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v2i8_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX9-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v2i8_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    v_mov_b32_e32 v0, s2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    flat_load_ushort v0, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v2, s4
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX8-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v2i8_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s0, s2
+; GFX7-NEXT:    s_mov_b32 s1, s3
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_load_ushort v1, off, s[0:3], 0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s4
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0xff
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v3, v0, vcc
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i8>, <2 x i8> addrspace(4)* %ptr
+  %insert = insertelement <2 x i8> %vec, i8 %val, i32 %idx
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v2i8_s_s(<2 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v2i8_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s3, 0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s3, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX9-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v2i8_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_ushort v0, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s3, 0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s3, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX8-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v2i8_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    v_mov_b32_e32 v2, s2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s3, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0xff
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 8, v0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s3, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    buffer_store_short v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i8>, <2 x i8> addrspace(1 )* %ptr
+  %insert = insertelement <2 x i8> %vec, i8 %val, i32 %idx
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v2i8_v_s(<2 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v2i8_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9-NEXT:    v_mov_b32_e32 v2, s3
+; GFX9-NEXT:    global_load_ushort v1, v[1:2], off
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, 8, v0
+; GFX9-NEXT:    v_or_b32_sdwa v2, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v2i8_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    v_mov_b32_e32 v1, s2
+; GFX8-NEXT:    v_mov_b32_e32 v2, s3
+; GFX8-NEXT:    flat_load_ushort v1, v[1:2]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
+; GFX8-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX8-NEXT:    v_lshlrev_b16_e32 v0, 8, v0
+; GFX8-NEXT:    v_or_b32_sdwa v2, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v2i8_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s0, s2
+; GFX7-NEXT:    s_mov_b32 s1, s3
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_load_ushort v1, off, s[0:3], 0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 0
+; GFX7-NEXT:    v_mov_b32_e32 v2, 0xff
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v3, v0, vcc
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i8>, <2 x i8> addrspace(4)* %ptr
+  %insert = insertelement <2 x i8> %vec, i8 %val, i32 %idx
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v2i8_s_v(<2 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v2i8_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9-NEXT:    v_mov_b32_e32 v2, s3
+; GFX9-NEXT:    global_load_ushort v1, v[1:2], off
+; GFX9-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, 8, v0
+; GFX9-NEXT:    v_or_b32_sdwa v2, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v2i8_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    v_mov_b32_e32 v1, s2
+; GFX8-NEXT:    v_mov_b32_e32 v2, s3
+; GFX8-NEXT:    flat_load_ushort v1, v[1:2]
+; GFX8-NEXT:    v_mov_b32_e32 v3, s4
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
+; GFX8-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX8-NEXT:    v_lshlrev_b16_e32 v0, 8, v0
+; GFX8-NEXT:    v_or_b32_sdwa v2, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v2i8_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s0, s2
+; GFX7-NEXT:    s_mov_b32 s1, s3
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_load_ushort v2, off, s[0:3], 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s4
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0xff
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v2
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v1, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v4, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i8>, <2 x i8> addrspace(4)* %ptr
+  %insert = insertelement <2 x i8> %vec, i8 %val, i32 %idx
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v2i8_v_v(<2 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v2i8_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NEXT:    global_load_ushort v2, v[2:3], off
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v0, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v1
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v3, v0, vcc
+; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX9-NEXT:    v_lshlrev_b16_e32 v0, 8, v0
+; GFX9-NEXT:    v_or_b32_sdwa v2, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v2i8_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    flat_load_ushort v2, v[2:3]
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 8, v2
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v0, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v1
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v3, v0, vcc
+; GFX8-NEXT:    v_and_b32_e32 v0, 0xff, v0
+; GFX8-NEXT:    v_lshlrev_b16_e32 v0, 8, v0
+; GFX8-NEXT:    v_or_b32_sdwa v2, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v2i8_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s0, s2
+; GFX7-NEXT:    s_mov_b32 s1, s3
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_load_ushort v2, off, s[0:3], 0
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
+; GFX7-NEXT:    v_mov_b32_e32 v3, 0xff
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v2
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v0, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v1
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v4, v0, vcc
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i8>, <2 x i8> addrspace(4)* %ptr
+  %insert = insertelement <2 x i8> %vec, i8 %val, i32 %idx
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v2i8_s_v(<2 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v2i8_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX9-NEXT:    v_mov_b32_e32 v3, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX9-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v2i8_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_ushort v0, v[0:1]
+; GFX8-NEXT:    v_mov_b32_e32 v3, s2
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX8-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX8-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v2i8_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    v_mov_b32_e32 v3, s2
+; GFX7-NEXT:    s_mov_b32 s2, 0
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    buffer_load_ushort v0, v[0:1], s[0:3], 0 addr64
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0xff
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v4, v3, vcc
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i8>, <2 x i8> addrspace(1)* %ptr
+  %insert = insertelement <2 x i8> %vec, i8 %val, i32 %idx
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v2i8_v_s(<2 x i8> addrspace(1)* %ptr, i8 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v2i8_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX9-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v2i8_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_ushort v0, v[0:1]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX8-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v2i8_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0xff
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 8, v0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    buffer_store_short v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i8>, <2 x i8> addrspace(1)* %ptr
+  %insert = insertelement <2 x i8> %vec, i8 %val, i32 %idx
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v2i8_v_v(<2 x i8> addrspace(1)* %ptr, i8 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v2i8_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_ushort v0, v[0:1], off
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v3
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX9-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_short v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v2i8_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_ushort v0, v[0:1]
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v3
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; GFX8-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
+; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_short v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v2i8_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s2, 0
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    buffer_load_ushort v0, v[0:1], s[0:3], 0 addr64
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0xff
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v3
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <2 x i8>, <2 x i8> addrspace(1)* %ptr
+  %insert = insertelement <2 x i8> %vec, i8 %val, i32 %idx
+  store <2 x i8> %insert, <2 x i8> addrspace(1)* null
+  ret void
+}
+
+; FIXME: 3 element load/store legalization
+; define amdgpu_ps void @insertelement_s_v3i8_s_s(<3 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 inreg %idx) {
+;   %vec = load <3 x i8>, <3 x i8> addrspace(4)* %ptr
+;   %insert = insertelement <3 x i8> %vec, i8 %val, i32 %idx
+;   store <3 x i8> %insert, <3 x i8> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_v_v3i8_s_s(<3 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 inreg %idx) {
+;   %vec = load <3 x i8>, <3 x i8> addrspace(1 )* %ptr
+;   %insert = insertelement <3 x i8> %vec, i8 %val, i32 %idx
+;   store <3 x i8> %insert, <3 x i8> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_s_v3i8_v_s(<3 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 inreg %idx) {
+;   %vec = load <3 x i8>, <3 x i8> addrspace(4)* %ptr
+;   %insert = insertelement <3 x i8> %vec, i8 %val, i32 %idx
+;   store <3 x i8> %insert, <3 x i8> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_s_v3i8_s_v(<3 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 %idx) {
+;   %vec = load <3 x i8>, <3 x i8> addrspace(4)* %ptr
+;   %insert = insertelement <3 x i8> %vec, i8 %val, i32 %idx
+;   store <3 x i8> %insert, <3 x i8> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_s_v3i8_v_v(<3 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 %idx) {
+;   %vec = load <3 x i8>, <3 x i8> addrspace(4)* %ptr
+;   %insert = insertelement <3 x i8> %vec, i8 %val, i32 %idx
+;   store <3 x i8> %insert, <3 x i8> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_v_v3i8_s_v(<3 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 %idx) {
+;   %vec = load <3 x i8>, <3 x i8> addrspace(1)* %ptr
+;   %insert = insertelement <3 x i8> %vec, i8 %val, i32 %idx
+;   store <3 x i8> %insert, <3 x i8> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_v_v3i8_v_s(<3 x i8> addrspace(1)* %ptr, i8 %val, i32 inreg %idx) {
+;   %vec = load <3 x i8>, <3 x i8> addrspace(1)* %ptr
+;   %insert = insertelement <3 x i8> %vec, i8 %val, i32 %idx
+;   store <3 x i8> %insert, <3 x i8> addrspace(1)* null
+;   ret void
+; }
+
+; define amdgpu_ps void @insertelement_v_v3i8_v_v(<3 x i8> addrspace(1)* %ptr, i8 %val, i32 %idx) {
+;   %vec = load <3 x i8>, <3 x i8> addrspace(1)* %ptr
+;   %insert = insertelement <3 x i8> %vec, i8 %val, i32 %idx
+;   store <3 x i8> %insert, <3 x i8> addrspace(1)* null
+;   ret void
+; }
+
+define amdgpu_ps void @insertelement_v_v4i8_s_s(<4 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v4i8_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    s_movk_i32 s1, 0xff
+; GFX9-NEXT:    s_and_b32 s3, s3, 3
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s1
+; GFX9-NEXT:    s_lshl_b32 s3, s3, 3
+; GFX9-NEXT:    s_lshl_b32 s2, s2, s3
+; GFX9-NEXT:    s_lshl_b32 s3, s1, s3
+; GFX9-NEXT:    s_not_b32 s3, s3
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, 8
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, s0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v4, v0, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v5, v0, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s1, v3
+; GFX9-NEXT:    v_or3_b32 v0, v0, v4, v5
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s3, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v3, v0, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v4, v0, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s1, v1
+; GFX9-NEXT:    v_or3_b32 v2, v0, v3, v4
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v4i8_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dword v0, v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v1, 8
+; GFX8-NEXT:    v_mov_b32_e32 v3, s0
+; GFX8-NEXT:    s_and_b32 s1, s3, 3
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX8-NEXT:    s_and_b32 s2, s2, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v5, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v6, v0, v3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX8-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX8-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v3, v0, v3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX8-NEXT:    v_or_b32_e32 v2, v0, v3
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v4i8_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s0, 0xff
+; GFX7-NEXT:    s_and_b32 s1, s3, 3
+; GFX7-NEXT:    s_and_b32 s2, s2, s0
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX7-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX7-NEXT:    s_lshl_b32 s1, s0, s1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 24, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_and_b32_e32 v0, s1, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, s2, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 24, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i8>, <4 x i8> addrspace(1 )* %ptr
+  %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
+  store <4 x i8> %insert, <4 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v4i8_v_s(<4 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v4i8_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s1, s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s6, 0xff
+; GFX9-NEXT:    v_and_b32_e32 v0, s6, v0
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s2, s1, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s6
+; GFX9-NEXT:    s_lshr_b32 s3, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s5, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s6
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s3, s6
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s5, s6
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s4, 3
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s3, s6, s2
+; GFX9-NEXT:    s_not_b32 s3, s3
+; GFX9-NEXT:    s_and_b32 s1, s1, s3
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_lshl_or_b32 v0, v0, s2, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, s0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v1, v0, s6, v1
+; GFX9-NEXT:    v_and_b32_sdwa v2, v0, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v2, v1, v2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v4i8_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s5, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s1, s0, 8
+; GFX8-NEXT:    s_and_b32 s1, s1, s5
+; GFX8-NEXT:    s_lshr_b32 s2, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s3, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s5
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s2, s5
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s3, s5
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s4, 3
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    s_lshl_b32 s1, s5, s1
+; GFX8-NEXT:    s_not_b32 s1, s1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    s_and_b32 s0, s0, s1
+; GFX8-NEXT:    v_or_b32_e32 v0, s0, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v2, s5
+; GFX8-NEXT:    v_or_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v3, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v2 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v4i8_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s5, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v0, s5, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s1, s0, 8
+; GFX7-NEXT:    s_and_b32 s1, s1, s5
+; GFX7-NEXT:    s_lshr_b32 s2, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s3, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s5
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s2, s5
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s3, s5
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s4, 3
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s1, v0
+; GFX7-NEXT:    s_lshl_b32 s1, s5, s1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_and_b32 s0, s0, s1
+; GFX7-NEXT:    v_or_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v1, s5, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s5, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s5, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s5, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i8>, <4 x i8> addrspace(4)* %ptr
+  %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
+  store <4 x i8> %insert, <4 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v4i8_s_v(<4 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v4i8_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s1, s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s6, 0xff
+; GFX9-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s2, s1, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s6
+; GFX9-NEXT:    s_lshr_b32 s3, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s5, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s6
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s3, s6
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s5, s6
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s4, s6
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v0, s2
+; GFX9-NEXT:    v_lshlrev_b32_e64 v0, v0, s6
+; GFX9-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX9-NEXT:    v_and_or_b32 v0, s1, v0, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, s0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v1, v0, s6, v1
+; GFX9-NEXT:    v_and_b32_sdwa v2, v0, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v2, v1, v2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v4i8_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s5, 0xff
+; GFX8-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s1, s0, 8
+; GFX8-NEXT:    s_and_b32 s1, s1, s5
+; GFX8-NEXT:    s_lshr_b32 s2, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s3, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s5
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s2, s5
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s3, s5
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s4, s5
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v0, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v0, v0, s5
+; GFX8-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v2, s5
+; GFX8-NEXT:    v_or_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v3, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v2 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v4i8_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s5, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s1, s0, 8
+; GFX7-NEXT:    s_and_b32 s1, s1, s5
+; GFX7-NEXT:    s_lshr_b32 s2, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s3, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s5
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s2, s5
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s3, s5
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s4, s5
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v0, s5, v0
+; GFX7-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v1, s5, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s5, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s5, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s5, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i8>, <4 x i8> addrspace(4)* %ptr
+  %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
+  store <4 x i8> %insert, <4 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v4i8_v_v(<4 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v4i8_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s1, s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s5, 0xff
+; GFX9-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s2, s1, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s5
+; GFX9-NEXT:    s_lshr_b32 s3, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s4, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s5
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s3, s5
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s4, s5
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s5
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_and_or_b32 v0, s1, v1, v0
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, s0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v1, v0, s5, v1
+; GFX9-NEXT:    v_and_b32_sdwa v2, v0, s5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v2, v1, v2, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v4i8_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s4, 0xff
+; GFX8-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s1, s0, 8
+; GFX8-NEXT:    s_and_b32 s1, s1, s4
+; GFX8-NEXT:    s_lshr_b32 s2, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s3, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s4
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s2, s4
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s3, s4
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s4
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX8-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v2, s4
+; GFX8-NEXT:    v_or_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v3, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v3
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v2 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v4i8_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dword s0, s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s4, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, s4, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s1, s0, 8
+; GFX7-NEXT:    s_and_b32 s1, s1, s4
+; GFX7-NEXT:    s_lshr_b32 s2, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s3, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s4
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s2, s4
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s3, s4
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s4, v1
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v1, s4, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s4, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s4, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s4, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i8>, <4 x i8> addrspace(4)* %ptr
+  %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
+  store <4 x i8> %insert, <4 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v4i8_s_v(<4 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v4i8_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    s_movk_i32 s1, 0xff
+; GFX9-NEXT:    v_and_b32_e32 v2, 3, v2
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v2
+; GFX9-NEXT:    v_lshlrev_b32_e64 v3, v2, s2
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v2, s1
+; GFX9-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX9-NEXT:    v_mov_b32_e32 v1, 8
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, s0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v5, v0, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v6, v0, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s1, v4
+; GFX9-NEXT:    v_or3_b32 v0, v0, v5, v6
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v2, v3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v3, v0, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v4, v0, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s1, v1
+; GFX9-NEXT:    v_or3_b32 v2, v0, v3, v4
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v4i8_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dword v0, v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v1, 8
+; GFX8-NEXT:    v_mov_b32_e32 v4, s0
+; GFX8-NEXT:    v_and_b32_e32 v2, 3, v2
+; GFX8-NEXT:    s_and_b32 s1, s2, s0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 3, v2
+; GFX8-NEXT:    v_lshlrev_b32_e64 v5, v2, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v2, s0
+; GFX8-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX8-NEXT:    v_mov_b32_e32 v3, 8
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v1, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v7, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v8, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v8
+; GFX8-NEXT:    v_and_b32_e32 v0, v0, v2
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v2, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v4, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, v0, v4
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v4i8_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s0, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v1, 3, v2
+; GFX7-NEXT:    s_and_b32 s1, s2, s0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s1, v1
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 16, v0
+; GFX7-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v4, s0, v4
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
+; GFX7-NEXT:    v_and_b32_e32 v5, s0, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 24, v5
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i8>, <4 x i8> addrspace(1)* %ptr
+  %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
+  store <4 x i8> %insert, <4 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v4i8_v_s(<4 x i8> addrspace(1)* %ptr, i8 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v4i8_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    s_movk_i32 s1, 0xff
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s2, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    s_lshl_b32 s2, s1, s2
+; GFX9-NEXT:    s_not_b32 s2, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, 8
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, s0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v4, v0, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v5, v0, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s1, v3
+; GFX9-NEXT:    v_or3_b32 v0, v0, v4, v5
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s2, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v3, v0, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v4, v0, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s1, v1
+; GFX9-NEXT:    v_or3_b32 v2, v0, v3, v4
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v4i8_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dword v0, v[0:1]
+; GFX8-NEXT:    s_and_b32 s1, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX8-NEXT:    v_mov_b32_e32 v5, s1
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v5, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 8
+; GFX8-NEXT:    v_mov_b32_e32 v4, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 8
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v6, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v7, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX8-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v2, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v4, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, v0, v4
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v4i8_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s0, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v2
+; GFX7-NEXT:    s_and_b32 s1, s2, 3
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, s1, v1
+; GFX7-NEXT:    s_lshl_b32 s1, s0, s1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_and_b32_e32 v4, s0, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v0, s1, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 24, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i8>, <4 x i8> addrspace(1)* %ptr
+  %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
+  store <4 x i8> %insert, <4 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v4i8_v_v(<4 x i8> addrspace(1)* %ptr, i8 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v4i8_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dword v0, v[0:1], off
+; GFX9-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    s_movk_i32 s1, 0xff
+; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v3, v3, s1
+; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX9-NEXT:    v_mov_b32_e32 v4, 8
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, s0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v6, v0, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v7, v0, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s1, v5
+; GFX9-NEXT:    v_or3_b32 v0, v0, v6, v7
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v3, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v3, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v5, v0, v1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v1, v2
+; GFX9-NEXT:    v_or3_b32 v2, v0, v3, v5
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    global_store_dword v[0:1], v2, off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v4i8_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dword v0, v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX8-NEXT:    v_mov_b32_e32 v6, s0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v3, v3, s0
+; GFX8-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX8-NEXT:    v_mov_b32_e32 v5, 8
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0xff
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v8, v0, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v6, v0, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v8
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX8-NEXT:    v_and_b32_e32 v0, v0, v3
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v5, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v3, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX8-NEXT:    v_or_b32_e32 v2, v0, v1
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0
+; GFX8-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v4i8_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s2, 0
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    buffer_load_dword v0, v[0:1], s[0:3], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s4, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX7-NEXT:    v_and_b32_e32 v2, s4, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v3, s4, v3
+; GFX7-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX7-NEXT:    v_mov_b32_e32 v1, 0xff
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
+; GFX7-NEXT:    v_and_b32_e32 v4, s4, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v5, s4, v5
+; GFX7-NEXT:    v_and_b32_e32 v0, s4, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_and_b32_e32 v6, s4, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 24, v6
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, v2, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, v3, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, v4, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <4 x i8>, <4 x i8> addrspace(1)* %ptr
+  %insert = insertelement <4 x i8> %vec, i8 %val, i32 %idx
+  store <4 x i8> %insert, <4 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v8i8_s_s(<8 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v8i8_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s10, 0xff
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s10
+; GFX9-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s3, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s6, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s7, s10
+; GFX9-NEXT:    s_lshr_b32 s8, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s8, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s9, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_lshr_b32 s2, s5, 2
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX9-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX9-NEXT:    s_and_b32 s5, s5, 3
+; GFX9-NEXT:    s_lshl_b32 s5, s5, 3
+; GFX9-NEXT:    s_and_b32 s4, s4, s10
+; GFX9-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX9-NEXT:    s_lshl_b32 s5, s10, s5
+; GFX9-NEXT:    s_not_b32 s5, s5
+; GFX9-NEXT:    s_and_b32 s3, s3, s5
+; GFX9-NEXT:    s_or_b32 s3, s3, s4
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 0
+; GFX9-NEXT:    s_cselect_b32 s0, s3, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX9-NEXT:    s_cselect_b32 s1, s3, s1
+; GFX9-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s10
+; GFX9-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s4, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s3, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s4, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_lshr_b32 s5, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s5, s10
+; GFX9-NEXT:    s_lshr_b32 s6, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s7, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s6, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s7, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v8i8_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s10, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX8-NEXT:    s_and_b32 s2, s2, s10
+; GFX8-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s3, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s6, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s7, s10
+; GFX8-NEXT:    s_lshr_b32 s8, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s9, s1, 24
+; GFX8-NEXT:    s_and_b32 s1, s1, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s8, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s9, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_lshr_b32 s2, s5, 2
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX8-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX8-NEXT:    s_and_b32 s5, s5, 3
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 3
+; GFX8-NEXT:    s_and_b32 s4, s4, s10
+; GFX8-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX8-NEXT:    s_lshl_b32 s5, s10, s5
+; GFX8-NEXT:    s_not_b32 s5, s5
+; GFX8-NEXT:    s_and_b32 s3, s3, s5
+; GFX8-NEXT:    s_or_b32 s3, s3, s4
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 0
+; GFX8-NEXT:    s_cselect_b32 s0, s3, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX8-NEXT:    s_cselect_b32 s1, s3, s1
+; GFX8-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX8-NEXT:    s_and_b32 s2, s2, s10
+; GFX8-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s4, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s3, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s4, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_lshr_b32 s5, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s5, s10
+; GFX8-NEXT:    s_lshr_b32 s6, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s7, s1, 24
+; GFX8-NEXT:    s_and_b32 s1, s1, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s6, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s7, s10
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v8i8_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s10, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX7-NEXT:    s_and_b32 s2, s2, s10
+; GFX7-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s10
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s3, s10
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s6, s10
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX7-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s7, s10
+; GFX7-NEXT:    s_lshr_b32 s8, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s9, s1, 24
+; GFX7-NEXT:    s_and_b32 s1, s1, s10
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_and_b32 s2, s8, s10
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_and_b32 s2, s9, s10
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_lshr_b32 s2, s5, 2
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX7-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX7-NEXT:    s_and_b32 s5, s5, 3
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 3
+; GFX7-NEXT:    s_and_b32 s4, s4, s10
+; GFX7-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX7-NEXT:    s_lshl_b32 s5, s10, s5
+; GFX7-NEXT:    s_not_b32 s5, s5
+; GFX7-NEXT:    s_and_b32 s3, s3, s5
+; GFX7-NEXT:    s_or_b32 s3, s3, s4
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 0
+; GFX7-NEXT:    s_cselect_b32 s4, s3, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX7-NEXT:    s_cselect_b32 s3, s3, s1
+; GFX7-NEXT:    s_lshr_b32 s2, s4, 8
+; GFX7-NEXT:    s_and_b32 s2, s2, s10
+; GFX7-NEXT:    s_lshr_b32 s5, s4, 16
+; GFX7-NEXT:    s_lshr_b32 s6, s4, 24
+; GFX7-NEXT:    s_and_b32 s4, s4, s10
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s2, s4, s2
+; GFX7-NEXT:    s_and_b32 s4, s5, s10
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX7-NEXT:    s_or_b32 s2, s2, s4
+; GFX7-NEXT:    s_and_b32 s4, s6, s10
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX7-NEXT:    s_lshr_b32 s7, s3, 8
+; GFX7-NEXT:    s_or_b32 s2, s2, s4
+; GFX7-NEXT:    s_and_b32 s4, s7, s10
+; GFX7-NEXT:    s_lshr_b32 s8, s3, 16
+; GFX7-NEXT:    s_lshr_b32 s9, s3, 24
+; GFX7-NEXT:    s_and_b32 s3, s3, s10
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX7-NEXT:    s_or_b32 s3, s3, s4
+; GFX7-NEXT:    s_and_b32 s4, s8, s10
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX7-NEXT:    s_or_b32 s3, s3, s4
+; GFX7-NEXT:    s_and_b32 s4, s9, s10
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX7-NEXT:    s_or_b32 s3, s3, s4
+; GFX7-NEXT:    v_mov_b32_e32 v0, s2
+; GFX7-NEXT:    v_mov_b32_e32 v1, s3
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i8>, <8 x i8> addrspace(4)* %ptr
+  %insert = insertelement <8 x i8> %vec, i8 %val, i32 %idx
+  store <8 x i8> %insert, <8 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v8i8_s_s(<8 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v8i8_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_mov_b32_e32 v2, 8
+; GFX9-NEXT:    s_movk_i32 s4, 0xff
+; GFX9-NEXT:    s_lshr_b32 s1, s3, 2
+; GFX9-NEXT:    s_and_b32 s3, s3, 3
+; GFX9-NEXT:    s_and_b32 s2, s2, s4
+; GFX9-NEXT:    s_lshl_b32 s3, s3, 3
+; GFX9-NEXT:    s_lshl_b32 s2, s2, s3
+; GFX9-NEXT:    s_lshl_b32 s3, s4, s3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX9-NEXT:    s_not_b32 s3, s3
+; GFX9-NEXT:    v_mov_b32_e32 v3, s2
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, s0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v6, v0, s4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v7, v0, s4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s4, v4
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, s4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v9, v1, s4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s4, v5
+; GFX9-NEXT:    v_or3_b32 v0, v0, v6, v7
+; GFX9-NEXT:    v_or3_b32 v1, v1, v8, v9
+; GFX9-NEXT:    v_cndmask_b32_e32 v4, v0, v1, vcc
+; GFX9-NEXT:    v_and_or_b32 v3, v4, s3, v3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v5, v0, s4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v6, v0, s4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s4, v3
+; GFX9-NEXT:    v_and_b32_sdwa v7, v1, s4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, s4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s4, v2
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_or3_b32 v0, v0, v5, v6
+; GFX9-NEXT:    v_or3_b32 v1, v1, v7, v8
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v8i8_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    v_mov_b32_e32 v3, 8
+; GFX8-NEXT:    v_mov_b32_e32 v4, s0
+; GFX8-NEXT:    s_lshr_b32 s1, s3, 2
+; GFX8-NEXT:    s_and_b32 s3, s3, 3
+; GFX8-NEXT:    s_lshl_b32 s3, s3, 3
+; GFX8-NEXT:    s_and_b32 s2, s2, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    s_lshl_b32 s2, s2, s3
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v3, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v7, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v8, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v9, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v10, v1, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v9
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v8
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v10
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
+; GFX8-NEXT:    v_and_b32_e32 v2, s0, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, s2, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v3, v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v6, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v7, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v8, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v4, v1, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v8
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v8i8_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s6, 0xff
+; GFX7-NEXT:    s_and_b32 s1, s3, 3
+; GFX7-NEXT:    s_lshr_b32 s0, s3, 2
+; GFX7-NEXT:    s_and_b32 s2, s2, s6
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX7-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX7-NEXT:    s_lshl_b32 s1, s6, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s0, 1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v2
+; GFX7-NEXT:    v_and_b32_e32 v5, s6, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v3
+; GFX7-NEXT:    v_and_b32_e32 v6, s6, v6
+; GFX7-NEXT:    v_and_b32_e32 v0, s6, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX7-NEXT:    v_and_b32_e32 v7, s6, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 24, v7
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v6
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v2, s1, v2
+; GFX7-NEXT:    v_or_b32_e32 v2, s2, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s0, 0
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v2
+; GFX7-NEXT:    v_and_b32_e32 v5, s6, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v3
+; GFX7-NEXT:    v_and_b32_e32 v6, s6, v6
+; GFX7-NEXT:    v_and_b32_e32 v0, s6, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX7-NEXT:    v_and_b32_e32 v7, s6, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 24, v7
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v6
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i8>, <8 x i8> addrspace(1 )* %ptr
+  %insert = insertelement <8 x i8> %vec, i8 %val, i32 %idx
+  store <8 x i8> %insert, <8 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v8i8_v_s(<8 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v8i8_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s10, 0xff
+; GFX9-NEXT:    v_and_b32_e32 v0, s10, v0
+; GFX9-NEXT:    s_mov_b32 s5, 8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s10
+; GFX9-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s3, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s6, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s7, s10
+; GFX9-NEXT:    s_lshr_b32 s8, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s8, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s9, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_lshr_b32 s2, s4, 2
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX9-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX9-NEXT:    s_and_b32 s4, s4, 3
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 3
+; GFX9-NEXT:    s_lshl_b32 s6, s10, s4
+; GFX9-NEXT:    s_not_b32 s6, s6
+; GFX9-NEXT:    s_and_b32 s3, s3, s6
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    v_lshl_or_b32 v2, v0, s4, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s5, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX9-NEXT:    v_and_or_b32 v2, v0, s10, v2
+; GFX9-NEXT:    v_and_b32_sdwa v4, v0, s10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s10 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v0, v2, v4, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s5, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v2, v1, s10, v2
+; GFX9-NEXT:    v_and_b32_sdwa v3, v1, s10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v1, s10 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v1, v2, v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v8i8_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s9, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX8-NEXT:    s_and_b32 s2, s2, s9
+; GFX8-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s5, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s3, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s5, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_lshr_b32 s6, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s6, s9
+; GFX8-NEXT:    s_lshr_b32 s7, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s8, s1, 24
+; GFX8-NEXT:    s_and_b32 s1, s1, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s7, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s8, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_lshr_b32 s2, s4, 2
+; GFX8-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX8-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX8-NEXT:    s_and_b32 s4, s4, 3
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 3
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    s_lshl_b32 s4, s9, s4
+; GFX8-NEXT:    s_not_b32 s4, s4
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    s_and_b32 s3, s3, s4
+; GFX8-NEXT:    v_or_b32_e32 v2, s3, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v4, s9
+; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v5, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v5
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX8-NEXT:    v_and_b32_sdwa v1, v1, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v8i8_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s9, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v0, s9, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX7-NEXT:    s_and_b32 s2, s2, s9
+; GFX7-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s5, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s3, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s5, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX7-NEXT:    s_lshr_b32 s6, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s6, s9
+; GFX7-NEXT:    s_lshr_b32 s7, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s8, s1, 24
+; GFX7-NEXT:    s_and_b32 s1, s1, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_and_b32 s2, s7, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_and_b32 s2, s8, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_lshr_b32 s2, s4, 2
+; GFX7-NEXT:    s_cmp_eq_u32 s2, 1
+; GFX7-NEXT:    s_cselect_b32 s3, s1, s0
+; GFX7-NEXT:    s_and_b32 s4, s4, 3
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
+; GFX7-NEXT:    s_lshl_b32 s4, s9, s4
+; GFX7-NEXT:    s_not_b32 s4, s4
+; GFX7-NEXT:    s_and_b32 s3, s3, s4
+; GFX7-NEXT:    v_or_b32_e32 v2, s3, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s2, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s9, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s9, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i8>, <8 x i8> addrspace(4)* %ptr
+  %insert = insertelement <8 x i8> %vec, i8 %val, i32 %idx
+  store <8 x i8> %insert, <8 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v8i8_s_v(<8 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v8i8_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s10, 0xff
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 2, v0
+; GFX9-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s10
+; GFX9-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s3, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s6, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s7, s10
+; GFX9-NEXT:    s_lshr_b32 s8, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s8, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s9, s10
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, s0
+; GFX9-NEXT:    v_mov_b32_e32 v3, s1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX9-NEXT:    s_and_b32 s2, s4, s10
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_lshlrev_b32_e64 v3, v0, s2
+; GFX9-NEXT:    v_lshlrev_b32_e64 v0, v0, s10
+; GFX9-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX9-NEXT:    v_and_or_b32 v3, v1, v0, v3
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX9-NEXT:    s_mov_b32 s5, 8
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s5, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX9-NEXT:    v_and_or_b32 v2, v0, s10, v2
+; GFX9-NEXT:    v_and_b32_sdwa v4, v0, s10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s10 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v0, v2, v4, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s5, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v2, v1, s10, v2
+; GFX9-NEXT:    v_and_b32_sdwa v3, v1, s10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v1, s10 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v1, v2, v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v8i8_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s9, 0xff
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 2, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX8-NEXT:    s_and_b32 s2, s2, s9
+; GFX8-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s5, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s3, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s5, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_lshr_b32 s6, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s6, s9
+; GFX8-NEXT:    s_lshr_b32 s7, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s8, s1, 24
+; GFX8-NEXT:    s_and_b32 s1, s1, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s7, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s8, s9
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s0
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX8-NEXT:    s_and_b32 s2, s4, s9
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX8-NEXT:    v_lshlrev_b32_e64 v3, v0, s2
+; GFX8-NEXT:    v_lshlrev_b32_e64 v0, v0, s9
+; GFX8-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX8-NEXT:    v_or_b32_e32 v3, v0, v3
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v4, s9
+; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v5, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v5
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX8-NEXT:    v_and_b32_sdwa v1, v1, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v8i8_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s9, 0xff
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 2, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX7-NEXT:    s_and_b32 s2, s2, s9
+; GFX7-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s5, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s3, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s5, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX7-NEXT:    s_lshr_b32 s6, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s6, s9
+; GFX7-NEXT:    s_lshr_b32 s7, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s8, s1, 24
+; GFX7-NEXT:    s_and_b32 s1, s1, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_and_b32 s2, s7, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_and_b32 s2, s8, s9
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    v_mov_b32_e32 v1, s0
+; GFX7-NEXT:    v_mov_b32_e32 v3, s1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX7-NEXT:    s_and_b32 s2, s4, s9
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX7-NEXT:    v_lshl_b32_e32 v3, s2, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v0, s9, v0
+; GFX7-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_or_b32_e32 v3, v0, v3
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v2
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s9, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s9, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s9, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i8>, <8 x i8> addrspace(4)* %ptr
+  %insert = insertelement <8 x i8> %vec, i8 %val, i32 %idx
+  store <8 x i8> %insert, <8 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v8i8_v_v(<8 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v8i8_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s9, 0xff
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 2, v1
+; GFX9-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX9-NEXT:    s_and_b32 s2, s2, s9
+; GFX9-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s5, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s9
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s3, s9
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s5, s9
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_lshr_b32 s6, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s2
+; GFX9-NEXT:    s_and_b32 s2, s6, s9
+; GFX9-NEXT:    s_lshr_b32 s7, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s8, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s9
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s7, s9
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    s_and_b32 s2, s8, s9
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX9-NEXT:    s_or_b32 s1, s1, s2
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s9
+; GFX9-NEXT:    v_mov_b32_e32 v3, s0
+; GFX9-NEXT:    v_mov_b32_e32 v4, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_and_or_b32 v3, v3, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX9-NEXT:    s_mov_b32 s4, 8
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX9-NEXT:    v_and_or_b32 v2, v0, s9, v2
+; GFX9-NEXT:    v_and_b32_sdwa v4, v0, s9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s9 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v0, v2, v4, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s4, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v2, v1, s9, v2
+; GFX9-NEXT:    v_and_b32_sdwa v3, v1, s9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v1, s9 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v1, v2, v3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v8i8_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s8, 0xff
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 2, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX8-NEXT:    s_and_b32 s2, s2, s8
+; GFX8-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s4, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s8
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s3, s8
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s4, s8
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_lshr_b32 s5, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s2
+; GFX8-NEXT:    s_and_b32 s2, s5, s8
+; GFX8-NEXT:    s_lshr_b32 s6, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s7, s1, 24
+; GFX8-NEXT:    s_and_b32 s1, s1, s8
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s6, s8
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    s_and_b32 s2, s7, s8
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX8-NEXT:    s_or_b32 s1, s1, s2
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s8
+; GFX8-NEXT:    v_mov_b32_e32 v3, s0
+; GFX8-NEXT:    v_mov_b32_e32 v4, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, v3, v1
+; GFX8-NEXT:    v_or_b32_e32 v3, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v4, s8
+; GFX8-NEXT:    v_or_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v5, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v5
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v2, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_sdwa v2, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX8-NEXT:    v_and_b32_sdwa v1, v1, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v8i8_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s8, 0xff
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 2, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s2, s0, 8
+; GFX7-NEXT:    s_and_b32 s2, s2, s8
+; GFX7-NEXT:    s_lshr_b32 s3, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s4, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s8
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s3, s8
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s4, s8
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX7-NEXT:    s_lshr_b32 s5, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s2
+; GFX7-NEXT:    s_and_b32 s2, s5, s8
+; GFX7-NEXT:    s_lshr_b32 s6, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s7, s1, 24
+; GFX7-NEXT:    s_and_b32 s1, s1, s8
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 8
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_and_b32 s2, s6, s8
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    s_and_b32 s2, s7, s8
+; GFX7-NEXT:    s_lshl_b32 s2, s2, 24
+; GFX7-NEXT:    s_or_b32 s1, s1, s2
+; GFX7-NEXT:    v_and_b32_e32 v0, s8, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s8, v1
+; GFX7-NEXT:    v_mov_b32_e32 v3, s0
+; GFX7-NEXT:    v_mov_b32_e32 v4, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v2
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v3, v1
+; GFX7-NEXT:    v_or_b32_e32 v3, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s8, v2
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s8, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s8, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s8, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s8, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s8, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s8, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s8, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i8>, <8 x i8> addrspace(4)* %ptr
+  %insert = insertelement <8 x i8> %vec, i8 %val, i32 %idx
+  store <8 x i8> %insert, <8 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v8i8_s_v(<8 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v8i8_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_mov_b32_e32 v3, 8
+; GFX9-NEXT:    s_movk_i32 s3, 0xff
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 2, v2
+; GFX9-NEXT:    v_and_b32_e32 v2, 3, v2
+; GFX9-NEXT:    s_and_b32 s1, s2, s3
+; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v2
+; GFX9-NEXT:    v_lshlrev_b32_e64 v5, v2, s1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v2, s3
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX9-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v6, s0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v7, v3, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v8, v0, s3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v9, v0, s3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s3, v6
+; GFX9-NEXT:    v_and_b32_sdwa v10, v1, s3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v11, v1, s3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s3, v7
+; GFX9-NEXT:    v_or3_b32 v0, v0, v8, v9
+; GFX9-NEXT:    v_or3_b32 v1, v1, v10, v11
+; GFX9-NEXT:    v_cndmask_b32_e32 v6, v0, v1, vcc
+; GFX9-NEXT:    v_and_or_b32 v2, v6, v2, v5
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v4
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v5, v0, s3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v6, v0, s3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s3, v2
+; GFX9-NEXT:    v_and_b32_sdwa v7, v1, s3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, s3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s3, v3
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_or3_b32 v0, v0, v5, v6
+; GFX9-NEXT:    v_or3_b32 v1, v1, v7, v8
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v8i8_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v3, 8
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_mov_b32_e32 v5, s0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 2, v2
+; GFX8-NEXT:    v_and_b32_e32 v2, 3, v2
+; GFX8-NEXT:    s_and_b32 s1, s2, s0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 3, v2
+; GFX8-NEXT:    v_lshlrev_b32_e64 v7, v2, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v2, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v6
+; GFX8-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v6
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v9, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v3, v3, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v8, v4, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v10, v0, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v11, v0, v5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v12, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v13, v1, v5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v10
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v12
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v11
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v13
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
+; GFX8-NEXT:    v_and_b32_e32 v2, v3, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v7
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v3, v4, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v6, v0, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v7, v0, v5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v8, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v5, v1, v5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v8
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v5
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v8i8_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s3, 0xff
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 2, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, 3, v2
+; GFX7-NEXT:    s_and_b32 s0, s2, s3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 3, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v4, s0, v2
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s3, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v3
+; GFX7-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v3
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v5, s3, v5
+; GFX7-NEXT:    v_and_b32_e32 v8, s3, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v6, s3, v6
+; GFX7-NEXT:    v_and_b32_e32 v9, s3, v9
+; GFX7-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
+; GFX7-NEXT:    v_and_b32_e32 v1, s3, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 8, v8
+; GFX7-NEXT:    v_and_b32_e32 v7, s3, v7
+; GFX7-NEXT:    v_and_b32_e32 v10, s3, v10
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v9, 16, v9
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 24, v7
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v10, 24, v10
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v9
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v10
+; GFX7-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v2, v5, v2
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s3, v2
+; GFX7-NEXT:    v_and_b32_e32 v5, s3, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s3, v3
+; GFX7-NEXT:    v_and_b32_e32 v6, s3, v6
+; GFX7-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, s3, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
+; GFX7-NEXT:    v_and_b32_e32 v4, s3, v4
+; GFX7-NEXT:    v_and_b32_e32 v7, s3, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 24, v7
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v6
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i8>, <8 x i8> addrspace(1)* %ptr
+  %insert = insertelement <8 x i8> %vec, i8 %val, i32 %idx
+  store <8 x i8> %insert, <8 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v8i8_v_s(<8 x i8> addrspace(1)* %ptr, i8 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v8i8_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_mov_b32_e32 v3, 8
+; GFX9-NEXT:    s_movk_i32 s3, 0xff
+; GFX9-NEXT:    s_lshr_b32 s1, s2, 2
+; GFX9-NEXT:    s_and_b32 s2, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s2, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    s_lshl_b32 s2, s3, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX9-NEXT:    s_not_b32 s2, s2
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, s0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v6, v0, s3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v7, v0, s3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s3, v4
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, s3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v9, v1, s3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s3, v5
+; GFX9-NEXT:    v_or3_b32 v0, v0, v6, v7
+; GFX9-NEXT:    v_or3_b32 v1, v1, v8, v9
+; GFX9-NEXT:    v_cndmask_b32_e32 v4, v0, v1, vcc
+; GFX9-NEXT:    v_and_or_b32 v2, v4, s2, v2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v5, v0, s3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v6, v0, s3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s3, v2
+; GFX9-NEXT:    v_and_b32_sdwa v7, v1, s3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, s3 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s3, v3
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_or3_b32 v0, v0, v5, v6
+; GFX9-NEXT:    v_or3_b32 v1, v1, v7, v8
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v8i8_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    s_lshr_b32 s1, s2, 2
+; GFX8-NEXT:    s_and_b32 s2, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s2, s2, 3
+; GFX8-NEXT:    v_mov_b32_e32 v6, s2
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v6, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v3, 8
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_mov_b32_e32 v5, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s1, 1
+; GFX8-NEXT:    s_not_b32 s0, s0
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v3, v3, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v6, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v8, v0, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v9, v0, v5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v10, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v11, v1, v5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v8
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v10
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v9
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v11
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
+; GFX8-NEXT:    v_and_b32_e32 v3, s0, v3
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s1, 0
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v3, v4, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v6, v0, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v7, v0, v5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v8, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v5, v1, v5 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v8
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v5
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v8i8_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s3, 0xff
+; GFX7-NEXT:    s_and_b32 s1, s2, 3
+; GFX7-NEXT:    s_lshr_b32 s0, s2, 2
+; GFX7-NEXT:    v_and_b32_e32 v2, s3, v2
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, s1, v2
+; GFX7-NEXT:    s_lshl_b32 s1, s3, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s0, 1
+; GFX7-NEXT:    s_not_b32 s1, s1
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s3, v3
+; GFX7-NEXT:    v_and_b32_e32 v6, s3, v6
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 24, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v4, s3, v4
+; GFX7-NEXT:    v_and_b32_e32 v7, s3, v7
+; GFX7-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
+; GFX7-NEXT:    v_and_b32_e32 v1, s3, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 8, v6
+; GFX7-NEXT:    v_and_b32_e32 v5, s3, v5
+; GFX7-NEXT:    v_and_b32_e32 v8, s3, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 24, v5
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 24, v8
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v8
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v3, s1, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s0, 0
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, s3, v2
+; GFX7-NEXT:    v_and_b32_e32 v5, s3, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v3, s3, v3
+; GFX7-NEXT:    v_and_b32_e32 v6, s3, v6
+; GFX7-NEXT:    v_and_b32_e32 v0, s3, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, s3, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
+; GFX7-NEXT:    v_and_b32_e32 v4, s3, v4
+; GFX7-NEXT:    v_and_b32_e32 v7, s3, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 24, v7
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v6
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i8>, <8 x i8> addrspace(1)* %ptr
+  %insert = insertelement <8 x i8> %vec, i8 %val, i32 %idx
+  store <8 x i8> %insert, <8 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v8i8_v_v(<8 x i8> addrspace(1)* %ptr, i8 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v8i8_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_mov_b32_e32 v5, 8
+; GFX9-NEXT:    s_movk_i32 s1, 0xff
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 2, v3
+; GFX9-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0xff
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v3, v3, v4
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v6
+; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v7, s0, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v8, v5, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v9, v0, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v10, v0, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s1, v7
+; GFX9-NEXT:    v_and_b32_sdwa v11, v1, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v12, v1, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s1, v8
+; GFX9-NEXT:    v_or3_b32 v0, v0, v9, v10
+; GFX9-NEXT:    v_or3_b32 v1, v1, v11, v12
+; GFX9-NEXT:    v_cndmask_b32_e32 v7, v0, v1, vcc
+; GFX9-NEXT:    v_and_or_b32 v2, v7, v3, v2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v6
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v5, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, v5, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v6, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v7, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, v4, v2
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v9, v1, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, v4, v3
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_or3_b32 v0, v0, v6, v7
+; GFX9-NEXT:    v_or3_b32 v1, v1, v8, v9
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-NEXT:    global_store_dwordx2 v[2:3], v[0:1], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v8i8_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v5, 8
+; GFX8-NEXT:    v_mov_b32_e32 v6, 8
+; GFX8-NEXT:    v_mov_b32_e32 v7, s0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 2, v3
+; GFX8-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0xff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v3, v3, v4
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX8-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v8
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v9, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v10, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v5, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v9, v6, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v11, v0, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v12, v0, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v13, v1, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v7, v1, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v11
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v13
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v12
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX8-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX8-NEXT:    v_and_b32_e32 v3, v5, v3
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v6, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v3, v6, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v5, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v7, v0, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v8, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v4, v1, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v8
+; GFX8-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX8-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v8i8_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s6, 0
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s0, 0xff
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 2, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX7-NEXT:    v_mov_b32_e32 v4, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v5
+; GFX7-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v6, s0, v6
+; GFX7-NEXT:    v_and_b32_e32 v9, s0, v9
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 24, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v7, s0, v7
+; GFX7-NEXT:    v_and_b32_e32 v10, s0, v10
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 8, v6
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v9, 8, v9
+; GFX7-NEXT:    v_and_b32_e32 v8, s0, v8
+; GFX7-NEXT:    v_and_b32_e32 v11, s0, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v10, 16, v10
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 24, v8
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v11, 24, v11
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v10
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v8
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v11
+; GFX7-NEXT:    v_cndmask_b32_e32 v6, v0, v1, vcc
+; GFX7-NEXT:    v_and_b32_e32 v3, v6, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v5
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v6, v6, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v7, v7, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 8, v6
+; GFX7-NEXT:    v_and_b32_e32 v5, v5, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, v8, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 24, v5
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <8 x i8>, <8 x i8> addrspace(1)* %ptr
+  %insert = insertelement <8 x i8> %vec, i8 %val, i32 %idx
+  store <8 x i8> %insert, <8 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v16i8_s_s(<16 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v16i8_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s18, 0xff
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s6, s0, 8
+; GFX9-NEXT:    s_and_b32 s6, s6, s18
+; GFX9-NEXT:    s_lshr_b32 s7, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s8, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s6
+; GFX9-NEXT:    s_and_b32 s6, s7, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s6
+; GFX9-NEXT:    s_and_b32 s6, s8, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s6
+; GFX9-NEXT:    s_and_b32 s6, s9, s18
+; GFX9-NEXT:    s_lshr_b32 s10, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s11, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s6
+; GFX9-NEXT:    s_and_b32 s6, s10, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s6
+; GFX9-NEXT:    s_and_b32 s6, s11, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX9-NEXT:    s_lshr_b32 s12, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s6
+; GFX9-NEXT:    s_and_b32 s6, s12, s18
+; GFX9-NEXT:    s_lshr_b32 s13, s2, 16
+; GFX9-NEXT:    s_lshr_b32 s14, s2, 24
+; GFX9-NEXT:    s_and_b32 s2, s2, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_or_b32 s2, s2, s6
+; GFX9-NEXT:    s_and_b32 s6, s13, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX9-NEXT:    s_or_b32 s2, s2, s6
+; GFX9-NEXT:    s_and_b32 s6, s14, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX9-NEXT:    s_lshr_b32 s15, s3, 8
+; GFX9-NEXT:    s_or_b32 s2, s2, s6
+; GFX9-NEXT:    s_and_b32 s6, s15, s18
+; GFX9-NEXT:    s_lshr_b32 s16, s3, 16
+; GFX9-NEXT:    s_lshr_b32 s17, s3, 24
+; GFX9-NEXT:    s_and_b32 s3, s3, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_or_b32 s3, s3, s6
+; GFX9-NEXT:    s_and_b32 s6, s16, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX9-NEXT:    s_or_b32 s3, s3, s6
+; GFX9-NEXT:    s_and_b32 s6, s17, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX9-NEXT:    s_or_b32 s3, s3, s6
+; GFX9-NEXT:    s_lshr_b32 s6, s5, 2
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX9-NEXT:    s_cselect_b32 s7, s1, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX9-NEXT:    s_cselect_b32 s7, s2, s7
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX9-NEXT:    s_cselect_b32 s7, s3, s7
+; GFX9-NEXT:    s_and_b32 s5, s5, 3
+; GFX9-NEXT:    s_lshl_b32 s5, s5, 3
+; GFX9-NEXT:    s_and_b32 s4, s4, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX9-NEXT:    s_lshl_b32 s5, s18, s5
+; GFX9-NEXT:    s_not_b32 s5, s5
+; GFX9-NEXT:    s_and_b32 s5, s7, s5
+; GFX9-NEXT:    s_or_b32 s4, s5, s4
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 0
+; GFX9-NEXT:    s_cselect_b32 s0, s4, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX9-NEXT:    s_cselect_b32 s1, s4, s1
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX9-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX9-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX9-NEXT:    s_lshr_b32 s4, s0, 8
+; GFX9-NEXT:    s_and_b32 s4, s4, s18
+; GFX9-NEXT:    s_lshr_b32 s5, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s4
+; GFX9-NEXT:    s_and_b32 s4, s5, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s4
+; GFX9-NEXT:    s_and_b32 s4, s6, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX9-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s4
+; GFX9-NEXT:    s_and_b32 s4, s7, s18
+; GFX9-NEXT:    s_lshr_b32 s8, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s4
+; GFX9-NEXT:    s_and_b32 s4, s8, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s4
+; GFX9-NEXT:    s_and_b32 s4, s9, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX9-NEXT:    s_lshr_b32 s10, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s4
+; GFX9-NEXT:    s_and_b32 s4, s10, s18
+; GFX9-NEXT:    s_lshr_b32 s11, s2, 16
+; GFX9-NEXT:    s_lshr_b32 s12, s2, 24
+; GFX9-NEXT:    s_and_b32 s2, s2, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX9-NEXT:    s_or_b32 s2, s2, s4
+; GFX9-NEXT:    s_and_b32 s4, s11, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-NEXT:    s_or_b32 s2, s2, s4
+; GFX9-NEXT:    s_and_b32 s4, s12, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX9-NEXT:    s_lshr_b32 s13, s3, 8
+; GFX9-NEXT:    s_or_b32 s2, s2, s4
+; GFX9-NEXT:    s_and_b32 s4, s13, s18
+; GFX9-NEXT:    s_lshr_b32 s14, s3, 16
+; GFX9-NEXT:    s_lshr_b32 s15, s3, 24
+; GFX9-NEXT:    s_and_b32 s3, s3, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX9-NEXT:    s_or_b32 s3, s3, s4
+; GFX9-NEXT:    s_and_b32 s4, s14, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-NEXT:    s_or_b32 s3, s3, s4
+; GFX9-NEXT:    s_and_b32 s4, s15, s18
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX9-NEXT:    s_or_b32 s3, s3, s4
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v16i8_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s18, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s6, s0, 8
+; GFX8-NEXT:    s_and_b32 s6, s6, s18
+; GFX8-NEXT:    s_lshr_b32 s7, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s8, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s6
+; GFX8-NEXT:    s_and_b32 s6, s7, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s6
+; GFX8-NEXT:    s_and_b32 s6, s8, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX8-NEXT:    s_lshr_b32 s9, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s6
+; GFX8-NEXT:    s_and_b32 s6, s9, s18
+; GFX8-NEXT:    s_lshr_b32 s10, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s11, s1, 24
+; GFX8-NEXT:    s_and_b32 s1, s1, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s6
+; GFX8-NEXT:    s_and_b32 s6, s10, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX8-NEXT:    s_or_b32 s1, s1, s6
+; GFX8-NEXT:    s_and_b32 s6, s11, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX8-NEXT:    s_lshr_b32 s12, s2, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s6
+; GFX8-NEXT:    s_and_b32 s6, s12, s18
+; GFX8-NEXT:    s_lshr_b32 s13, s2, 16
+; GFX8-NEXT:    s_lshr_b32 s14, s2, 24
+; GFX8-NEXT:    s_and_b32 s2, s2, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX8-NEXT:    s_or_b32 s2, s2, s6
+; GFX8-NEXT:    s_and_b32 s6, s13, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX8-NEXT:    s_or_b32 s2, s2, s6
+; GFX8-NEXT:    s_and_b32 s6, s14, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX8-NEXT:    s_lshr_b32 s15, s3, 8
+; GFX8-NEXT:    s_or_b32 s2, s2, s6
+; GFX8-NEXT:    s_and_b32 s6, s15, s18
+; GFX8-NEXT:    s_lshr_b32 s16, s3, 16
+; GFX8-NEXT:    s_lshr_b32 s17, s3, 24
+; GFX8-NEXT:    s_and_b32 s3, s3, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX8-NEXT:    s_or_b32 s3, s3, s6
+; GFX8-NEXT:    s_and_b32 s6, s16, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX8-NEXT:    s_or_b32 s3, s3, s6
+; GFX8-NEXT:    s_and_b32 s6, s17, s18
+; GFX8-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX8-NEXT:    s_or_b32 s3, s3, s6
+; GFX8-NEXT:    s_lshr_b32 s6, s5, 2
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX8-NEXT:    s_cselect_b32 s7, s1, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX8-NEXT:    s_cselect_b32 s7, s2, s7
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX8-NEXT:    s_cselect_b32 s7, s3, s7
+; GFX8-NEXT:    s_and_b32 s5, s5, 3
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 3
+; GFX8-NEXT:    s_and_b32 s4, s4, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX8-NEXT:    s_lshl_b32 s5, s18, s5
+; GFX8-NEXT:    s_not_b32 s5, s5
+; GFX8-NEXT:    s_and_b32 s5, s7, s5
+; GFX8-NEXT:    s_or_b32 s4, s5, s4
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 0
+; GFX8-NEXT:    s_cselect_b32 s0, s4, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX8-NEXT:    s_cselect_b32 s1, s4, s1
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX8-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX8-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX8-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX8-NEXT:    s_lshr_b32 s4, s0, 8
+; GFX8-NEXT:    s_and_b32 s4, s4, s18
+; GFX8-NEXT:    s_lshr_b32 s5, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s4
+; GFX8-NEXT:    s_and_b32 s4, s5, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s4
+; GFX8-NEXT:    s_and_b32 s4, s6, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX8-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s4
+; GFX8-NEXT:    s_and_b32 s4, s7, s18
+; GFX8-NEXT:    s_lshr_b32 s8, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s9, s1, 24
+; GFX8-NEXT:    s_and_b32 s1, s1, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s4
+; GFX8-NEXT:    s_and_b32 s4, s8, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX8-NEXT:    s_or_b32 s1, s1, s4
+; GFX8-NEXT:    s_and_b32 s4, s9, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX8-NEXT:    s_lshr_b32 s10, s2, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s4
+; GFX8-NEXT:    s_and_b32 s4, s10, s18
+; GFX8-NEXT:    s_lshr_b32 s11, s2, 16
+; GFX8-NEXT:    s_lshr_b32 s12, s2, 24
+; GFX8-NEXT:    s_and_b32 s2, s2, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX8-NEXT:    s_or_b32 s2, s2, s4
+; GFX8-NEXT:    s_and_b32 s4, s11, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX8-NEXT:    s_or_b32 s2, s2, s4
+; GFX8-NEXT:    s_and_b32 s4, s12, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX8-NEXT:    s_lshr_b32 s13, s3, 8
+; GFX8-NEXT:    s_or_b32 s2, s2, s4
+; GFX8-NEXT:    s_and_b32 s4, s13, s18
+; GFX8-NEXT:    s_lshr_b32 s14, s3, 16
+; GFX8-NEXT:    s_lshr_b32 s15, s3, 24
+; GFX8-NEXT:    s_and_b32 s3, s3, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX8-NEXT:    s_or_b32 s3, s3, s4
+; GFX8-NEXT:    s_and_b32 s4, s14, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX8-NEXT:    s_or_b32 s3, s3, s4
+; GFX8-NEXT:    s_and_b32 s4, s15, s18
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX8-NEXT:    s_or_b32 s3, s3, s4
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v16i8_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s18, 0xff
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s6, s0, 8
+; GFX7-NEXT:    s_and_b32 s6, s6, s18
+; GFX7-NEXT:    s_lshr_b32 s7, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s8, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s6
+; GFX7-NEXT:    s_and_b32 s6, s7, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s6
+; GFX7-NEXT:    s_and_b32 s6, s8, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX7-NEXT:    s_lshr_b32 s9, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s6
+; GFX7-NEXT:    s_and_b32 s6, s9, s18
+; GFX7-NEXT:    s_lshr_b32 s10, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s11, s1, 24
+; GFX7-NEXT:    s_and_b32 s1, s1, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX7-NEXT:    s_or_b32 s1, s1, s6
+; GFX7-NEXT:    s_and_b32 s6, s10, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX7-NEXT:    s_or_b32 s1, s1, s6
+; GFX7-NEXT:    s_and_b32 s6, s11, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX7-NEXT:    s_lshr_b32 s12, s2, 8
+; GFX7-NEXT:    s_or_b32 s1, s1, s6
+; GFX7-NEXT:    s_and_b32 s6, s12, s18
+; GFX7-NEXT:    s_lshr_b32 s13, s2, 16
+; GFX7-NEXT:    s_lshr_b32 s14, s2, 24
+; GFX7-NEXT:    s_and_b32 s2, s2, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX7-NEXT:    s_or_b32 s2, s2, s6
+; GFX7-NEXT:    s_and_b32 s6, s13, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX7-NEXT:    s_or_b32 s2, s2, s6
+; GFX7-NEXT:    s_and_b32 s6, s14, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX7-NEXT:    s_lshr_b32 s15, s3, 8
+; GFX7-NEXT:    s_or_b32 s2, s2, s6
+; GFX7-NEXT:    s_and_b32 s6, s15, s18
+; GFX7-NEXT:    s_lshr_b32 s16, s3, 16
+; GFX7-NEXT:    s_lshr_b32 s17, s3, 24
+; GFX7-NEXT:    s_and_b32 s3, s3, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX7-NEXT:    s_or_b32 s3, s3, s6
+; GFX7-NEXT:    s_and_b32 s6, s16, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX7-NEXT:    s_or_b32 s3, s3, s6
+; GFX7-NEXT:    s_and_b32 s6, s17, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX7-NEXT:    s_or_b32 s3, s3, s6
+; GFX7-NEXT:    s_lshr_b32 s6, s5, 2
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX7-NEXT:    s_cselect_b32 s7, s1, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX7-NEXT:    s_cselect_b32 s7, s2, s7
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX7-NEXT:    s_cselect_b32 s7, s3, s7
+; GFX7-NEXT:    s_and_b32 s5, s5, 3
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 3
+; GFX7-NEXT:    s_and_b32 s4, s4, s18
+; GFX7-NEXT:    s_lshl_b32 s4, s4, s5
+; GFX7-NEXT:    s_lshl_b32 s5, s18, s5
+; GFX7-NEXT:    s_not_b32 s5, s5
+; GFX7-NEXT:    s_and_b32 s5, s7, s5
+; GFX7-NEXT:    s_or_b32 s4, s5, s4
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 0
+; GFX7-NEXT:    s_cselect_b32 s5, s4, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX7-NEXT:    s_cselect_b32 s7, s4, s1
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX7-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX7-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX7-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX7-NEXT:    s_lshr_b32 s4, s5, 8
+; GFX7-NEXT:    s_and_b32 s4, s4, s18
+; GFX7-NEXT:    s_lshr_b32 s6, s5, 16
+; GFX7-NEXT:    s_lshr_b32 s8, s5, 24
+; GFX7-NEXT:    s_and_b32 s5, s5, s18
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX7-NEXT:    s_or_b32 s4, s5, s4
+; GFX7-NEXT:    s_and_b32 s5, s6, s18
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX7-NEXT:    s_lshr_b32 s9, s7, 8
+; GFX7-NEXT:    s_or_b32 s4, s4, s5
+; GFX7-NEXT:    s_and_b32 s5, s8, s18
+; GFX7-NEXT:    s_and_b32 s6, s9, s18
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX7-NEXT:    s_or_b32 s4, s4, s5
+; GFX7-NEXT:    s_lshr_b32 s10, s7, 16
+; GFX7-NEXT:    s_and_b32 s5, s7, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX7-NEXT:    s_or_b32 s5, s5, s6
+; GFX7-NEXT:    s_and_b32 s6, s10, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX7-NEXT:    s_lshr_b32 s11, s7, 24
+; GFX7-NEXT:    s_or_b32 s5, s5, s6
+; GFX7-NEXT:    s_and_b32 s6, s11, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX7-NEXT:    s_lshr_b32 s12, s2, 8
+; GFX7-NEXT:    s_or_b32 s5, s5, s6
+; GFX7-NEXT:    s_and_b32 s6, s12, s18
+; GFX7-NEXT:    s_lshr_b32 s13, s2, 16
+; GFX7-NEXT:    s_lshr_b32 s14, s2, 24
+; GFX7-NEXT:    s_and_b32 s2, s2, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX7-NEXT:    s_or_b32 s2, s2, s6
+; GFX7-NEXT:    s_and_b32 s6, s13, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX7-NEXT:    s_or_b32 s2, s2, s6
+; GFX7-NEXT:    s_and_b32 s6, s14, s18
+; GFX7-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX7-NEXT:    s_lshr_b32 s15, s3, 8
+; GFX7-NEXT:    s_or_b32 s6, s2, s6
+; GFX7-NEXT:    s_lshr_b32 s16, s3, 16
+; GFX7-NEXT:    s_lshr_b32 s17, s3, 24
+; GFX7-NEXT:    s_and_b32 s2, s3, s18
+; GFX7-NEXT:    s_and_b32 s3, s15, s18
+; GFX7-NEXT:    s_lshl_b32 s3, s3, 8
+; GFX7-NEXT:    s_or_b32 s2, s2, s3
+; GFX7-NEXT:    s_and_b32 s3, s16, s18
+; GFX7-NEXT:    s_lshl_b32 s3, s3, 16
+; GFX7-NEXT:    s_or_b32 s2, s2, s3
+; GFX7-NEXT:    s_and_b32 s3, s17, s18
+; GFX7-NEXT:    s_lshl_b32 s3, s3, 24
+; GFX7-NEXT:    s_or_b32 s7, s2, s3
+; GFX7-NEXT:    v_mov_b32_e32 v0, s4
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s5
+; GFX7-NEXT:    v_mov_b32_e32 v2, s6
+; GFX7-NEXT:    v_mov_b32_e32 v3, s7
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i8>, <16 x i8> addrspace(4)* %ptr
+  %insert = insertelement <16 x i8> %vec, i8 %val, i32 %idx
+  store <16 x i8> %insert, <16 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v16i8_s_s(<16 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v16i8_s_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_mov_b32_e32 v4, 8
+; GFX9-NEXT:    s_movk_i32 s6, 0xff
+; GFX9-NEXT:    s_and_b32 s1, s3, 3
+; GFX9-NEXT:    s_lshr_b32 s4, s3, 2
+; GFX9-NEXT:    s_and_b32 s2, s2, s6
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX9-NEXT:    s_lshl_b32 s2, s2, s1
+; GFX9-NEXT:    s_lshl_b32 s1, s6, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX9-NEXT:    s_not_b32 s5, s1
+; GFX9-NEXT:    v_mov_b32_e32 v5, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v2
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v6, s0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v7, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 8, v3
+; GFX9-NEXT:    v_and_b32_sdwa v10, v0, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v11, v0, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s6, v6
+; GFX9-NEXT:    v_and_b32_sdwa v12, v1, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v13, v1, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s6, v7
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v8, v4, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v14, v2, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v15, v2, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v2, v2, s6, v8
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v9, v4, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_or3_b32 v0, v0, v10, v11
+; GFX9-NEXT:    v_or3_b32 v1, v1, v12, v13
+; GFX9-NEXT:    v_and_b32_sdwa v16, v3, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v17, v3, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v3, v3, s6, v9
+; GFX9-NEXT:    v_or3_b32 v2, v2, v14, v15
+; GFX9-NEXT:    v_cndmask_b32_e32 v6, v0, v1, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX9-NEXT:    v_or3_b32 v3, v3, v16, v17
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v6, v2, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v6, v3, s[2:3]
+; GFX9-NEXT:    v_and_or_b32 v5, v6, s5, v5
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v6, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v7, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, v4, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v9, v0, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v10, v0, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v0, v0, s6, v5
+; GFX9-NEXT:    v_and_b32_sdwa v11, v1, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v12, v1, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s6, v6
+; GFX9-NEXT:    v_and_b32_sdwa v13, v2, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v14, v2, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v2, v2, s6, v7
+; GFX9-NEXT:    v_and_b32_sdwa v15, v3, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v16, v3, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v3, v3, s6, v4
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_or3_b32 v0, v0, v9, v10
+; GFX9-NEXT:    v_or3_b32 v1, v1, v11, v12
+; GFX9-NEXT:    v_or3_b32 v2, v2, v13, v14
+; GFX9-NEXT:    v_or3_b32 v3, v3, v15, v16
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v16i8_s_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_mov_b32_e32 v5, 8
+; GFX8-NEXT:    v_mov_b32_e32 v6, s0
+; GFX8-NEXT:    s_and_b32 s1, s3, 3
+; GFX8-NEXT:    s_lshr_b32 s4, s3, 2
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX8-NEXT:    s_and_b32 s2, s2, s0
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX8-NEXT:    s_lshl_b32 s5, s2, s1
+; GFX8-NEXT:    s_not_b32 s6, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 8, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v9, 8, v2
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v7, v5, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v10, 8, v3
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v8, v5, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v11, v0, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v12, v0, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v13, v1, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v14, v1, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v11
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v13
+; GFX8-NEXT:    v_and_b32_sdwa v15, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v16, v2, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v2, v2, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v9, v5, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v15
+; GFX8-NEXT:    v_and_b32_sdwa v17, v3, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v18, v3, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v12
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v14
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v17
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v16
+; GFX8-NEXT:    v_cndmask_b32_e32 v4, v0, v1, vcc
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v18
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v3, s[2:3]
+; GFX8-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX8-NEXT:    v_or_b32_e32 v4, s5, v4
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[2:3]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 8, v2
+; GFX8-NEXT:    v_lshrrev_b32_e32 v9, 8, v3
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v5, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v7, v5, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v8, v5, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v5, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v10, v0, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v11, v0, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v12, v1, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v13, v1, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v1, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v14, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v15, v2, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v2, v2, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v16, v3, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v6, v3, v6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v10
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v12
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v14
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v16
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v11
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v13
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v15
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v6
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v16i8_s_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s10, 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s6, 0xff
+; GFX7-NEXT:    s_and_b32 s0, s3, 3
+; GFX7-NEXT:    s_lshr_b32 s4, s3, 2
+; GFX7-NEXT:    s_and_b32 s1, s2, s6
+; GFX7-NEXT:    s_lshl_b32 s0, s0, 3
+; GFX7-NEXT:    s_lshl_b32 s5, s1, s0
+; GFX7-NEXT:    s_lshl_b32 s0, s6, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX7-NEXT:    s_not_b32 s7, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 8, v2
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX7-NEXT:    v_and_b32_e32 v7, s6, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 24, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 8, v3
+; GFX7-NEXT:    v_and_b32_e32 v5, s6, v5
+; GFX7-NEXT:    v_and_b32_e32 v8, s6, v8
+; GFX7-NEXT:    v_and_b32_e32 v10, s6, v10
+; GFX7-NEXT:    v_and_b32_e32 v0, s6, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 8, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 16, v3
+; GFX7-NEXT:    v_and_b32_e32 v6, s6, v6
+; GFX7-NEXT:    v_and_b32_e32 v9, s6, v9
+; GFX7-NEXT:    v_and_b32_e32 v11, s6, v11
+; GFX7-NEXT:    v_and_b32_e32 v13, s6, v13
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v10, 8, v10
+; GFX7-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 16, v8
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 24, v3
+; GFX7-NEXT:    v_and_b32_e32 v12, s6, v12
+; GFX7-NEXT:    v_and_b32_e32 v14, s6, v14
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v13, 8, v13
+; GFX7-NEXT:    v_lshlrev_b32_e32 v6, 24, v6
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v9, 24, v9
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v11, 16, v11
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v10
+; GFX7-NEXT:    v_and_b32_e32 v15, s6, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v12, 24, v12
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v14, 16, v14
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v13
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v6
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v15, 24, v15
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v14
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v12
+; GFX7-NEXT:    v_cndmask_b32_e32 v4, v0, v1, vcc
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v15
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v2, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v3, s[2:3]
+; GFX7-NEXT:    v_and_b32_e32 v4, s7, v4
+; GFX7-NEXT:    v_or_b32_e32 v4, s5, v4
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[2:3]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s6, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v10
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 24, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v12
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v13
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 24, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v14
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i8>, <16 x i8> addrspace(1 )* %ptr
+  %insert = insertelement <16 x i8> %vec, i8 %val, i32 %idx
+  store <16 x i8> %insert, <16 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v16i8_v_s(<16 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_s_v16i8_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s18, 0xff
+; GFX9-NEXT:    v_and_b32_e32 v0, s18, v0
+; GFX9-NEXT:    s_mov_b32 s5, 8
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s6, s0, 8
+; GFX9-NEXT:    s_and_b32 s6, s6, s18
+; GFX9-NEXT:    s_lshr_b32 s7, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s8, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s6
+; GFX9-NEXT:    s_and_b32 s6, s7, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s6
+; GFX9-NEXT:    s_and_b32 s6, s8, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s6
+; GFX9-NEXT:    s_and_b32 s6, s9, s18
+; GFX9-NEXT:    s_lshr_b32 s10, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s11, s1, 24
+; GFX9-NEXT:    s_and_b32 s1, s1, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s6
+; GFX9-NEXT:    s_and_b32 s6, s10, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX9-NEXT:    s_or_b32 s1, s1, s6
+; GFX9-NEXT:    s_and_b32 s6, s11, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX9-NEXT:    s_lshr_b32 s12, s2, 8
+; GFX9-NEXT:    s_or_b32 s1, s1, s6
+; GFX9-NEXT:    s_and_b32 s6, s12, s18
+; GFX9-NEXT:    s_lshr_b32 s13, s2, 16
+; GFX9-NEXT:    s_lshr_b32 s14, s2, 24
+; GFX9-NEXT:    s_and_b32 s2, s2, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_or_b32 s2, s2, s6
+; GFX9-NEXT:    s_and_b32 s6, s13, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX9-NEXT:    s_or_b32 s2, s2, s6
+; GFX9-NEXT:    s_and_b32 s6, s14, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX9-NEXT:    s_lshr_b32 s15, s3, 8
+; GFX9-NEXT:    s_or_b32 s2, s2, s6
+; GFX9-NEXT:    s_and_b32 s6, s15, s18
+; GFX9-NEXT:    s_lshr_b32 s16, s3, 16
+; GFX9-NEXT:    s_lshr_b32 s17, s3, 24
+; GFX9-NEXT:    s_and_b32 s3, s3, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 8
+; GFX9-NEXT:    s_or_b32 s3, s3, s6
+; GFX9-NEXT:    s_and_b32 s6, s16, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 16
+; GFX9-NEXT:    s_or_b32 s3, s3, s6
+; GFX9-NEXT:    s_and_b32 s6, s17, s18
+; GFX9-NEXT:    s_lshl_b32 s6, s6, 24
+; GFX9-NEXT:    s_or_b32 s3, s3, s6
+; GFX9-NEXT:    s_lshr_b32 s6, s4, 2
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 1
+; GFX9-NEXT:    s_cselect_b32 s7, s1, s0
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 2
+; GFX9-NEXT:    s_cselect_b32 s7, s2, s7
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 3
+; GFX9-NEXT:    s_cselect_b32 s7, s3, s7
+; GFX9-NEXT:    s_and_b32 s4, s4, 3
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 3
+; GFX9-NEXT:    s_lshl_b32 s8, s18, s4
+; GFX9-NEXT:    s_not_b32 s8, s8
+; GFX9-NEXT:    s_and_b32 s7, s7, s8
+; GFX9-NEXT:    v_mov_b32_e32 v1, s7
+; GFX9-NEXT:    v_lshl_or_b32 v4, v0, s4, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s6, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s6, 1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s6, 2
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s6, 3
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, s5, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v4, v0, s18, v4
+; GFX9-NEXT:    v_and_b32_sdwa v8, v0, s18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s18 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v0, v4, v8, v0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX9-NEXT:    v_mov_b32_e32 v4, 8
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
+; GFX9-NEXT:    v_and_or_b32 v5, v1, s18, v5
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, s18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v1, s18 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v1, v5, v8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v5, v2, s18, v5
+; GFX9-NEXT:    v_and_b32_sdwa v6, v2, s18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v2, v2, s18 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v2, v5, v6, v2
+; GFX9-NEXT:    v_and_or_b32 v4, v3, s18, v4
+; GFX9-NEXT:    v_and_b32_sdwa v5, v3, s18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v3, v3, s18 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v3, v4, v5, v3
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v16i8_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s17, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v8, 8
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s5, s0, 8
+; GFX8-NEXT:    s_and_b32 s5, s5, s17
+; GFX8-NEXT:    s_lshr_b32 s6, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s7, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s5
+; GFX8-NEXT:    s_and_b32 s5, s6, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s5
+; GFX8-NEXT:    s_and_b32 s5, s7, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX8-NEXT:    s_lshr_b32 s8, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s5
+; GFX8-NEXT:    s_and_b32 s5, s8, s17
+; GFX8-NEXT:    s_lshr_b32 s9, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s10, s1, 24
+; GFX8-NEXT:    s_and_b32 s1, s1, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s5
+; GFX8-NEXT:    s_and_b32 s5, s9, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX8-NEXT:    s_or_b32 s1, s1, s5
+; GFX8-NEXT:    s_and_b32 s5, s10, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX8-NEXT:    s_lshr_b32 s11, s2, 8
+; GFX8-NEXT:    s_or_b32 s1, s1, s5
+; GFX8-NEXT:    s_and_b32 s5, s11, s17
+; GFX8-NEXT:    s_lshr_b32 s12, s2, 16
+; GFX8-NEXT:    s_lshr_b32 s13, s2, 24
+; GFX8-NEXT:    s_and_b32 s2, s2, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX8-NEXT:    s_or_b32 s2, s2, s5
+; GFX8-NEXT:    s_and_b32 s5, s12, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX8-NEXT:    s_or_b32 s2, s2, s5
+; GFX8-NEXT:    s_and_b32 s5, s13, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX8-NEXT:    s_lshr_b32 s14, s3, 8
+; GFX8-NEXT:    s_or_b32 s2, s2, s5
+; GFX8-NEXT:    s_and_b32 s5, s14, s17
+; GFX8-NEXT:    s_lshr_b32 s15, s3, 16
+; GFX8-NEXT:    s_lshr_b32 s16, s3, 24
+; GFX8-NEXT:    s_and_b32 s3, s3, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX8-NEXT:    s_or_b32 s3, s3, s5
+; GFX8-NEXT:    s_and_b32 s5, s15, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX8-NEXT:    s_or_b32 s3, s3, s5
+; GFX8-NEXT:    s_and_b32 s5, s16, s17
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX8-NEXT:    s_or_b32 s3, s3, s5
+; GFX8-NEXT:    s_lshr_b32 s5, s4, 2
+; GFX8-NEXT:    s_cmp_eq_u32 s5, 1
+; GFX8-NEXT:    s_cselect_b32 s6, s1, s0
+; GFX8-NEXT:    s_cmp_eq_u32 s5, 2
+; GFX8-NEXT:    s_cselect_b32 s6, s2, s6
+; GFX8-NEXT:    s_cmp_eq_u32 s5, 3
+; GFX8-NEXT:    s_cselect_b32 s6, s3, s6
+; GFX8-NEXT:    s_and_b32 s4, s4, 3
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 3
+; GFX8-NEXT:    v_mov_b32_e32 v1, s4
+; GFX8-NEXT:    s_lshl_b32 s4, s17, s4
+; GFX8-NEXT:    s_not_b32 s4, s4
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    s_and_b32 s4, s6, s4
+; GFX8-NEXT:    v_or_b32_e32 v4, s4, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v1, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v2, s2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GFX8-NEXT:    v_mov_b32_e32 v3, s3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 3
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v8, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v8, s17
+; GFX8-NEXT:    v_or_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v9, v0, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v9
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v4, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_sdwa v5, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v9, v1, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v9
+; GFX8-NEXT:    v_and_b32_sdwa v1, v1, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v5, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v3
+; GFX8-NEXT:    v_or_b32_sdwa v5, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v6, v2, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v6
+; GFX8-NEXT:    v_and_b32_sdwa v2, v2, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v5, v2
+; GFX8-NEXT:    v_or_b32_sdwa v4, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v5, v3, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v5
+; GFX8-NEXT:    v_and_b32_sdwa v3, v3, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v3, v4, v3
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v16i8_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s17, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v0, s17, v0
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s5, s0, 8
+; GFX7-NEXT:    s_and_b32 s5, s5, s17
+; GFX7-NEXT:    s_lshr_b32 s6, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s7, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s5
+; GFX7-NEXT:    s_and_b32 s5, s6, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s5
+; GFX7-NEXT:    s_and_b32 s5, s7, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX7-NEXT:    s_lshr_b32 s8, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s5
+; GFX7-NEXT:    s_and_b32 s5, s8, s17
+; GFX7-NEXT:    s_lshr_b32 s9, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s10, s1, 24
+; GFX7-NEXT:    s_and_b32 s1, s1, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX7-NEXT:    s_or_b32 s1, s1, s5
+; GFX7-NEXT:    s_and_b32 s5, s9, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX7-NEXT:    s_or_b32 s1, s1, s5
+; GFX7-NEXT:    s_and_b32 s5, s10, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX7-NEXT:    s_lshr_b32 s11, s2, 8
+; GFX7-NEXT:    s_or_b32 s1, s1, s5
+; GFX7-NEXT:    s_and_b32 s5, s11, s17
+; GFX7-NEXT:    s_lshr_b32 s12, s2, 16
+; GFX7-NEXT:    s_lshr_b32 s13, s2, 24
+; GFX7-NEXT:    s_and_b32 s2, s2, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX7-NEXT:    s_or_b32 s2, s2, s5
+; GFX7-NEXT:    s_and_b32 s5, s12, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX7-NEXT:    s_or_b32 s2, s2, s5
+; GFX7-NEXT:    s_and_b32 s5, s13, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX7-NEXT:    s_lshr_b32 s14, s3, 8
+; GFX7-NEXT:    s_or_b32 s2, s2, s5
+; GFX7-NEXT:    s_and_b32 s5, s14, s17
+; GFX7-NEXT:    s_lshr_b32 s15, s3, 16
+; GFX7-NEXT:    s_lshr_b32 s16, s3, 24
+; GFX7-NEXT:    s_and_b32 s3, s3, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX7-NEXT:    s_or_b32 s3, s3, s5
+; GFX7-NEXT:    s_and_b32 s5, s15, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX7-NEXT:    s_or_b32 s3, s3, s5
+; GFX7-NEXT:    s_and_b32 s5, s16, s17
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX7-NEXT:    s_or_b32 s3, s3, s5
+; GFX7-NEXT:    s_lshr_b32 s5, s4, 2
+; GFX7-NEXT:    s_cmp_eq_u32 s5, 1
+; GFX7-NEXT:    s_cselect_b32 s6, s1, s0
+; GFX7-NEXT:    s_cmp_eq_u32 s5, 2
+; GFX7-NEXT:    s_cselect_b32 s6, s2, s6
+; GFX7-NEXT:    s_cmp_eq_u32 s5, 3
+; GFX7-NEXT:    s_cselect_b32 s6, s3, s6
+; GFX7-NEXT:    s_and_b32 s4, s4, 3
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s4, v0
+; GFX7-NEXT:    s_lshl_b32 s4, s17, s4
+; GFX7-NEXT:    s_not_b32 s4, s4
+; GFX7-NEXT:    s_and_b32 s4, s6, s4
+; GFX7-NEXT:    v_or_b32_e32 v4, s4, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 0
+; GFX7-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s1
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 1
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v2, s2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 2
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v3, s3
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s5, 3
+; GFX7-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s17, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s17, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v10
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 24, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s17, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v12
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v13
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 24, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s17, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v14
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    s_mov_b64 s[0:1], 0
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    s_mov_b32 s2, -1
+; GFX7-NEXT:    s_mov_b32 s3, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i8>, <16 x i8> addrspace(4)* %ptr
+  %insert = insertelement <16 x i8> %vec, i8 %val, i32 %idx
+  store <16 x i8> %insert, <16 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v16i8_s_v(<16 x i8> addrspace(4)* inreg %ptr, i8 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v16i8_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s18, 0xff
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 2, v0
+; GFX9-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s5, s0, 8
+; GFX9-NEXT:    s_and_b32 s5, s5, s18
+; GFX9-NEXT:    s_lshr_b32 s7, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s8, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s18
+; GFX9-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s5
+; GFX9-NEXT:    s_and_b32 s5, s7, s18
+; GFX9-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s5
+; GFX9-NEXT:    s_and_b32 s5, s8, s18
+; GFX9-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 8
+; GFX9-NEXT:    s_or_b32 s8, s0, s5
+; GFX9-NEXT:    s_lshr_b32 s10, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s11, s1, 24
+; GFX9-NEXT:    s_and_b32 s0, s1, s18
+; GFX9-NEXT:    s_and_b32 s1, s9, s18
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s10, s18
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s11, s18
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX9-NEXT:    s_lshr_b32 s12, s2, 8
+; GFX9-NEXT:    s_or_b32 s9, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s12, s18
+; GFX9-NEXT:    s_lshr_b32 s13, s2, 16
+; GFX9-NEXT:    s_and_b32 s0, s2, s18
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s13, s18
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s14, s2, 24
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s14, s18
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX9-NEXT:    s_lshr_b32 s15, s3, 8
+; GFX9-NEXT:    s_or_b32 s10, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s15, s18
+; GFX9-NEXT:    s_lshr_b32 s16, s3, 16
+; GFX9-NEXT:    s_and_b32 s0, s3, s18
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s16, s18
+; GFX9-NEXT:    s_lshr_b32 s17, s3, 24
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s17, s18
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX9-NEXT:    v_mov_b32_e32 v1, s8
+; GFX9-NEXT:    v_mov_b32_e32 v2, s9
+; GFX9-NEXT:    s_or_b32 s11, s0, s1
+; GFX9-NEXT:    v_mov_b32_e32 v3, s10
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX9-NEXT:    s_and_b32 s4, s4, s18
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v0, s4
+; GFX9-NEXT:    v_lshlrev_b32_e64 v0, v0, s18
+; GFX9-NEXT:    v_mov_b32_e32 v5, s11
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[2:3]
+; GFX9-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX9-NEXT:    v_and_or_b32 v5, v1, v0, v2
+; GFX9-NEXT:    v_mov_b32_e32 v0, s8
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX9-NEXT:    s_mov_b32 s6, 8
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s9
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, s6, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_mov_b32_e32 v2, s10
+; GFX9-NEXT:    v_mov_b32_e32 v3, s11
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX9-NEXT:    v_and_or_b32 v4, v0, s18, v4
+; GFX9-NEXT:    v_and_b32_sdwa v8, v0, s18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s18 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v0, v4, v8, v0
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX9-NEXT:    v_mov_b32_e32 v4, 8
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
+; GFX9-NEXT:    v_and_or_b32 v5, v1, s18, v5
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, s18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v1, s18 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v1, v5, v8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v5, v2, s18, v5
+; GFX9-NEXT:    v_and_b32_sdwa v6, v2, s18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v2, v2, s18 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v2, v5, v6, v2
+; GFX9-NEXT:    v_and_or_b32 v4, v3, s18, v4
+; GFX9-NEXT:    v_and_b32_sdwa v5, v3, s18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v3, v3, s18 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v3, v4, v5, v3
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v16i8_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s18, 0xff
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 2, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s5, s0, 8
+; GFX8-NEXT:    s_and_b32 s5, s5, s18
+; GFX8-NEXT:    s_lshr_b32 s6, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s7, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s18
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s5
+; GFX8-NEXT:    s_and_b32 s5, s6, s18
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s5
+; GFX8-NEXT:    s_and_b32 s5, s7, s18
+; GFX8-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX8-NEXT:    s_lshr_b32 s9, s1, 8
+; GFX8-NEXT:    s_or_b32 s8, s0, s5
+; GFX8-NEXT:    s_lshr_b32 s10, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s11, s1, 24
+; GFX8-NEXT:    s_and_b32 s0, s1, s18
+; GFX8-NEXT:    s_and_b32 s1, s9, s18
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s10, s18
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s11, s18
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    s_lshr_b32 s12, s2, 8
+; GFX8-NEXT:    s_or_b32 s9, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s12, s18
+; GFX8-NEXT:    s_lshr_b32 s13, s2, 16
+; GFX8-NEXT:    s_and_b32 s0, s2, s18
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s13, s18
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s14, s2, 24
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s14, s18
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    s_lshr_b32 s15, s3, 8
+; GFX8-NEXT:    s_or_b32 s10, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s15, s18
+; GFX8-NEXT:    s_lshr_b32 s16, s3, 16
+; GFX8-NEXT:    s_and_b32 s0, s3, s18
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s16, s18
+; GFX8-NEXT:    s_lshr_b32 s17, s3, 24
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s17, s18
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    v_mov_b32_e32 v1, s8
+; GFX8-NEXT:    v_mov_b32_e32 v2, s9
+; GFX8-NEXT:    s_or_b32 s11, s0, s1
+; GFX8-NEXT:    v_mov_b32_e32 v3, s10
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX8-NEXT:    s_and_b32 s4, s4, s18
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v0, s4
+; GFX8-NEXT:    v_lshlrev_b32_e64 v0, v0, s18
+; GFX8-NEXT:    v_mov_b32_e32 v5, s11
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[2:3]
+; GFX8-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX8-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX8-NEXT:    v_or_b32_e32 v5, v0, v2
+; GFX8-NEXT:    v_mov_b32_e32 v0, s8
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX8-NEXT:    v_mov_b32_e32 v8, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v8, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v8, s18
+; GFX8-NEXT:    v_mov_b32_e32 v1, s9
+; GFX8-NEXT:    v_or_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v9, v0, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_mov_b32_e32 v2, s10
+; GFX8-NEXT:    v_mov_b32_e32 v3, s11
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v9
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX8-NEXT:    v_or_b32_e32 v0, v4, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_sdwa v5, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v9, v1, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v9
+; GFX8-NEXT:    v_and_b32_sdwa v1, v1, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v5, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v3
+; GFX8-NEXT:    v_or_b32_sdwa v5, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v6, v2, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v6
+; GFX8-NEXT:    v_and_b32_sdwa v2, v2, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v5, v2
+; GFX8-NEXT:    v_or_b32_sdwa v4, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v5, v3, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v5
+; GFX8-NEXT:    v_and_b32_sdwa v3, v3, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v3, v4, v3
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v16i8_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s18, 0xff
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 2, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, 3, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s5, s0, 8
+; GFX7-NEXT:    s_and_b32 s5, s5, s18
+; GFX7-NEXT:    s_lshr_b32 s6, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s7, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s18
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s5
+; GFX7-NEXT:    s_and_b32 s5, s6, s18
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s5
+; GFX7-NEXT:    s_and_b32 s5, s7, s18
+; GFX7-NEXT:    s_lshl_b32 s5, s5, 24
+; GFX7-NEXT:    s_lshr_b32 s9, s1, 8
+; GFX7-NEXT:    s_or_b32 s8, s0, s5
+; GFX7-NEXT:    s_lshr_b32 s10, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s11, s1, 24
+; GFX7-NEXT:    s_and_b32 s0, s1, s18
+; GFX7-NEXT:    s_and_b32 s1, s9, s18
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s10, s18
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s11, s18
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    s_lshr_b32 s12, s2, 8
+; GFX7-NEXT:    s_or_b32 s9, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s12, s18
+; GFX7-NEXT:    s_lshr_b32 s13, s2, 16
+; GFX7-NEXT:    s_and_b32 s0, s2, s18
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s13, s18
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s14, s2, 24
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s14, s18
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    s_lshr_b32 s15, s3, 8
+; GFX7-NEXT:    s_or_b32 s10, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s15, s18
+; GFX7-NEXT:    s_lshr_b32 s16, s3, 16
+; GFX7-NEXT:    s_and_b32 s0, s3, s18
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s16, s18
+; GFX7-NEXT:    s_lshr_b32 s17, s3, 24
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s17, s18
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    v_mov_b32_e32 v1, s8
+; GFX7-NEXT:    v_mov_b32_e32 v2, s9
+; GFX7-NEXT:    s_or_b32 s11, s0, s1
+; GFX7-NEXT:    v_mov_b32_e32 v3, s10
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
+; GFX7-NEXT:    s_and_b32 s4, s4, s18
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s4, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v0, s18, v0
+; GFX7-NEXT:    v_mov_b32_e32 v5, s11
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v5, s[2:3]
+; GFX7-NEXT:    v_xor_b32_e32 v0, -1, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_or_b32_e32 v5, v0, v2
+; GFX7-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s9
+; GFX7-NEXT:    v_mov_b32_e32 v2, s10
+; GFX7-NEXT:    v_mov_b32_e32 v3, s11
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v4
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s18, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s18, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v10
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 24, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s18, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v12
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v13
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 24, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s18, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v14
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s18, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i8>, <16 x i8> addrspace(4)* %ptr
+  %insert = insertelement <16 x i8> %vec, i8 %val, i32 %idx
+  store <16 x i8> %insert, <16 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_s_v16i8_v_v(<16 x i8> addrspace(4)* inreg %ptr, i8 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_s_v16i8_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX9-NEXT:    s_movk_i32 s17, 0xff
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 2, v1
+; GFX9-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b32 s4, s0, 8
+; GFX9-NEXT:    s_and_b32 s4, s4, s17
+; GFX9-NEXT:    s_lshr_b32 s5, s0, 16
+; GFX9-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX9-NEXT:    s_and_b32 s0, s0, s17
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s4
+; GFX9-NEXT:    s_and_b32 s4, s5, s17
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s4
+; GFX9-NEXT:    s_and_b32 s4, s6, s17
+; GFX9-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX9-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX9-NEXT:    s_or_b32 s4, s0, s4
+; GFX9-NEXT:    s_lshr_b32 s9, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s10, s1, 24
+; GFX9-NEXT:    s_and_b32 s0, s1, s17
+; GFX9-NEXT:    s_and_b32 s1, s7, s17
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s9, s17
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s10, s17
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX9-NEXT:    s_lshr_b32 s11, s2, 8
+; GFX9-NEXT:    s_or_b32 s5, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s11, s17
+; GFX9-NEXT:    s_lshr_b32 s12, s2, 16
+; GFX9-NEXT:    s_and_b32 s0, s2, s17
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s12, s17
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX9-NEXT:    s_lshr_b32 s13, s2, 24
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s13, s17
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX9-NEXT:    s_lshr_b32 s14, s3, 8
+; GFX9-NEXT:    s_or_b32 s6, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s14, s17
+; GFX9-NEXT:    s_lshr_b32 s15, s3, 16
+; GFX9-NEXT:    s_and_b32 s0, s3, s17
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s15, s17
+; GFX9-NEXT:    s_lshr_b32 s16, s3, 24
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX9-NEXT:    s_or_b32 s0, s0, s1
+; GFX9-NEXT:    s_and_b32 s1, s16, s17
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX9-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NEXT:    s_or_b32 s7, s0, s1
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX9-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s17
+; GFX9-NEXT:    v_mov_b32_e32 v6, s7
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[2:3]
+; GFX9-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX9-NEXT:    v_and_or_b32 v5, v2, v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NEXT:    v_mov_b32_e32 v2, s6
+; GFX9-NEXT:    v_mov_b32_e32 v3, s7
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX9-NEXT:    s_mov_b32 s8, 8
+; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, s8, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX9-NEXT:    v_and_or_b32 v4, v0, s17, v4
+; GFX9-NEXT:    v_and_b32_sdwa v8, v0, s17 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, s17 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v0, v4, v8, v0
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX9-NEXT:    v_mov_b32_e32 v4, 8
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
+; GFX9-NEXT:    v_and_or_b32 v5, v1, s17, v5
+; GFX9-NEXT:    v_and_b32_sdwa v8, v1, s17 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v1, s17 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v1, v5, v8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v5, v2, s17, v5
+; GFX9-NEXT:    v_and_b32_sdwa v6, v2, s17 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v2, v2, s17 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v2, v5, v6, v2
+; GFX9-NEXT:    v_and_or_b32 v4, v3, s17, v4
+; GFX9-NEXT:    v_and_b32_sdwa v5, v3, s17 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v3, v3, s17 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v3, v4, v5, v3
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_s_v16i8_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX8-NEXT:    s_movk_i32 s16, 0xff
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 2, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    s_lshr_b32 s4, s0, 8
+; GFX8-NEXT:    s_and_b32 s4, s4, s16
+; GFX8-NEXT:    s_lshr_b32 s5, s0, 16
+; GFX8-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX8-NEXT:    s_and_b32 s0, s0, s16
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s4
+; GFX8-NEXT:    s_and_b32 s4, s5, s16
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s4
+; GFX8-NEXT:    s_and_b32 s4, s6, s16
+; GFX8-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX8-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX8-NEXT:    s_or_b32 s4, s0, s4
+; GFX8-NEXT:    s_lshr_b32 s8, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s9, s1, 24
+; GFX8-NEXT:    s_and_b32 s0, s1, s16
+; GFX8-NEXT:    s_and_b32 s1, s7, s16
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s8, s16
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s9, s16
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    s_lshr_b32 s10, s2, 8
+; GFX8-NEXT:    s_or_b32 s5, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s10, s16
+; GFX8-NEXT:    s_lshr_b32 s11, s2, 16
+; GFX8-NEXT:    s_and_b32 s0, s2, s16
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s11, s16
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_lshr_b32 s12, s2, 24
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s12, s16
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    s_lshr_b32 s13, s3, 8
+; GFX8-NEXT:    s_or_b32 s6, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s13, s16
+; GFX8-NEXT:    s_lshr_b32 s14, s3, 16
+; GFX8-NEXT:    s_and_b32 s0, s3, s16
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s14, s16
+; GFX8-NEXT:    s_lshr_b32 s15, s3, 24
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX8-NEXT:    s_or_b32 s0, s0, s1
+; GFX8-NEXT:    s_and_b32 s1, s15, s16
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX8-NEXT:    v_mov_b32_e32 v2, s4
+; GFX8-NEXT:    v_mov_b32_e32 v3, s5
+; GFX8-NEXT:    s_or_b32 s7, s0, s1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX8-NEXT:    v_mov_b32_e32 v5, s6
+; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_e64 v1, v1, s16
+; GFX8-NEXT:    v_mov_b32_e32 v6, s7
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[2:3]
+; GFX8-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX8-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX8-NEXT:    v_or_b32_e32 v5, v1, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s4
+; GFX8-NEXT:    v_mov_b32_e32 v1, s5
+; GFX8-NEXT:    v_mov_b32_e32 v2, s6
+; GFX8-NEXT:    v_mov_b32_e32 v3, s7
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX8-NEXT:    v_mov_b32_e32 v8, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v8, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v8, s16
+; GFX8-NEXT:    v_or_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v9, v0, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v9
+; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX8-NEXT:    v_or_b32_e32 v0, v4, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX8-NEXT:    v_mov_b32_e32 v4, 8
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_sdwa v5, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v9, v1, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v2
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v9
+; GFX8-NEXT:    v_and_b32_sdwa v1, v1, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v1, v5, v1
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v3
+; GFX8-NEXT:    v_or_b32_sdwa v5, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v6, v2, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v4, v4, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v6
+; GFX8-NEXT:    v_and_b32_sdwa v2, v2, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v5, v2
+; GFX8-NEXT:    v_or_b32_sdwa v4, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v5, v3, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v5
+; GFX8-NEXT:    v_and_b32_sdwa v3, v3, v8 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v3, v4, v3
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_s_v16i8_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x0
+; GFX7-NEXT:    s_movk_i32 s17, 0xff
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 2, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, 3, v1
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v4
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_lshr_b32 s4, s0, 8
+; GFX7-NEXT:    s_and_b32 s4, s4, s17
+; GFX7-NEXT:    s_lshr_b32 s5, s0, 16
+; GFX7-NEXT:    s_lshr_b32 s6, s0, 24
+; GFX7-NEXT:    s_and_b32 s0, s0, s17
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s4
+; GFX7-NEXT:    s_and_b32 s4, s5, s17
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s4
+; GFX7-NEXT:    s_and_b32 s4, s6, s17
+; GFX7-NEXT:    s_lshl_b32 s4, s4, 24
+; GFX7-NEXT:    s_lshr_b32 s7, s1, 8
+; GFX7-NEXT:    s_or_b32 s8, s0, s4
+; GFX7-NEXT:    s_lshr_b32 s9, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s10, s1, 24
+; GFX7-NEXT:    s_and_b32 s0, s1, s17
+; GFX7-NEXT:    s_and_b32 s1, s7, s17
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s9, s17
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s10, s17
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    s_lshr_b32 s11, s2, 8
+; GFX7-NEXT:    s_or_b32 s9, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s11, s17
+; GFX7-NEXT:    s_lshr_b32 s12, s2, 16
+; GFX7-NEXT:    s_and_b32 s0, s2, s17
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s12, s17
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_lshr_b32 s13, s2, 24
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s13, s17
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    s_lshr_b32 s14, s3, 8
+; GFX7-NEXT:    s_or_b32 s10, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s14, s17
+; GFX7-NEXT:    s_lshr_b32 s15, s3, 16
+; GFX7-NEXT:    s_and_b32 s0, s3, s17
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 8
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s15, s17
+; GFX7-NEXT:    s_lshr_b32 s16, s3, 24
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 16
+; GFX7-NEXT:    s_or_b32 s0, s0, s1
+; GFX7-NEXT:    s_and_b32 s1, s16, s17
+; GFX7-NEXT:    s_lshl_b32 s1, s1, 24
+; GFX7-NEXT:    v_mov_b32_e32 v2, s8
+; GFX7-NEXT:    v_mov_b32_e32 v3, s9
+; GFX7-NEXT:    s_or_b32 s11, s0, s1
+; GFX7-NEXT:    v_mov_b32_e32 v5, s10
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 3, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, s17, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_lshl_b32_e32 v1, s17, v1
+; GFX7-NEXT:    v_mov_b32_e32 v6, s11
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v4
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[2:3]
+; GFX7-NEXT:    v_xor_b32_e32 v1, -1, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v2, v1
+; GFX7-NEXT:    v_or_b32_e32 v5, v1, v0
+; GFX7-NEXT:    v_mov_b32_e32 v0, s8
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v4
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[4:5]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 8, v0
+; GFX7-NEXT:    v_mov_b32_e32 v1, s9
+; GFX7-NEXT:    v_mov_b32_e32 v2, s10
+; GFX7-NEXT:    v_mov_b32_e32 v3, s11
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v4
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[2:3]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s17, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s17, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v10
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 24, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s17, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v12
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v13
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 24, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s17, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v14
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s17, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_mov_b64 s[4:5], 0
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i8>, <16 x i8> addrspace(4)* %ptr
+  %insert = insertelement <16 x i8> %vec, i8 %val, i32 %idx
+  store <16 x i8> %insert, <16 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v16i8_s_v(<16 x i8> addrspace(1)* %ptr, i8 inreg %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v16i8_s_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[3:6], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_mov_b32_e32 v0, 8
+; GFX9-NEXT:    s_movk_i32 s6, 0xff
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 2, v2
+; GFX9-NEXT:    v_and_b32_e32 v2, 3, v2
+; GFX9-NEXT:    s_and_b32 s1, s2, s6
+; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v2
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v7, v2, s1
+; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v2, s6
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v1
+; GFX9-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v1
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 8, v4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v10, 8, v5
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v8, s0, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v9, v0, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v11, 8, v6
+; GFX9-NEXT:    v_and_b32_sdwa v12, v3, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v13, v3, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v3, v3, s6, v8
+; GFX9-NEXT:    v_and_b32_sdwa v14, v4, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v15, v4, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v4, v4, s6, v9
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v10, v0, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v16, v5, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v17, v5, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v5, v5, s6, v10
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v11, v0, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_or3_b32 v3, v3, v12, v13
+; GFX9-NEXT:    v_or3_b32 v4, v4, v14, v15
+; GFX9-NEXT:    v_and_b32_sdwa v18, v6, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v19, v6, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v6, v6, s6, v11
+; GFX9-NEXT:    v_or3_b32 v5, v5, v16, v17
+; GFX9-NEXT:    v_cndmask_b32_e32 v8, v3, v4, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v1
+; GFX9-NEXT:    v_or3_b32 v6, v6, v18, v19
+; GFX9-NEXT:    v_cndmask_b32_e64 v8, v8, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v8, v8, v6, s[2:3]
+; GFX9-NEXT:    v_and_or_b32 v2, v8, v2, v7
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v3, v2, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v4, v2, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v5, v2, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v6, v2, s[2:3]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v2
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v6, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v7, v0, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v0, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v9, v1, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v10, v1, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v1, v1, s6, v5
+; GFX9-NEXT:    v_and_or_b32 v5, v2, s6, v0
+; GFX9-NEXT:    v_and_b32_sdwa v11, v3, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v12, v3, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v13, v4, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v14, v4, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v15, v2, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v16, v2, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v4, v4, s6, v7
+; GFX9-NEXT:    v_and_or_b32 v3, v3, s6, v6
+; GFX9-NEXT:    v_or3_b32 v0, v1, v9, v10
+; GFX9-NEXT:    v_or3_b32 v1, v3, v11, v12
+; GFX9-NEXT:    v_or3_b32 v2, v4, v13, v14
+; GFX9-NEXT:    v_or3_b32 v3, v5, v15, v16
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v16i8_s_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[3:6], v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v0, 8
+; GFX8-NEXT:    v_mov_b32_e32 v1, 8
+; GFX8-NEXT:    v_mov_b32_e32 v7, s0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 2, v2
+; GFX8-NEXT:    v_and_b32_e32 v2, 3, v2
+; GFX8-NEXT:    s_and_b32 s1, s2, s0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 3, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX8-NEXT:    v_lshlrev_b32_e64 v9, v2, s1
+; GFX8-NEXT:    v_lshlrev_b32_e64 v2, v2, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v8
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v8
+; GFX8-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v8
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v11, 8, v4
+; GFX8-NEXT:    v_lshrrev_b32_e32 v10, 8, v3
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v0, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v12, 8, v5
+; GFX8-NEXT:    v_lshrrev_b32_e32 v13, 8, v6
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v11, v1, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v14, v3, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v15, v3, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v16, v4, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v4, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v12, v1, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v13, v1, v13 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v14
+; GFX8-NEXT:    v_and_b32_sdwa v17, v4, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v16
+; GFX8-NEXT:    v_and_b32_sdwa v18, v5, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v4, v5, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v19, v5, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v10, v6, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v5, v6, v13 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v18
+; GFX8-NEXT:    v_and_b32_sdwa v6, v6, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v10
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v15
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v17
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v6
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v19
+; GFX8-NEXT:    v_cndmask_b32_e32 v6, v0, v3, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v6, v4, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v6, v5, s[2:3]
+; GFX8-NEXT:    v_and_b32_e32 v2, v6, v2
+; GFX8-NEXT:    v_or_b32_e32 v2, v2, v9
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v2, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[2:3]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 8, v4
+; GFX8-NEXT:    v_lshrrev_b32_e32 v9, 8, v2
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v6, v1, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v8, v1, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v1, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v12, v3, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v13, v3, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v14, v4, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v15, v4, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v16, v2, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v4, v4, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v10, v0, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v11, v0, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v7, v2, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v12
+; GFX8-NEXT:    v_or_b32_e32 v3, v4, v14
+; GFX8-NEXT:    v_or_b32_e32 v4, v1, v16
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v10
+; GFX8-NEXT:    v_or_b32_e32 v1, v2, v13
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v15
+; GFX8-NEXT:    v_or_b32_e32 v3, v4, v7
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v11
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v16i8_s_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s10, 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[3:6], v[0:1], s[8:11], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s6, 0xff
+; GFX7-NEXT:    v_lshrrev_b32_e32 v17, 2, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, 3, v2
+; GFX7-NEXT:    s_and_b32 s0, s2, s6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 3, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v17
+; GFX7-NEXT:    v_lshl_b32_e32 v18, s0, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v17
+; GFX7-NEXT:    v_lshl_b32_e32 v2, s6, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v17
+; GFX7-NEXT:    v_xor_b32_e32 v2, -1, v2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v17
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v0, 8, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 8, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 16, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 8, v5
+; GFX7-NEXT:    v_and_b32_e32 v0, s6, v0
+; GFX7-NEXT:    v_and_b32_e32 v8, s6, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 16, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 8, v6
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_and_b32_e32 v9, s6, v9
+; GFX7-NEXT:    v_and_b32_e32 v11, s6, v11
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 8, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 24, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 16, v6
+; GFX7-NEXT:    v_and_b32_e32 v7, s6, v7
+; GFX7-NEXT:    v_and_b32_e32 v10, s6, v10
+; GFX7-NEXT:    v_and_b32_e32 v12, s6, v12
+; GFX7-NEXT:    v_and_b32_e32 v14, s6, v14
+; GFX7-NEXT:    v_or_b32_e32 v0, v3, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_and_b32_e32 v5, s6, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v9, 16, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v11, 8, v11
+; GFX7-NEXT:    v_or_b32_e32 v3, v4, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v16, 24, v6
+; GFX7-NEXT:    v_and_b32_e32 v13, s6, v13
+; GFX7-NEXT:    v_and_b32_e32 v15, s6, v15
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 24, v7
+; GFX7-NEXT:    v_and_b32_e32 v6, s6, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v10, 24, v10
+; GFX7-NEXT:    v_or_b32_e32 v1, v3, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v12, 16, v12
+; GFX7-NEXT:    v_or_b32_e32 v4, v5, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v14, 8, v14
+; GFX7-NEXT:    v_and_b32_e32 v16, s6, v16
+; GFX7-NEXT:    v_lshlrev_b32_e32 v13, 24, v13
+; GFX7-NEXT:    v_or_b32_e32 v3, v4, v12
+; GFX7-NEXT:    v_lshlrev_b32_e32 v15, 16, v15
+; GFX7-NEXT:    v_or_b32_e32 v5, v6, v14
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v7
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v10
+; GFX7-NEXT:    v_or_b32_e32 v4, v5, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v16, 24, v16
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v13
+; GFX7-NEXT:    v_cndmask_b32_e32 v5, v0, v1, vcc
+; GFX7-NEXT:    v_or_b32_e32 v4, v4, v16
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v5, v3, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v5, v4, s[2:3]
+; GFX7-NEXT:    v_and_b32_e32 v2, v5, v2
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v18
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v2, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v2, s[2:3]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, s6, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v7
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 24, v3
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v10
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v12
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 24, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 16, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 24, v4
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v13
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v14
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i8>, <16 x i8> addrspace(1)* %ptr
+  %insert = insertelement <16 x i8> %vec, i8 %val, i32 %idx
+  store <16 x i8> %insert, <16 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v16i8_v_s(<16 x i8> addrspace(1)* %ptr, i8 %val, i32 inreg %idx) {
+; GFX9-LABEL: insertelement_v_v16i8_v_s:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[3:6], v[0:1], off
+; GFX9-NEXT:    s_and_b32 s1, s2, 3
+; GFX9-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, s1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_mov_b32_e32 v0, 8
+; GFX9-NEXT:    s_movk_i32 s6, 0xff
+; GFX9-NEXT:    s_lshr_b32 s4, s2, 2
+; GFX9-NEXT:    s_lshl_b32 s1, s6, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX9-NEXT:    s_not_b32 s5, s1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 8, v3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v4
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, s0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v5
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v7, v0, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 8, v6
+; GFX9-NEXT:    v_and_b32_sdwa v10, v3, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v11, v3, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v2, v3, s6, v2
+; GFX9-NEXT:    v_and_b32_sdwa v12, v4, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v13, v4, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v3, v4, s6, v7
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v8, v0, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v14, v5, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v15, v5, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v4, v5, s6, v8
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v9, v0, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_or3_b32 v2, v2, v10, v11
+; GFX9-NEXT:    v_or3_b32 v3, v3, v12, v13
+; GFX9-NEXT:    v_and_b32_sdwa v16, v6, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v17, v6, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v5, v6, s6, v9
+; GFX9-NEXT:    v_or3_b32 v4, v4, v14, v15
+; GFX9-NEXT:    v_cndmask_b32_e32 v6, v2, v3, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX9-NEXT:    v_or3_b32 v5, v5, v16, v17
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v6, v4, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, v6, v5, s[2:3]
+; GFX9-NEXT:    v_and_or_b32 v1, v6, s5, v1
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v1, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, v4, v1, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v1, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v5, v1, s[2:3]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v5, 8, v2
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v1
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v5, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v6, v0, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v7, v0, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v0, v0, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v9, v2, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v10, v2, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v2, v2, s6, v5
+; GFX9-NEXT:    v_and_or_b32 v5, v1, s6, v0
+; GFX9-NEXT:    v_and_b32_sdwa v11, v3, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v12, v3, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v13, v4, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v14, v4, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v15, v1, s6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v16, v1, s6 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v3, v3, s6, v6
+; GFX9-NEXT:    v_and_or_b32 v4, v4, s6, v7
+; GFX9-NEXT:    v_or3_b32 v0, v2, v9, v10
+; GFX9-NEXT:    v_or3_b32 v1, v3, v11, v12
+; GFX9-NEXT:    v_or3_b32 v2, v4, v13, v14
+; GFX9-NEXT:    v_or3_b32 v3, v5, v15, v16
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v16i8_v_s:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[3:6], v[0:1]
+; GFX8-NEXT:    s_and_b32 s1, s2, 3
+; GFX8-NEXT:    s_lshl_b32 s1, s1, 3
+; GFX8-NEXT:    v_mov_b32_e32 v8, s1
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v8, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_mov_b32_e32 v0, 8
+; GFX8-NEXT:    v_mov_b32_e32 v1, 8
+; GFX8-NEXT:    v_mov_b32_e32 v7, s0
+; GFX8-NEXT:    s_lshr_b32 s4, s2, 2
+; GFX8-NEXT:    s_lshl_b32 s0, s0, s1
+; GFX8-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX8-NEXT:    s_not_b32 s5, s0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 8, v3
+; GFX8-NEXT:    v_lshrrev_b32_e32 v9, 8, v4
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v0, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v10, 8, v5
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v8, v1, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v11, 8, v6
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v9, v1, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v12, v3, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v13, v3, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v14, v4, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v4, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v12
+; GFX8-NEXT:    v_and_b32_sdwa v15, v4, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v14
+; GFX8-NEXT:    v_and_b32_sdwa v16, v5, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v4, v5, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v10, v1, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v17, v5, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v16
+; GFX8-NEXT:    v_and_b32_sdwa v18, v6, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v5, v6, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v13
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v15
+; GFX8-NEXT:    v_and_b32_sdwa v19, v6, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v18
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v17
+; GFX8-NEXT:    v_cndmask_b32_e32 v6, v0, v3, vcc
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v19
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v6, v4, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v6, v6, v5, s[2:3]
+; GFX8-NEXT:    v_and_b32_e32 v6, s5, v6
+; GFX8-NEXT:    v_or_b32_e32 v2, v6, v2
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v2, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v4, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[2:3]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
+; GFX8-NEXT:    v_lshrrev_b32_e32 v8, 8, v4
+; GFX8-NEXT:    v_lshrrev_b32_e32 v9, 8, v2
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v1, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v6, v1, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v8, v1, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v1, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v12, v3, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v13, v3, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v14, v4, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v15, v4, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v16, v2, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v4, v4, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v10, v0, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v11, v0, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v0, v0, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v7, v2, v7 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v12
+; GFX8-NEXT:    v_or_b32_e32 v3, v4, v14
+; GFX8-NEXT:    v_or_b32_e32 v4, v1, v16
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v10
+; GFX8-NEXT:    v_or_b32_e32 v1, v2, v13
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v15
+; GFX8-NEXT:    v_or_b32_e32 v3, v4, v7
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v11
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v16i8_v_s:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s10, 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[3:6], v[0:1], s[8:11], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s6, 0xff
+; GFX7-NEXT:    v_and_b32_e32 v0, s6, v2
+; GFX7-NEXT:    s_and_b32 s0, s2, 3
+; GFX7-NEXT:    s_lshr_b32 s4, s2, 2
+; GFX7-NEXT:    s_lshl_b32 s0, s0, 3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, s0, v0
+; GFX7-NEXT:    s_lshl_b32 s0, s6, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 vcc, s4, 1
+; GFX7-NEXT:    s_not_b32 s5, s0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], s4, 2
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, 3
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 8, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 8, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 16, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 8, v5
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_and_b32_e32 v8, s6, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 24, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 16, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 8, v6
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v2
+; GFX7-NEXT:    v_and_b32_e32 v9, s6, v9
+; GFX7-NEXT:    v_and_b32_e32 v11, s6, v11
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v8, 8, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 24, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 16, v6
+; GFX7-NEXT:    v_and_b32_e32 v7, s6, v7
+; GFX7-NEXT:    v_and_b32_e32 v10, s6, v10
+; GFX7-NEXT:    v_and_b32_e32 v12, s6, v12
+; GFX7-NEXT:    v_and_b32_e32 v14, s6, v14
+; GFX7-NEXT:    v_or_b32_e32 v1, v3, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_and_b32_e32 v5, s6, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v9, 16, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v11, 8, v11
+; GFX7-NEXT:    v_or_b32_e32 v3, v4, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v16, 24, v6
+; GFX7-NEXT:    v_and_b32_e32 v13, s6, v13
+; GFX7-NEXT:    v_and_b32_e32 v15, s6, v15
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v7, 24, v7
+; GFX7-NEXT:    v_and_b32_e32 v6, s6, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v10, 24, v10
+; GFX7-NEXT:    v_or_b32_e32 v2, v3, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v12, 16, v12
+; GFX7-NEXT:    v_or_b32_e32 v4, v5, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v14, 8, v14
+; GFX7-NEXT:    v_and_b32_e32 v16, s6, v16
+; GFX7-NEXT:    v_lshlrev_b32_e32 v13, 24, v13
+; GFX7-NEXT:    v_or_b32_e32 v3, v4, v12
+; GFX7-NEXT:    v_lshlrev_b32_e32 v15, 16, v15
+; GFX7-NEXT:    v_or_b32_e32 v5, v6, v14
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v7
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v10
+; GFX7-NEXT:    v_or_b32_e32 v4, v5, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v16, 24, v16
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v13
+; GFX7-NEXT:    v_cndmask_b32_e32 v5, v1, v2, vcc
+; GFX7-NEXT:    v_or_b32_e32 v4, v4, v16
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v5, v3, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v5, v5, v4, s[2:3]
+; GFX7-NEXT:    v_and_b32_e32 v5, s5, v5
+; GFX7-NEXT:    v_or_b32_e32 v0, v5, v0
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], s4, 0
+; GFX7-NEXT:    v_cndmask_b32_e64 v1, v1, v0, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e32 v2, v2, v0, vcc
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v3, v0, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v4, v0, s[2:3]
+; GFX7-NEXT:    v_lshrrev_b32_e32 v0, 8, v1
+; GFX7-NEXT:    v_and_b32_e32 v0, s6, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
+; GFX7-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v6
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 24, v2
+; GFX7-NEXT:    v_and_b32_e32 v1, s6, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v7
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v9
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 24, v3
+; GFX7-NEXT:    v_and_b32_e32 v2, s6, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v10
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v12
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 24, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 16, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 24, v4
+; GFX7-NEXT:    v_and_b32_e32 v3, s6, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v13
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v14
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, s6, v15
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i8>, <16 x i8> addrspace(1)* %ptr
+  %insert = insertelement <16 x i8> %vec, i8 %val, i32 %idx
+  store <16 x i8> %insert, <16 x i8> addrspace(1)* null
+  ret void
+}
+
+define amdgpu_ps void @insertelement_v_v16i8_v_v(<16 x i8> addrspace(1)* %ptr, i8 %val, i32 %idx) {
+; GFX9-LABEL: insertelement_v_v16i8_v_v:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    global_load_dwordx4 v[4:7], v[0:1], off
+; GFX9-NEXT:    s_mov_b32 s0, 8
+; GFX9-NEXT:    v_mov_b32_e32 v1, 8
+; GFX9-NEXT:    s_movk_i32 s1, 0xff
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 2, v3
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0xff
+; GFX9-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v8
+; GFX9-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v3, v3, v0
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v8
+; GFX9-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v8
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 8, v4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v10, 8, v5
+; GFX9-NEXT:    v_lshrrev_b32_e32 v11, 8, v6
+; GFX9-NEXT:    v_lshrrev_b32_e32 v12, 8, v7
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v9, s0, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v10, v1, v10 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v13, v4, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v14, v4, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v4, v4, s1, v9
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v12, v1, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v15, v5, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v16, v5, s1 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v5, v5, s1, v10
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v11, v1, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v17, v6, s1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v18, v6, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v6, v6, s1, v11
+; GFX9-NEXT:    v_and_b32_sdwa v19, v7, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v9, v7, v0, v12
+; GFX9-NEXT:    v_and_b32_sdwa v7, v7, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v4, v4, v13, v14
+; GFX9-NEXT:    v_or3_b32 v5, v5, v15, v16
+; GFX9-NEXT:    v_or3_b32 v7, v9, v19, v7
+; GFX9-NEXT:    v_or3_b32 v6, v6, v17, v18
+; GFX9-NEXT:    v_cndmask_b32_e32 v9, v4, v5, vcc
+; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v8
+; GFX9-NEXT:    v_cndmask_b32_e64 v9, v9, v6, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v9, v9, v7, s[2:3]
+; GFX9-NEXT:    v_and_or_b32 v2, v9, v3, v2
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v4, v2, s[4:5]
+; GFX9-NEXT:    v_cndmask_b32_e32 v4, v5, v2, vcc
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v6, v2, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v7, v2, s[2:3]
+; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 8, v4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v8, 8, v5
+; GFX9-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
+; GFX9-NEXT:    v_lshrrev_b32_e32 v9, 8, v2
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v7, v1, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v8, v1, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v6, v1, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_b32_sdwa v12, v4, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v13, v4, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v14, v5, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v15, v5, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b32_sdwa v1, v1, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX9-NEXT:    v_and_or_b32 v4, v4, v0, v7
+; GFX9-NEXT:    v_and_or_b32 v5, v5, v0, v8
+; GFX9-NEXT:    v_and_b32_sdwa v10, v3, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v11, v3, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_and_or_b32 v3, v3, v0, v6
+; GFX9-NEXT:    v_and_or_b32 v6, v2, v0, v1
+; GFX9-NEXT:    v_and_b32_sdwa v16, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v17, v2, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX9-NEXT:    v_or3_b32 v1, v4, v12, v13
+; GFX9-NEXT:    v_or3_b32 v2, v5, v14, v15
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    v_or3_b32 v0, v3, v10, v11
+; GFX9-NEXT:    v_or3_b32 v3, v6, v16, v17
+; GFX9-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9-NEXT:    global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-NEXT:    s_endpgm
+;
+; GFX8-LABEL: insertelement_v_v16i8_v_v:
+; GFX8:       ; %bb.0:
+; GFX8-NEXT:    flat_load_dwordx4 v[4:7], v[0:1]
+; GFX8-NEXT:    s_movk_i32 s0, 0xff
+; GFX8-NEXT:    v_mov_b32_e32 v1, 8
+; GFX8-NEXT:    v_mov_b32_e32 v8, 8
+; GFX8-NEXT:    v_mov_b32_e32 v9, s0
+; GFX8-NEXT:    v_mov_b32_e32 v0, 0xff
+; GFX8-NEXT:    v_lshrrev_b32_e32 v10, 2, v3
+; GFX8-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v10
+; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v10
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v3, v3, v0
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v10
+; GFX8-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX8-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v10
+; GFX8-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX8-NEXT:    v_lshrrev_b32_e32 v12, 8, v5
+; GFX8-NEXT:    v_lshrrev_b32_e32 v11, 8, v4
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v1, v1, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshrrev_b32_e32 v13, 8, v6
+; GFX8-NEXT:    v_lshrrev_b32_e32 v14, 8, v7
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v12, v8, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_and_b32_sdwa v15, v4, v9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v16, v4, v9 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v17, v5, v9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v4, v5, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v13, v8, v13 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v14, v8, v14 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v15
+; GFX8-NEXT:    v_and_b32_sdwa v9, v5, v9 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v17
+; GFX8-NEXT:    v_and_b32_sdwa v18, v6, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v5, v6, v13 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v19, v6, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v11, v7, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v6, v7, v14 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v18
+; GFX8-NEXT:    v_and_b32_sdwa v7, v7, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v6, v6, v11
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v16
+; GFX8-NEXT:    v_or_b32_e32 v4, v4, v9
+; GFX8-NEXT:    v_or_b32_e32 v6, v6, v7
+; GFX8-NEXT:    v_or_b32_e32 v5, v5, v19
+; GFX8-NEXT:    v_cndmask_b32_e32 v7, v1, v4, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v7, v5, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v7, v7, v6, s[2:3]
+; GFX8-NEXT:    v_and_b32_e32 v3, v7, v3
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX8-NEXT:    v_cndmask_b32_e32 v3, v4, v2, vcc
+; GFX8-NEXT:    v_cndmask_b32_e64 v4, v5, v2, s[0:1]
+; GFX8-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[4:5]
+; GFX8-NEXT:    v_cndmask_b32_e64 v2, v6, v2, s[2:3]
+; GFX8-NEXT:    v_lshrrev_b32_e32 v5, 8, v1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v6, 8, v3
+; GFX8-NEXT:    v_lshrrev_b32_e32 v7, 8, v4
+; GFX8-NEXT:    v_lshrrev_b32_e32 v9, 8, v2
+; GFX8-NEXT:    v_and_b32_sdwa v10, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v11, v1, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v12, v3, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v13, v3, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v14, v4, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v15, v4, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v16, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v17, v2, v0 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v0, v8, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v5, v8, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v6, v8, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_lshlrev_b32_sdwa v7, v8, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
+; GFX8-NEXT:    v_or_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v1, v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v3, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_sdwa v2, v2, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
+; GFX8-NEXT:    v_or_b32_e32 v3, v3, v14
+; GFX8-NEXT:    v_or_b32_e32 v4, v2, v16
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v10
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v12
+; GFX8-NEXT:    v_or_b32_e32 v2, v3, v15
+; GFX8-NEXT:    v_or_b32_e32 v3, v4, v17
+; GFX8-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8-NEXT:    v_or_b32_e32 v0, v0, v11
+; GFX8-NEXT:    v_or_b32_e32 v1, v1, v13
+; GFX8-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT:    s_endpgm
+;
+; GFX7-LABEL: insertelement_v_v16i8_v_v:
+; GFX7:       ; %bb.0:
+; GFX7-NEXT:    s_mov_b32 s10, 0
+; GFX7-NEXT:    s_mov_b32 s11, 0xf000
+; GFX7-NEXT:    s_mov_b64 s[8:9], 0
+; GFX7-NEXT:    buffer_load_dwordx4 v[4:7], v[0:1], s[8:11], 0 addr64
+; GFX7-NEXT:    s_movk_i32 s0, 0xff
+; GFX7-NEXT:    v_mov_b32_e32 v8, 0xff
+; GFX7-NEXT:    v_lshrrev_b32_e32 v19, 2, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, 3, v3
+; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v19
+; GFX7-NEXT:    v_and_b32_e32 v2, v2, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 3, v3
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, v3, v8
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[2:3], 3, v19
+; GFX7-NEXT:    v_xor_b32_e32 v3, -1, v3
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[4:5], 0, v19
+; GFX7-NEXT:    s_mov_b32 s10, -1
+; GFX7-NEXT:    s_waitcnt vmcnt(0)
+; GFX7-NEXT:    v_lshrrev_b32_e32 v0, 8, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 8, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v1, 16, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 16, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 8, v6
+; GFX7-NEXT:    v_and_b32_e32 v0, s0, v0
+; GFX7-NEXT:    v_and_b32_e32 v10, s0, v10
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 24, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 24, v5
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 16, v6
+; GFX7-NEXT:    v_lshrrev_b32_e32 v16, 8, v7
+; GFX7-NEXT:    v_and_b32_e32 v1, s0, v1
+; GFX7-NEXT:    v_and_b32_e32 v11, s0, v11
+; GFX7-NEXT:    v_and_b32_e32 v13, v13, v8
+; GFX7-NEXT:    v_and_b32_e32 v4, s0, v4
+; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v5, s0, v5
+; GFX7-NEXT:    v_lshlrev_b32_e32 v10, 8, v10
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 24, v6
+; GFX7-NEXT:    v_lshrrev_b32_e32 v17, 16, v7
+; GFX7-NEXT:    v_and_b32_e32 v9, s0, v9
+; GFX7-NEXT:    v_and_b32_e32 v12, s0, v12
+; GFX7-NEXT:    v_and_b32_e32 v14, v14, v8
+; GFX7-NEXT:    v_and_b32_e32 v16, v16, v8
+; GFX7-NEXT:    v_and_b32_e32 v6, s0, v6
+; GFX7-NEXT:    v_or_b32_e32 v0, v4, v0
+; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v11, 16, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v13, 8, v13
+; GFX7-NEXT:    v_or_b32_e32 v4, v5, v10
+; GFX7-NEXT:    v_lshrrev_b32_e32 v18, 24, v7
+; GFX7-NEXT:    v_and_b32_e32 v15, v15, v8
+; GFX7-NEXT:    v_and_b32_e32 v17, v17, v8
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v1
+; GFX7-NEXT:    v_lshlrev_b32_e32 v9, 24, v9
+; GFX7-NEXT:    v_and_b32_e32 v7, v7, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v12, 24, v12
+; GFX7-NEXT:    v_or_b32_e32 v1, v4, v11
+; GFX7-NEXT:    v_lshlrev_b32_e32 v14, 16, v14
+; GFX7-NEXT:    v_or_b32_e32 v5, v6, v13
+; GFX7-NEXT:    v_lshlrev_b32_e32 v16, 8, v16
+; GFX7-NEXT:    v_and_b32_e32 v18, v18, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v15, 24, v15
+; GFX7-NEXT:    v_or_b32_e32 v4, v5, v14
+; GFX7-NEXT:    v_lshlrev_b32_e32 v17, 16, v17
+; GFX7-NEXT:    v_or_b32_e32 v6, v7, v16
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v9
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v12
+; GFX7-NEXT:    v_or_b32_e32 v5, v6, v17
+; GFX7-NEXT:    v_lshlrev_b32_e32 v18, 24, v18
+; GFX7-NEXT:    v_or_b32_e32 v4, v4, v15
+; GFX7-NEXT:    v_cndmask_b32_e32 v6, v0, v1, vcc
+; GFX7-NEXT:    v_cmp_eq_u32_e64 s[0:1], 2, v19
+; GFX7-NEXT:    v_or_b32_e32 v5, v5, v18
+; GFX7-NEXT:    v_cndmask_b32_e64 v6, v6, v4, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v6, v6, v5, s[2:3]
+; GFX7-NEXT:    v_and_b32_e32 v3, v6, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v3, v2
+; GFX7-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[4:5]
+; GFX7-NEXT:    v_cndmask_b32_e64 v3, v4, v2, s[0:1]
+; GFX7-NEXT:    v_cndmask_b32_e64 v4, v5, v2, s[2:3]
+; GFX7-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
+; GFX7-NEXT:    v_and_b32_e32 v2, v2, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
+; GFX7-NEXT:    v_lshrrev_b32_e32 v6, 24, v0
+; GFX7-NEXT:    v_and_b32_e32 v0, v0, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, v5, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, v6, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v7, 8, v1
+; GFX7-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, v7, v8
+; GFX7-NEXT:    v_lshrrev_b32_e32 v9, 16, v1
+; GFX7-NEXT:    v_lshrrev_b32_e32 v10, 24, v1
+; GFX7-NEXT:    v_and_b32_e32 v1, v1, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, v9, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_and_b32_e32 v2, v10, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 24, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v11, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-NEXT:    v_lshrrev_b32_e32 v12, 16, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v13, 24, v3
+; GFX7-NEXT:    v_and_b32_e32 v2, v3, v8
+; GFX7-NEXT:    v_and_b32_e32 v3, v11, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, v12, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_and_b32_e32 v3, v13, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 24, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v14, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v2, v2, v3
+; GFX7-NEXT:    v_lshrrev_b32_e32 v15, 16, v4
+; GFX7-NEXT:    v_lshrrev_b32_e32 v16, 24, v4
+; GFX7-NEXT:    v_and_b32_e32 v3, v4, v8
+; GFX7-NEXT:    v_and_b32_e32 v4, v14, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, v15, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    v_and_b32_e32 v4, v16, v8
+; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 24, v4
+; GFX7-NEXT:    v_or_b32_e32 v3, v3, v4
+; GFX7-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX7-NEXT:    s_endpgm
+  %vec = load <16 x i8>, <16 x i8> addrspace(1)* %ptr
+  %insert = insertelement <16 x i8> %vec, i8 %val, i32 %idx
+  store <16 x i8> %insert, <16 x i8> addrspace(1)* null
+  ret void
+}