[AMDGPU] Add G16 support to image instructions

[Flakebi]

Summary

Add G16 feature for GFX10 and refactor lowerImage to be easier to read
and support 16 bit derivatives.

[Flakebi]

Forgot to run formatter

[arsenm]

Can you also add this to the GlobalISel legalization

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

5912–5936

ST->hasR128A16() && ST->hasGFX10A16(). We should probably add a hasA16 that checks both

5968

ST->hasG16()

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.encode.ll

3

Needs gfx9 checks too

[Flakebi]

Thank you for the fast review!

I added G16 to GlobalISel legalization. Is there a part where the image instructions get selected for GlobalISel? I didn’t find it and I guess it doesn’t pick up G16 instructions automatically.

gfx9 does not have G16 (addresses are 32 bit, derivatives 16 bit) only A16 (addresses and derivatives are 16 bit), so it’s tested in unsupported-image-g16.ll.

[arsenm]

In D76836#1944335, @Flakebi wrote:

Thank you for the fast review!

I added G16 to GlobalISel legalization. Is there a part where the image instructions get selected for GlobalISel? I didn’t find it and I guess it doesn’t pick up G16 instructions automatically.

gfx9 does not have G16 (addresses are 32 bit, derivatives 16 bit) only A16 (addresses and derivatives are 16 bit), so it’s tested in unsupported-image-g16.ll.

The GLobalISEl selection is stuck in review in D74316

[Flakebi]

What is the best way to handle A16 and G16 in instruction selection for GlobalISel?

They are only encoded in the types of arguments and not in the intrinsic id. In legalization, we can simply check the type of derivative and address arguments but afterwards, all arguments are packed into a singe, large value for the non-nsa case or several 32 bit values for nsa.
I think it’s hard to detect if this was originally A16 or even only G16 with 32 bit addresses.

Should I encode this information into metadata in the legalization step and use that in instruction selection?

I think this would look like I.getMetadata("amdgpu.a16") and I.getMetadata("amdgpu.g16").

[nhaehnle]

In D76836#1956855, @Flakebi wrote:

What is the best way to handle A16 and G16 in instruction selection for GlobalISel?

They are only encoded in the types of arguments and not in the intrinsic id. In legalization, we can simply check the type of derivative and address arguments but afterwards, all arguments are packed into a singe, large value for the non-nsa case or several 32 bit values for nsa.
I think it’s hard to detect if this was originally A16 or even only G16 with 32 bit addresses.

I agree that this is sub-optimal, though metadata isn't the right way to go here.

IIRC the current approach is to intuit what the original intent was based on NoReg left over by GlobalISel legalization and the like. However, now that we have G_AMDGPU_xxx MIR opcodes for image instructions after legalization, one option would be to add a magic operand to those machine instructions which encode the A16- and G16-ness.

[Flakebi]

As Nicolai suggested, I added an operand which encodes A16- and G16-ness.

A16 and G16 seem to be working now with SDag and GlobalISel.

[Flakebi]

Fixed UseNSA with A16 in GlobalISel, which decided based on the number of address components instead of dwords/registers.

[nhaehnle]

LGTM

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

5756

Maybe state that the comment uses "hi,lo" notation?

[Flakebi]

Add comment as Nicolai suggested

llvm/lib/Target/AMDGPU/AMDGPU.td

 >;
 
 def FeatureGFX10A16 : SubtargetFeature<"a16",
   "HasGFX10A16",
   "true",
   "Support gfx10-style A16 for 16-bit coordinates/gradients/lod/clamp/mip image operands"
 >;
 
+def FeatureG16 : SubtargetFeature<"g16",
+  "HasG16",
+  "true",
+  "Support G16 for 16-bit gradient image operands"
+>;
+
 def FeatureNSAEncoding : SubtargetFeature<"nsa-encoding",
   "HasNSAEncoding",
   "true",
   "Support NSA encoding for image instructions"
 >;
 
 def FeatureIntClamp : SubtargetFeature<"int-clamp-insts",
   "HasIntClamp",
def FeatureGFX10 : GCNSubtargetFeatureGeneration<"GFX10",
    FeatureApertureRegs, FeatureGFX9Insts, FeatureGFX10Insts, FeatureVOP3P,
    FeatureMovrel, FeatureFastFMAF32, FeatureDPP, FeatureIntClamp,
    FeatureSDWA, FeatureSDWAOmod, FeatureSDWAScalar, FeatureSDWASdst,
    FeatureFlatInstOffsets, FeatureFlatGlobalInsts, FeatureFlatScratchInsts,
    FeatureAddNoCarryInsts, FeatureFmaMixInsts, FeatureGFX8Insts,
    FeatureNoSdstCMPX, FeatureVscnt, FeatureRegisterBanking,
    FeatureVOP3Literal, FeatureDPP8,
    FeatureNoDataDepHazard, FeaturePkFmacF16Inst, FeatureDoesNotSupportSRAMECC,
-   FeatureGFX10A16, FeatureFastDenormalF32
+   FeatureGFX10A16, FeatureFastDenormalF32, FeatureG16
   ]
 >;
 
 class FeatureSet<list<SubtargetFeature> Features_> {
   list<SubtargetFeature> Features = Features_;
 }
 
 def FeatureISAVersion6_0_0 : FeatureSet<[FeatureSouthernIslands,
def HasDPP8 : Predicate<"Subtarget->hasDPP8()">,
   AssemblerPredicate<(all_of (not FeatureGCN3Encoding), FeatureGFX10Insts, FeatureDPP8)>;
 
 def HasR128A16 : Predicate<"Subtarget->hasR128A16()">,
   AssemblerPredicate<(all_of FeatureR128A16)>;
 
 def HasGFX10A16 : Predicate<"Subtarget->hasGFX10A16()">,
   AssemblerPredicate<(all_of FeatureGFX10A16)>;
 
+def HasG16 : Predicate<"Subtarget->hasG16()">,
+  AssemblerPredicate<(all_of FeatureG16)>;
+
 def HasDPP16 : Predicate<"Subtarget->hasDPP()">,
   AssemblerPredicate<(all_of (not FeatureGCN3Encoding), FeatureGFX10Insts, FeatureDPP)>;
 
 def HasIntClamp : Predicate<"Subtarget->hasIntClamp()">,
   AssemblerPredicate<(all_of FeatureIntClamp)>;
 
 def HasMadMixInsts : Predicate<"Subtarget->hasMadMixInsts()">,
   AssemblerPredicate<(all_of FeatureMadMixInsts)>;

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

bool AMDGPUInstructionSelector::selectImageIntrinsic(
   const AMDGPU::MIMGDimInfo *DimInfo = AMDGPU::getMIMGDimInfo(Intr->Dim);
   const AMDGPU::MIMGLZMappingInfo *LZMappingInfo =
       AMDGPU::getMIMGLZMappingInfo(Intr->BaseOpcode);
   const AMDGPU::MIMGMIPMappingInfo *MIPMappingInfo =
       AMDGPU::getMIMGMIPMappingInfo(Intr->BaseOpcode);
   unsigned IntrOpcode = Intr->BaseOpcode;
   const bool IsGFX10 = STI.getGeneration() >= AMDGPUSubtarget::GFX10;
 
-  const LLT S16 = LLT::scalar(16);
   const int VAddrIdx = getImageVAddrIdxBegin(BaseOpcode,
                                              MI.getNumExplicitDefs());
   int NumVAddr, NumGradients;
   std::tie(NumVAddr, NumGradients) = getImageNumVAddr(Intr, BaseOpcode);
 
-  const LLT AddrTy = MRI->getType(MI.getOperand(VAddrIdx).getReg());
-  const bool IsA16 = AddrTy.getScalarType() == S16;
-
   Register VDataIn, VDataOut;
   LLT VDataTy;
   int NumVDataDwords = -1;
   bool IsD16 = false;
 
   // XXX - Can we just get the second to last argument for ctrl?
   unsigned CtrlIdx; // Index of texfailctrl argument
   bool Unorm;
   if (!BaseOpcode->Sampler) {
     Unorm = true;
     CtrlIdx = VAddrIdx + NumVAddr + 1;
   } else {
     Unorm = MI.getOperand(VAddrIdx + NumVAddr + 2).getImm() != 0;
     CtrlIdx = VAddrIdx + NumVAddr + 3;
   }
 
   bool TFE;
   bool LWE;
   bool IsTexFail = false;
   if (!parseTexFail(MI.getOperand(CtrlIdx).getImm(), TFE, LWE, IsTexFail))
     return false;
 
+  const int Flags = MI.getOperand(CtrlIdx + 2).getImm();
+  const bool IsA16 = (Flags & 1) != 0;
+  const bool IsG16 = (Flags & 2) != 0;
+
+  // A16 implies 16 bit gradients
+  if (IsA16 && !IsG16)
+    return false;
+
   unsigned DMask = 0;
   unsigned DMaskLanes = 0;
 
   if (BaseOpcode->Atomic) {
     VDataOut = MI.getOperand(0).getReg();
     VDataIn = MI.getOperand(2).getReg();
     LLT Ty = MRI->getType(VDataIn);
 
bool AMDGPUInstructionSelector::selectImageIntrinsic(
   if (MIPMappingInfo) {
     const MachineOperand &Lod = MI.getOperand(VAddrIdx + NumVAddr - 1);
     if (Lod.isImm()) {
       assert(Lod.getImm() == 0);
       IntrOpcode = MIPMappingInfo->NONMIP;  // set new opcode to variant without _mip
     }
   }
 
+  // Set G16 opcode
+  if (IsG16 && !IsA16) {
+    const AMDGPU::MIMGG16MappingInfo *G16MappingInfo =
+        AMDGPU::getMIMGG16MappingInfo(Intr->BaseOpcode);
+    assert(G16MappingInfo);
+    IntrOpcode = G16MappingInfo->G16; // set opcode to variant with _g16
+  }
+
   // TODO: Check this in verifier.
   assert((!IsTexFail || DMaskLanes >= 1) && "should have legalized this");
 
   bool GLC = false;
   bool SLC = false;
   bool DLC = false;
   if (BaseOpcode->Atomic) {
     GLC = true; // TODO no-return optimization

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

bool AMDGPULegalizerInfo::legalizeBufferAtomic(MachineInstr &MI,
   MI.eraseFromParent();
   return true;
 }
 
 /// Turn a set of s16 typed registers in \p A16AddrRegs into a dword sized
 /// vector with s16 typed elements.
 static void packImageA16AddressToDwords(MachineIRBuilder &B, MachineInstr &MI,
                                         SmallVectorImpl<Register> &PackedAddrs,
-                                        int AddrIdx, int DimIdx, int NumVAddrs,
+                                        int AddrIdx, int DimIdx, int EndIdx,
                                         int NumGradients) {
   const LLT S16 = LLT::scalar(16);
   const LLT V2S16 = LLT::vector(2, 16);
 
-  for (int I = AddrIdx; I < AddrIdx + NumVAddrs; ++I) {
+  for (int I = AddrIdx; I < EndIdx; ++I) {
     MachineOperand &SrcOp = MI.getOperand(I);
     if (!SrcOp.isReg())
       continue; // _L to _LZ may have eliminated this.
 
     Register AddrReg = SrcOp.getReg();
 
     if (I < DimIdx) {
       AddrReg = B.buildBitcast(V2S16, AddrReg).getReg(0);
       PackedAddrs.push_back(AddrReg);
     } else {
       // Dz/dh, dz/dv and the last odd coord are packed with undef. Also, in 1D,
       // derivatives dx/dh and dx/dv are packed with undef.
-      if (((I + 1) >= (AddrIdx + NumVAddrs)) ||
+      if (((I + 1) >= EndIdx) ||
           ((NumGradients / 2) % 2 == 1 &&
            (I == DimIdx + (NumGradients / 2) - 1 ||
             I == DimIdx + NumGradients - 1)) ||
           // Check for _L to _LZ optimization
           !MI.getOperand(I + 1).isReg()) {
         PackedAddrs.push_back(
             B.buildBuildVector(V2S16, {AddrReg, B.buildUndef(S16).getReg(0)})
                 .getReg(0));
bool AMDGPULegalizerInfo::legalizeImageIntrinsic(
   MachineRegisterInfo *MRI = B.getMRI();
   const LLT S32 = LLT::scalar(32);
   const LLT S16 = LLT::scalar(16);
   const LLT V2S16 = LLT::vector(2, 16);
 
   // Index of first address argument
   const int AddrIdx = getImageVAddrIdxBegin(BaseOpcode, NumDefs);
 
-  // Check for 16 bit addresses and pack if true.
-  int DimIdx = AddrIdx + BaseOpcode->NumExtraArgs;
-  LLT AddrTy = MRI->getType(MI.getOperand(DimIdx).getReg());
-  const bool IsA16 = AddrTy == S16;
-
   int NumVAddrs, NumGradients;
   std::tie(NumVAddrs, NumGradients) = getImageNumVAddr(ImageDimIntr, BaseOpcode);
   const int DMaskIdx = BaseOpcode->Atomic ? -1 :
     getDMaskIdx(BaseOpcode, NumDefs);
   unsigned DMask = 0;
 
+  // Check for 16 bit addresses and pack if true.
+  int DimIdx = AddrIdx + BaseOpcode->NumExtraArgs;
+  LLT GradTy = MRI->getType(MI.getOperand(DimIdx).getReg());
+  LLT AddrTy = MRI->getType(MI.getOperand(DimIdx + NumGradients).getReg());
+  const bool IsG16 = GradTy == S16;
+  const bool IsA16 = AddrTy == S16;
+
   int DMaskLanes = 0;
   if (!BaseOpcode->Atomic) {
     DMask = MI.getOperand(DMaskIdx).getImm();
     if (BaseOpcode->Gather4) {
       DMaskLanes = 4;
     } else if (DMask != 0) {
       DMaskLanes = countPopulation(DMask);
     } else if (!IsTFE && !BaseOpcode->Store) {
if (mi_match(MI.getOperand(LodIdx).getReg(), *MRI, m_ICst(ConstantLod))) {
         // TODO: Change intrinsic opcode and remove operand instead or replacing
         // it with 0, as the _L to _LZ handling is done above.
         MI.getOperand(LodIdx).ChangeToImmediate(0);
         --CorrectedNumVAddrs;
       }
     }
   }
 
-  // If the register allocator cannot place the address registers contiguously
-  // without introducing moves, then using the non-sequential address encoding
-  // is always preferable, since it saves VALU instructions and is usually a
-  // wash in terms of code size or even better.
-  //
-  // However, we currently have no way of hinting to the register allocator
-  // that MIMG addresses should be placed contiguously when it is possible to
-  // do so, so force non-NSA for the common 2-address case as a heuristic.
-  //
-  // SIShrinkInstructions will convert NSA encodings to non-NSA after register
-  // allocation when possible.
-  const bool UseNSA = CorrectedNumVAddrs >= 3 && ST.hasNSAEncoding();
-
   // Rewrite the addressing register layout before doing anything else.
+  if (IsA16 || IsG16) {
     if (IsA16) {
-    // FIXME: this feature is missing from gfx10. When that is fixed, this check
-    // should be introduced.
-    if (!ST.hasR128A16() && !ST.hasGFX10A16())
+      // Target must support the feature and gradients need to be 16 bit too
+      if (!ST.hasA16() || !IsG16)
+        return false;
+    } else if (!ST.hasG16())
       return false;
 
     if (NumVAddrs > 1) {
       SmallVector<Register, 4> PackedRegs;
-      packImageA16AddressToDwords(B, MI, PackedRegs, AddrIdx, DimIdx, NumVAddrs,
-                                  NumGradients);
+      // Don't compress addresses for G16
+      const int PackEndIdx =
+          IsA16 ? (AddrIdx + NumVAddrs) : (DimIdx + NumGradients);
+      packImageA16AddressToDwords(B, MI, PackedRegs, AddrIdx, DimIdx,
+                                  PackEndIdx, NumGradients);
+
+      if (!IsA16) {
+        // Add uncompressed address
+        for (int I = DimIdx + NumGradients; I != AddrIdx + NumVAddrs; ++I) {
+          int AddrReg = MI.getOperand(I).getReg();
+          assert(B.getMRI()->getType(AddrReg) == LLT::scalar(32));
+          PackedRegs.push_back(AddrReg);
+        }
+      }
+
+      // See also below in the non-a16 branch
+      const bool UseNSA = PackedRegs.size() >= 3 && ST.hasNSAEncoding();
 
       if (!UseNSA && PackedRegs.size() > 1) {
         LLT PackedAddrTy = LLT::vector(2 * PackedRegs.size(), 16);
         auto Concat = B.buildConcatVectors(PackedAddrTy, PackedRegs);
         PackedRegs[0] = Concat.getReg(0);
         PackedRegs.resize(1);
       }
 
       const int NumPacked = PackedRegs.size();
       for (int I = 0; I != NumVAddrs; ++I) {
         MachineOperand &SrcOp = MI.getOperand(AddrIdx + I);
         if (!SrcOp.isReg()) {
           assert(SrcOp.isImm() && SrcOp.getImm() == 0);
           continue;
         }
 
         assert(SrcOp.getReg() != AMDGPU::NoRegister);
 
         if (I < NumPacked)
           SrcOp.setReg(PackedRegs[I]);
         else
           SrcOp.setReg(AMDGPU::NoRegister);
       }
     }
-  } else if (!UseNSA && NumVAddrs > 1) {
+  } else {
+    // If the register allocator cannot place the address registers contiguously
+    // without introducing moves, then using the non-sequential address encoding
+    // is always preferable, since it saves VALU instructions and is usually a
+    // wash in terms of code size or even better.
+    //
+    // However, we currently have no way of hinting to the register allocator
+    // that MIMG addresses should be placed contiguously when it is possible to
+    // do so, so force non-NSA for the common 2-address case as a heuristic.
+    //
+    // SIShrinkInstructions will convert NSA encodings to non-NSA after register
+    // allocation when possible.
+    const bool UseNSA = CorrectedNumVAddrs >= 3 && ST.hasNSAEncoding();
+
+    if (!UseNSA && NumVAddrs > 1)
       convertImageAddrToPacked(B, MI, AddrIdx, NumVAddrs);
   }
 
+  int Flags = 0;
+  if (IsA16)
+    Flags |= 1;
+  if (IsG16)
+    Flags |= 2;
+  MI.addOperand(MachineOperand::CreateImm(Flags));
 
   if (BaseOpcode->Store) { // No TFE for stores?
     // TODO: Handle dmask trim
     Register VData = MI.getOperand(1).getReg();
     LLT Ty = MRI->getType(VData);
     if (!Ty.isVector() || Ty.getElementType() != S16)
       return true;
 

llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h

protected:
   bool HasSDWAScalar;
   bool HasSDWASdst;
   bool HasSDWAMac;
   bool HasSDWAOutModsVOPC;
   bool HasDPP;
   bool HasDPP8;
   bool HasR128A16;
   bool HasGFX10A16;
+  bool HasG16;
   bool HasNSAEncoding;
   bool HasDLInsts;
   bool HasDot1Insts;
   bool HasDot2Insts;
   bool HasDot3Insts;
   bool HasDot4Insts;
   bool HasDot5Insts;
   bool HasDot6Insts;
public:
   bool hasR128A16() const {
     return HasR128A16;
   }
 
   bool hasGFX10A16() const {
     return HasGFX10A16;
   }
 
+  bool hasA16() const { return hasR128A16() || hasGFX10A16(); }
+
+  bool hasG16() const { return HasG16; }
+
   bool hasOffset3fBug() const {
     return HasOffset3fBug;
   }
 
   bool hasNSAEncoding() const {
     return HasNSAEncoding;
   }
 

llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp

GCNSubtarget::GCNSubtarget(const Triple &TT, StringRef GPU, StringRef FS,
     HasSDWAScalar(false),
     HasSDWASdst(false),
     HasSDWAMac(false),
     HasSDWAOutModsVOPC(false),
     HasDPP(false),
     HasDPP8(false),
     HasR128A16(false),
     HasGFX10A16(false),
+    HasG16(false),
     HasNSAEncoding(false),
     HasDLInsts(false),
     HasDot1Insts(false),
     HasDot2Insts(false),
     HasDot3Insts(false),
     HasDot4Insts(false),
     HasDot5Insts(false),
     HasDot6Insts(false),

llvm/lib/Target/AMDGPU/MIMGInstructions.td

class MIMGBaseOpcode : PredicateControl {
   MIMGBaseOpcode BaseOpcode = !cast<MIMGBaseOpcode>(NAME);
   bit Store = 0;
   bit Atomic = 0;
   bit AtomicX2 = 0; // (f)cmpswap
   bit Sampler = 0;
   bit Gather4 = 0;
   bits<8> NumExtraArgs = 0;
   bit Gradients = 0;
+  bit G16 = 0;
   bit Coordinates = 1;
   bit LodOrClampOrMip = 0;
   bit HasD16 = 0;
 }
 
 def MIMGBaseOpcode : GenericEnum {
   let FilterClass = "MIMGBaseOpcode";
 }
 
 def MIMGBaseOpcodesTable : GenericTable {
   let FilterClass = "MIMGBaseOpcode";
   let CppTypeName = "MIMGBaseOpcodeInfo";
-  let Fields = ["BaseOpcode", "Store", "Atomic", "AtomicX2", "Sampler", "Gather4",
-                "NumExtraArgs", "Gradients", "Coordinates", "LodOrClampOrMip",
-                "HasD16"];
+  let Fields = ["BaseOpcode", "Store", "Atomic", "AtomicX2", "Sampler",
+                "Gather4", "NumExtraArgs", "Gradients", "G16", "Coordinates",
+                "LodOrClampOrMip", "HasD16"];
   GenericEnum TypeOf_BaseOpcode = MIMGBaseOpcode;
 
   let PrimaryKey = ["BaseOpcode"];
   let PrimaryKeyName = "getMIMGBaseOpcodeInfo";
 }
 
 def MIMGDim : GenericEnum {
   let FilterClass = "AMDGPUDimProps";
def MIMGMIPMappingTable : GenericTable {
   let Fields = ["MIP", "NONMIP"];
   GenericEnum TypeOf_MIP = MIMGBaseOpcode;
   GenericEnum TypeOf_NONMIP = MIMGBaseOpcode;
 
   let PrimaryKey = ["MIP"];
   let PrimaryKeyName = "getMIMGMIPMappingInfo";
 }
 
+class MIMGG16Mapping<MIMGBaseOpcode g, MIMGBaseOpcode g16> {
+  MIMGBaseOpcode G = g;
+  MIMGBaseOpcode G16 = g16;
+}
+
+def MIMGG16MappingTable : GenericTable {
+  let FilterClass = "MIMGG16Mapping";
+  let CppTypeName = "MIMGG16MappingInfo";
+  let Fields = ["G", "G16"];
+  GenericEnum TypeOf_G = MIMGBaseOpcode;
+  GenericEnum TypeOf_G16 = MIMGBaseOpcode;
+
+  let PrimaryKey = ["G"];
+  let PrimaryKeyName = "getMIMGG16MappingInfo";
+}
+
 class MIMG_Base <dag outs, string dns = "">
   : InstSI <outs, (ins), "", []> {
 
   let VM_CNT = 1;
   let EXP_CNT = 1;
   let MIMG = 1;
   let Uses = [EXEC];
   let mayLoad = 1;
class MIMG_Sampler_BaseOpcode<AMDGPUSampleVariant sample>
   : MIMGBaseOpcode {
   let Sampler = 1;
   let NumExtraArgs = !size(sample.ExtraAddrArgs);
   let Gradients = sample.Gradients;
   let LodOrClampOrMip = !ne(sample.LodOrClamp, "");
 }
 
 multiclass MIMG_Sampler <bits<8> op, AMDGPUSampleVariant sample, bit wqm = 0,
-                         bit isGetLod = 0,
-                         string asm = "image_sample"#sample.LowerCaseMod> {
+                         bit isG16 = 0, bit isGetLod = 0,
+                         string asm = "image_sample"#sample.LowerCaseMod#!if(isG16, "_g16", "")> {
   def "" : MIMG_Sampler_BaseOpcode<sample> {
     let HasD16 = !if(isGetLod, 0, 1);
+    let G16 = isG16;
   }
 
   let BaseOpcode = !cast<MIMGBaseOpcode>(NAME), WQM = wqm,
       mayLoad = !if(isGetLod, 0, 1) in {
     let VDataDwords = 1 in
     defm _V1 : MIMG_Sampler_Src_Helper<op, asm, sample, VGPR_32, 1>;
     let VDataDwords = 2 in
     defm _V2 : MIMG_Sampler_Src_Helper<op, asm, sample, VReg_64>;

 defm IMAGE_ATOMIC_XOR : MIMG_Atomic <mimg<0x1a>, "image_atomic_xor">;
 defm IMAGE_ATOMIC_INC : MIMG_Atomic <mimg<0x1b>, "image_atomic_inc">;
 defm IMAGE_ATOMIC_DEC : MIMG_Atomic <mimg<0x1c>, "image_atomic_dec">;
 //let FPAtomic = 1 in {
 //def IMAGE_ATOMIC_FCMPSWAP : MIMG_NoPattern_ <"image_atomic_fcmpswap", 0x0000001d, 1>; -- not on VI
 //def IMAGE_ATOMIC_FMIN : MIMG_NoPattern_ <"image_atomic_fmin", 0x0000001e>; -- not on VI
 //def IMAGE_ATOMIC_FMAX : MIMG_NoPattern_ <"image_atomic_fmax", 0x0000001f>; -- not on VI
 //} // End let FPAtomic = 1
-defm IMAGE_SAMPLE           : MIMG_Sampler_WQM <0x00000020, AMDGPUSample>;
-defm IMAGE_SAMPLE_CL        : MIMG_Sampler_WQM <0x00000021, AMDGPUSample_cl>;
-defm IMAGE_SAMPLE_D         : MIMG_Sampler <0x00000022, AMDGPUSample_d>;
-defm IMAGE_SAMPLE_D_CL      : MIMG_Sampler <0x00000023, AMDGPUSample_d_cl>;
-defm IMAGE_SAMPLE_L         : MIMG_Sampler <0x00000024, AMDGPUSample_l>;
-defm IMAGE_SAMPLE_B         : MIMG_Sampler_WQM <0x00000025, AMDGPUSample_b>;
-defm IMAGE_SAMPLE_B_CL      : MIMG_Sampler_WQM <0x00000026, AMDGPUSample_b_cl>;
-defm IMAGE_SAMPLE_LZ        : MIMG_Sampler <0x00000027, AMDGPUSample_lz>;
-defm IMAGE_SAMPLE_C         : MIMG_Sampler_WQM <0x00000028, AMDGPUSample_c>;
-defm IMAGE_SAMPLE_C_CL      : MIMG_Sampler_WQM <0x00000029, AMDGPUSample_c_cl>;
-defm IMAGE_SAMPLE_C_D       : MIMG_Sampler <0x0000002a, AMDGPUSample_c_d>;
-defm IMAGE_SAMPLE_C_D_CL    : MIMG_Sampler <0x0000002b, AMDGPUSample_c_d_cl>;
-defm IMAGE_SAMPLE_C_L       : MIMG_Sampler <0x0000002c, AMDGPUSample_c_l>;
-defm IMAGE_SAMPLE_C_B       : MIMG_Sampler_WQM <0x0000002d, AMDGPUSample_c_b>;
-defm IMAGE_SAMPLE_C_B_CL    : MIMG_Sampler_WQM <0x0000002e, AMDGPUSample_c_b_cl>;
-defm IMAGE_SAMPLE_C_LZ      : MIMG_Sampler <0x0000002f, AMDGPUSample_c_lz>;
-defm IMAGE_SAMPLE_O         : MIMG_Sampler_WQM <0x00000030, AMDGPUSample_o>;
-defm IMAGE_SAMPLE_CL_O      : MIMG_Sampler_WQM <0x00000031, AMDGPUSample_cl_o>;
-defm IMAGE_SAMPLE_D_O       : MIMG_Sampler <0x00000032, AMDGPUSample_d_o>;
-defm IMAGE_SAMPLE_D_CL_O    : MIMG_Sampler <0x00000033, AMDGPUSample_d_cl_o>;
-defm IMAGE_SAMPLE_L_O       : MIMG_Sampler <0x00000034, AMDGPUSample_l_o>;
-defm IMAGE_SAMPLE_B_O       : MIMG_Sampler_WQM <0x00000035, AMDGPUSample_b_o>;
-defm IMAGE_SAMPLE_B_CL_O    : MIMG_Sampler_WQM <0x00000036, AMDGPUSample_b_cl_o>;
-defm IMAGE_SAMPLE_LZ_O      : MIMG_Sampler <0x00000037, AMDGPUSample_lz_o>;
-defm IMAGE_SAMPLE_C_O       : MIMG_Sampler_WQM <0x00000038, AMDGPUSample_c_o>;
-defm IMAGE_SAMPLE_C_CL_O    : MIMG_Sampler_WQM <0x00000039, AMDGPUSample_c_cl_o>;
-defm IMAGE_SAMPLE_C_D_O     : MIMG_Sampler <0x0000003a, AMDGPUSample_c_d_o>;
-defm IMAGE_SAMPLE_C_D_CL_O  : MIMG_Sampler <0x0000003b, AMDGPUSample_c_d_cl_o>;
-defm IMAGE_SAMPLE_C_L_O     : MIMG_Sampler <0x0000003c, AMDGPUSample_c_l_o>;
-defm IMAGE_SAMPLE_C_B_CL_O  : MIMG_Sampler_WQM <0x0000003e, AMDGPUSample_c_b_cl_o>;
-defm IMAGE_SAMPLE_C_B_O     : MIMG_Sampler_WQM <0x0000003d, AMDGPUSample_c_b_o>;
-defm IMAGE_SAMPLE_C_LZ_O    : MIMG_Sampler <0x0000003f, AMDGPUSample_c_lz_o>;
-defm IMAGE_GATHER4          : MIMG_Gather_WQM <0x00000040, AMDGPUSample>;
-defm IMAGE_GATHER4_CL       : MIMG_Gather_WQM <0x00000041, AMDGPUSample_cl>;
-defm IMAGE_GATHER4_L        : MIMG_Gather <0x00000044, AMDGPUSample_l>;
-defm IMAGE_GATHER4_B        : MIMG_Gather_WQM <0x00000045, AMDGPUSample_b>;
-defm IMAGE_GATHER4_B_CL     : MIMG_Gather_WQM <0x00000046, AMDGPUSample_b_cl>;
-defm IMAGE_GATHER4_LZ       : MIMG_Gather <0x00000047, AMDGPUSample_lz>;
-defm IMAGE_GATHER4_C        : MIMG_Gather_WQM <0x00000048, AMDGPUSample_c>;
-defm IMAGE_GATHER4_C_CL     : MIMG_Gather_WQM <0x00000049, AMDGPUSample_c_cl>;
-defm IMAGE_GATHER4_C_L      : MIMG_Gather <0x0000004c, AMDGPUSample_c_l>;
-defm IMAGE_GATHER4_C_B      : MIMG_Gather_WQM <0x0000004d, AMDGPUSample_c_b>;
-defm IMAGE_GATHER4_C_B_CL   : MIMG_Gather_WQM <0x0000004e, AMDGPUSample_c_b_cl>;
-defm IMAGE_GATHER4_C_LZ     : MIMG_Gather <0x0000004f, AMDGPUSample_c_lz>;
-defm IMAGE_GATHER4_O        : MIMG_Gather_WQM <0x00000050, AMDGPUSample_o>;
-defm IMAGE_GATHER4_CL_O     : MIMG_Gather_WQM <0x00000051, AMDGPUSample_cl_o>;
-defm IMAGE_GATHER4_L_O      : MIMG_Gather <0x00000054, AMDGPUSample_l_o>;
-defm IMAGE_GATHER4_B_O      : MIMG_Gather_WQM <0x00000055, AMDGPUSample_b_o>;
-defm IMAGE_GATHER4_B_CL_O   : MIMG_Gather <0x00000056, AMDGPUSample_b_cl_o>;
-defm IMAGE_GATHER4_LZ_O     : MIMG_Gather <0x00000057, AMDGPUSample_lz_o>;
-defm IMAGE_GATHER4_C_O      : MIMG_Gather_WQM <0x00000058, AMDGPUSample_c_o>;
-defm IMAGE_GATHER4_C_CL_O   : MIMG_Gather_WQM <0x00000059, AMDGPUSample_c_cl_o>;
-defm IMAGE_GATHER4_C_L_O    : MIMG_Gather <0x0000005c, AMDGPUSample_c_l_o>;
-defm IMAGE_GATHER4_C_B_O    : MIMG_Gather_WQM <0x0000005d, AMDGPUSample_c_b_o>;
-defm IMAGE_GATHER4_C_B_CL_O : MIMG_Gather_WQM <0x0000005e, AMDGPUSample_c_b_cl_o>;
-defm IMAGE_GATHER4_C_LZ_O   : MIMG_Gather <0x0000005f, AMDGPUSample_c_lz_o>;
+defm IMAGE_SAMPLE               : MIMG_Sampler_WQM <0x00000020, AMDGPUSample>;
+defm IMAGE_SAMPLE_CL            : MIMG_Sampler_WQM <0x00000021, AMDGPUSample_cl>;
+defm IMAGE_SAMPLE_D             : MIMG_Sampler <0x00000022, AMDGPUSample_d>;
+defm IMAGE_SAMPLE_D_CL          : MIMG_Sampler <0x00000023, AMDGPUSample_d_cl>;
+defm IMAGE_SAMPLE_D_G16         : MIMG_Sampler <0x000000a2, AMDGPUSample_d, 0, 1>;
+defm IMAGE_SAMPLE_D_CL_G16      : MIMG_Sampler <0x000000a3, AMDGPUSample_d_cl, 0, 1>;
+defm IMAGE_SAMPLE_L             : MIMG_Sampler <0x00000024, AMDGPUSample_l>;
+defm IMAGE_SAMPLE_B             : MIMG_Sampler_WQM <0x00000025, AMDGPUSample_b>;
+defm IMAGE_SAMPLE_B_CL          : MIMG_Sampler_WQM <0x00000026, AMDGPUSample_b_cl>;
+defm IMAGE_SAMPLE_LZ            : MIMG_Sampler <0x00000027, AMDGPUSample_lz>;
+defm IMAGE_SAMPLE_C             : MIMG_Sampler_WQM <0x00000028, AMDGPUSample_c>;
+defm IMAGE_SAMPLE_C_CL          : MIMG_Sampler_WQM <0x00000029, AMDGPUSample_c_cl>;
+defm IMAGE_SAMPLE_C_D           : MIMG_Sampler <0x0000002a, AMDGPUSample_c_d>;
+defm IMAGE_SAMPLE_C_D_CL        : MIMG_Sampler <0x0000002b, AMDGPUSample_c_d_cl>;
+defm IMAGE_SAMPLE_C_D_G16       : MIMG_Sampler <0x000000aa, AMDGPUSample_c_d, 0, 1>;
+defm IMAGE_SAMPLE_C_D_CL_G16    : MIMG_Sampler <0x000000ab, AMDGPUSample_c_d_cl, 0, 1>;
+defm IMAGE_SAMPLE_C_L           : MIMG_Sampler <0x0000002c, AMDGPUSample_c_l>;
+defm IMAGE_SAMPLE_C_B           : MIMG_Sampler_WQM <0x0000002d, AMDGPUSample_c_b>;
+defm IMAGE_SAMPLE_C_B_CL        : MIMG_Sampler_WQM <0x0000002e, AMDGPUSample_c_b_cl>;
+defm IMAGE_SAMPLE_C_LZ          : MIMG_Sampler <0x0000002f, AMDGPUSample_c_lz>;
+defm IMAGE_SAMPLE_O             : MIMG_Sampler_WQM <0x00000030, AMDGPUSample_o>;
+defm IMAGE_SAMPLE_CL_O          : MIMG_Sampler_WQM <0x00000031, AMDGPUSample_cl_o>;
+defm IMAGE_SAMPLE_D_O           : MIMG_Sampler <0x00000032, AMDGPUSample_d_o>;
+defm IMAGE_SAMPLE_D_CL_O        : MIMG_Sampler <0x00000033, AMDGPUSample_d_cl_o>;
+defm IMAGE_SAMPLE_D_O_G16       : MIMG_Sampler <0x000000b2, AMDGPUSample_d_o, 0, 1>;
+defm IMAGE_SAMPLE_D_CL_O_G16    : MIMG_Sampler <0x000000b3, AMDGPUSample_d_cl_o, 0, 1>;
+defm IMAGE_SAMPLE_L_O           : MIMG_Sampler <0x00000034, AMDGPUSample_l_o>;
+defm IMAGE_SAMPLE_B_O           : MIMG_Sampler_WQM <0x00000035, AMDGPUSample_b_o>;
+defm IMAGE_SAMPLE_B_CL_O        : MIMG_Sampler_WQM <0x00000036, AMDGPUSample_b_cl_o>;
+defm IMAGE_SAMPLE_LZ_O          : MIMG_Sampler <0x00000037, AMDGPUSample_lz_o>;
+defm IMAGE_SAMPLE_C_O           : MIMG_Sampler_WQM <0x00000038, AMDGPUSample_c_o>;
+defm IMAGE_SAMPLE_C_CL_O        : MIMG_Sampler_WQM <0x00000039, AMDGPUSample_c_cl_o>;
+defm IMAGE_SAMPLE_C_D_O         : MIMG_Sampler <0x0000003a, AMDGPUSample_c_d_o>;
+defm IMAGE_SAMPLE_C_D_CL_O      : MIMG_Sampler <0x0000003b, AMDGPUSample_c_d_cl_o>;
+defm IMAGE_SAMPLE_C_D_O_G16     : MIMG_Sampler <0x000000ba, AMDGPUSample_c_d_o, 0, 1>;
+defm IMAGE_SAMPLE_C_D_CL_O_G16  : MIMG_Sampler <0x000000bb, AMDGPUSample_c_d_cl_o, 0, 1>;
+defm IMAGE_SAMPLE_C_L_O         : MIMG_Sampler <0x0000003c, AMDGPUSample_c_l_o>;
+defm IMAGE_SAMPLE_C_B_CL_O      : MIMG_Sampler_WQM <0x0000003e, AMDGPUSample_c_b_cl_o>;
+defm IMAGE_SAMPLE_C_B_O         : MIMG_Sampler_WQM <0x0000003d, AMDGPUSample_c_b_o>;
+defm IMAGE_SAMPLE_C_LZ_O        : MIMG_Sampler <0x0000003f, AMDGPUSample_c_lz_o>;
+defm IMAGE_GATHER4              : MIMG_Gather_WQM <0x00000040, AMDGPUSample>;
+defm IMAGE_GATHER4_CL           : MIMG_Gather_WQM <0x00000041, AMDGPUSample_cl>;
+defm IMAGE_GATHER4_L            : MIMG_Gather <0x00000044, AMDGPUSample_l>;
+defm IMAGE_GATHER4_B            : MIMG_Gather_WQM <0x00000045, AMDGPUSample_b>;
+defm IMAGE_GATHER4_B_CL         : MIMG_Gather_WQM <0x00000046, AMDGPUSample_b_cl>;
+defm IMAGE_GATHER4_LZ           : MIMG_Gather <0x00000047, AMDGPUSample_lz>;
+defm IMAGE_GATHER4_C            : MIMG_Gather_WQM <0x00000048, AMDGPUSample_c>;
+defm IMAGE_GATHER4_C_CL         : MIMG_Gather_WQM <0x00000049, AMDGPUSample_c_cl>;
+defm IMAGE_GATHER4_C_L          : MIMG_Gather <0x0000004c, AMDGPUSample_c_l>;
+defm IMAGE_GATHER4_C_B          : MIMG_Gather_WQM <0x0000004d, AMDGPUSample_c_b>;
+defm IMAGE_GATHER4_C_B_CL       : MIMG_Gather_WQM <0x0000004e, AMDGPUSample_c_b_cl>;
+defm IMAGE_GATHER4_C_LZ         : MIMG_Gather <0x0000004f, AMDGPUSample_c_lz>;
+defm IMAGE_GATHER4_O            : MIMG_Gather_WQM <0x00000050, AMDGPUSample_o>;
+defm IMAGE_GATHER4_CL_O         : MIMG_Gather_WQM <0x00000051, AMDGPUSample_cl_o>;
+defm IMAGE_GATHER4_L_O          : MIMG_Gather <0x00000054, AMDGPUSample_l_o>;
+defm IMAGE_GATHER4_B_O          : MIMG_Gather_WQM <0x00000055, AMDGPUSample_b_o>;
+defm IMAGE_GATHER4_B_CL_O       : MIMG_Gather <0x00000056, AMDGPUSample_b_cl_o>;
+defm IMAGE_GATHER4_LZ_O         : MIMG_Gather <0x00000057, AMDGPUSample_lz_o>;
+defm IMAGE_GATHER4_C_O          : MIMG_Gather_WQM <0x00000058, AMDGPUSample_c_o>;
+defm IMAGE_GATHER4_C_CL_O       : MIMG_Gather_WQM <0x00000059, AMDGPUSample_c_cl_o>;
+defm IMAGE_GATHER4_C_L_O        : MIMG_Gather <0x0000005c, AMDGPUSample_c_l_o>;
+defm IMAGE_GATHER4_C_B_O        : MIMG_Gather_WQM <0x0000005d, AMDGPUSample_c_b_o>;
+defm IMAGE_GATHER4_C_B_CL_O     : MIMG_Gather_WQM <0x0000005e, AMDGPUSample_c_b_cl_o>;
+defm IMAGE_GATHER4_C_LZ_O       : MIMG_Gather <0x0000005f, AMDGPUSample_c_lz_o>;
 
-defm IMAGE_GET_LOD          : MIMG_Sampler <0x00000060, AMDGPUSample, 1, 1, "image_get_lod">;
+defm IMAGE_GET_LOD              : MIMG_Sampler <0x00000060, AMDGPUSample, 1, 0, 1, "image_get_lod">;
 
-defm IMAGE_SAMPLE_CD        : MIMG_Sampler <0x00000068, AMDGPUSample_cd>;
-defm IMAGE_SAMPLE_CD_CL     : MIMG_Sampler <0x00000069, AMDGPUSample_cd_cl>;
-defm IMAGE_SAMPLE_C_CD      : MIMG_Sampler <0x0000006a, AMDGPUSample_c_cd>;
-defm IMAGE_SAMPLE_C_CD_CL   : MIMG_Sampler <0x0000006b, AMDGPUSample_c_cd_cl>;
-defm IMAGE_SAMPLE_CD_O      : MIMG_Sampler <0x0000006c, AMDGPUSample_cd_o>;
-defm IMAGE_SAMPLE_CD_CL_O   : MIMG_Sampler <0x0000006d, AMDGPUSample_cd_cl_o>;
-defm IMAGE_SAMPLE_C_CD_O    : MIMG_Sampler <0x0000006e, AMDGPUSample_c_cd_o>;
-defm IMAGE_SAMPLE_C_CD_CL_O : MIMG_Sampler <0x0000006f, AMDGPUSample_c_cd_cl_o>;
+defm IMAGE_SAMPLE_CD            : MIMG_Sampler <0x00000068, AMDGPUSample_cd>;
+defm IMAGE_SAMPLE_CD_CL         : MIMG_Sampler <0x00000069, AMDGPUSample_cd_cl>;
+defm IMAGE_SAMPLE_C_CD          : MIMG_Sampler <0x0000006a, AMDGPUSample_c_cd>;
+defm IMAGE_SAMPLE_C_CD_CL       : MIMG_Sampler <0x0000006b, AMDGPUSample_c_cd_cl>;
+defm IMAGE_SAMPLE_CD_O          : MIMG_Sampler <0x0000006c, AMDGPUSample_cd_o>;
+defm IMAGE_SAMPLE_CD_CL_O       : MIMG_Sampler <0x0000006d, AMDGPUSample_cd_cl_o>;
+defm IMAGE_SAMPLE_C_CD_O        : MIMG_Sampler <0x0000006e, AMDGPUSample_c_cd_o>;
+defm IMAGE_SAMPLE_C_CD_CL_O     : MIMG_Sampler <0x0000006f, AMDGPUSample_c_cd_cl_o>;
+defm IMAGE_SAMPLE_CD_G16        : MIMG_Sampler <0x000000e8, AMDGPUSample_cd, 0, 1>;
+defm IMAGE_SAMPLE_CD_CL_G16     : MIMG_Sampler <0x000000e9, AMDGPUSample_cd_cl, 0, 1>;
+defm IMAGE_SAMPLE_C_CD_G16      : MIMG_Sampler <0x000000ea, AMDGPUSample_c_cd, 0, 1>;
+defm IMAGE_SAMPLE_C_CD_CL_G16   : MIMG_Sampler <0x000000eb, AMDGPUSample_c_cd_cl, 0, 1>;
+defm IMAGE_SAMPLE_CD_O_G16      : MIMG_Sampler <0x000000ec, AMDGPUSample_cd_o, 0, 1>;
+defm IMAGE_SAMPLE_CD_CL_O_G16   : MIMG_Sampler <0x000000ed, AMDGPUSample_cd_cl_o, 0, 1>;
+defm IMAGE_SAMPLE_C_CD_O_G16    : MIMG_Sampler <0x000000ee, AMDGPUSample_c_cd_o, 0, 1>;
+defm IMAGE_SAMPLE_C_CD_CL_O_G16 : MIMG_Sampler <0x000000ef, AMDGPUSample_c_cd_cl_o, 0, 1>;
 //def IMAGE_RSRC256 : MIMG_NoPattern_RSRC256 <"image_rsrc256", 0x0000007e>;
 //def IMAGE_SAMPLER : MIMG_NoPattern_ <"image_sampler", 0x0000007f>;
 
 /********** ========================================= **********/
 /********** Table of dimension-aware image intrinsics **********/
 /********** ========================================= **********/
 
 class ImageDimIntrinsicInfo<AMDGPUImageDimIntrinsic I> {

 def : MIMGLZMapping<IMAGE_GATHER4_L, IMAGE_GATHER4_LZ>;
 def : MIMGLZMapping<IMAGE_GATHER4_C_L, IMAGE_GATHER4_C_LZ>;
 def : MIMGLZMapping<IMAGE_GATHER4_L_O, IMAGE_GATHER4_LZ_O>;
 def : MIMGLZMapping<IMAGE_GATHER4_C_L_O, IMAGE_GATHER4_C_LZ_O>;
 
 // MIP to NONMIP Optimization Mapping
 def : MIMGMIPMapping<IMAGE_LOAD_MIP, IMAGE_LOAD>;
 def : MIMGMIPMapping<IMAGE_STORE_MIP, IMAGE_STORE>;
+
+// G to G16 Optimization Mapping
+def : MIMGG16Mapping<IMAGE_SAMPLE_D, IMAGE_SAMPLE_D_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_D_CL, IMAGE_SAMPLE_D_CL_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_C_D, IMAGE_SAMPLE_C_D_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_C_D_CL, IMAGE_SAMPLE_C_D_CL_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_D_O, IMAGE_SAMPLE_D_O_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_D_CL_O, IMAGE_SAMPLE_D_CL_O_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_C_D_O, IMAGE_SAMPLE_C_D_O_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_C_D_CL_O, IMAGE_SAMPLE_C_D_CL_O_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_CD, IMAGE_SAMPLE_CD_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_CD_CL, IMAGE_SAMPLE_CD_CL_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_C_CD, IMAGE_SAMPLE_C_CD_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_C_CD_CL, IMAGE_SAMPLE_C_CD_CL_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_CD_O, IMAGE_SAMPLE_CD_O_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_CD_CL_O, IMAGE_SAMPLE_CD_CL_O_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_C_CD_O, IMAGE_SAMPLE_C_CD_O_G16>;
+def : MIMGG16Mapping<IMAGE_SAMPLE_C_CD_CL_O, IMAGE_SAMPLE_C_CD_CL_O_G16>;

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

static bool parseTexFail(SDValue TexFailCtrl, SelectionDAG &DAG, SDValue *TFE,
   *TFE = DAG.getTargetConstant((Value & 0x1) ? 1 : 0, DL, MVT::i32);
   Value &= ~(uint64_t)0x1;
   *LWE = DAG.getTargetConstant((Value & 0x2) ? 1 : 0, DL, MVT::i32);
   Value &= ~(uint64_t)0x2;
 
   return Value == 0;
 }
 
+static void packImageA16AddressToDwords(SelectionDAG &DAG, SDValue Op,
+                                        MVT PackVectorVT,
+                                        SmallVectorImpl<SDValue> &PackedAddrs,
+                                        unsigned DimIdx, unsigned EndIdx,
+                                        unsigned NumGradients) {
+  SDLoc DL(Op);
+  for (unsigned I = DimIdx; I < EndIdx; I++) {
+    SDValue Addr = Op.getOperand(I);
+
+    // Gradients are packed with undef for each coordinate.
+    // In <hi 16 bit>,<lo 16 bit> notation, the registers look like this:
+    // 1D: undef,dx/dh; undef,dx/dv
+    // 2D: dy/dh,dx/dh; dy/dv,dx/dv
+    // 3D: dy/dh,dx/dh; undef,dz/dh; dy/dv,dx/dv; undef,dz/dv
+    if (((I + 1) >= EndIdx) ||
+        ((NumGradients / 2) % 2 == 1 && (I == DimIdx + (NumGradients / 2) - 1 ||
+                                         I == DimIdx + NumGradients - 1))) {
+      if (Addr.getValueType() != MVT::i16)
+        Addr = DAG.getBitcast(MVT::i16, Addr);
+      Addr = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i32, Addr);
+    } else {
+      Addr = DAG.getBuildVector(PackVectorVT, DL, {Addr, Op.getOperand(I + 1)});
+      I++;
+    }
+    Addr = DAG.getBitcast(MVT::f32, Addr);
+    PackedAddrs.push_back(Addr);
+  }
+}
+
 SDValue SITargetLowering::lowerImage(SDValue Op,
                                      const AMDGPU::ImageDimIntrinsicInfo *Intr,
                                      SelectionDAG &DAG) const {
   SDLoc DL(Op);
   MachineFunction &MF = DAG.getMachineFunction();
   const GCNSubtarget* ST = &MF.getSubtarget<GCNSubtarget>();
   const AMDGPU::MIMGBaseOpcodeInfo *BaseOpcode =
       AMDGPU::getMIMGBaseOpcodeInfo(Intr->BaseOpcode);
   const AMDGPU::MIMGDimInfo *DimInfo = AMDGPU::getMIMGDimInfo(Intr->Dim);
   const AMDGPU::MIMGLZMappingInfo *LZMappingInfo =
       AMDGPU::getMIMGLZMappingInfo(Intr->BaseOpcode);
   const AMDGPU::MIMGMIPMappingInfo *MIPMappingInfo =
       AMDGPU::getMIMGMIPMappingInfo(Intr->BaseOpcode);
   unsigned IntrOpcode = Intr->BaseOpcode;
   bool IsGFX10 = Subtarget->getGeneration() >= AMDGPUSubtarget::GFX10;
 
   SmallVector<EVT, 3> ResultTypes(Op->value_begin(), Op->value_end());
   SmallVector<EVT, 3> OrigResultTypes(Op->value_begin(), Op->value_end());
   bool IsD16 = false;
+  bool IsG16 = false;
   bool IsA16 = false;
   SDValue VData;
   int NumVDataDwords;
   bool AdjustRetType = false;
 
   unsigned AddrIdx; // Index of first address argument
   unsigned DMask;
   unsigned DMaskLanes = 0;
if (auto ConstantLod =
          dyn_cast<ConstantSDNode>(Op.getOperand(AddrIdx+NumVAddrs-1))) {
       if (ConstantLod->isNullValue()) {
         IntrOpcode = MIPMappingInfo->NONMIP;  // set new opcode to variant without _mip
         NumMIVAddrs--;               // remove 'lod'
       }
     }
   }
 
-  // Check for 16 bit addresses and pack if true.
+  // Push back extra arguments.
+  for (unsigned I = 0; I < BaseOpcode->NumExtraArgs; I++)
+    VAddrs.push_back(Op.getOperand(AddrIdx + I));
+
+  // Check for 16 bit addresses or derivatives and pack if true.
   unsigned DimIdx = AddrIdx + BaseOpcode->NumExtraArgs;
+  unsigned CoordIdx = DimIdx + NumGradients;
+  unsigned CoordsEnd = AddrIdx + NumMIVAddrs;
+
   MVT VAddrVT = Op.getOperand(DimIdx).getSimpleValueType();
-  const MVT VAddrScalarVT = VAddrVT.getScalarType();
-  if (((VAddrScalarVT == MVT::f16) || (VAddrScalarVT == MVT::i16))) {
-    // Illegal to use a16 images
-    if (!ST->hasFeature(AMDGPU::FeatureR128A16) && !ST->hasFeature(AMDGPU::FeatureGFX10A16))
+  MVT VAddrScalarVT = VAddrVT.getScalarType();
+  MVT PackVectorVT = VAddrScalarVT == MVT::f16 ? MVT::v2f16 : MVT::v2i16;
+  IsG16 = VAddrScalarVT == MVT::f16 || VAddrScalarVT == MVT::i16;
+
+  VAddrVT = Op.getOperand(CoordIdx).getSimpleValueType();
+  VAddrScalarVT = VAddrVT.getScalarType();
+  IsA16 = VAddrScalarVT == MVT::f16 || VAddrScalarVT == MVT::i16;
+  if (IsA16 || IsG16) {
+    if (IsA16) {
+      if (!ST->hasA16()) {
+        LLVM_DEBUG(dbgs() << "Failed to lower image intrinsic: Target does not "
+                             "support 16 bit addresses\n");
         return Op;
+      }
+      if (!IsG16) {
+        LLVM_DEBUG(
+            dbgs() << "Failed to lower image intrinsic: 16 bit addresses "
+                      "need 16 bit derivatives but got 32 bit derivatives\n");
+        return Op;
+      }
+    } else if (!ST->hasG16()) {
+      LLVM_DEBUG(dbgs() << "Failed to lower image intrinsic: Target does not "
+                           "support 16 bit derivatives\n");
+      return Op;
+    }
 
-    IsA16 = true;
-    const MVT VectorVT = VAddrScalarVT == MVT::f16 ? MVT::v2f16 : MVT::v2i16;
-    for (unsigned i = AddrIdx; i < (AddrIdx + NumMIVAddrs); ++i) {
-      SDValue AddrLo;
-      // Push back extra arguments.
-      if (i < DimIdx) {
-        AddrLo = Op.getOperand(i);
-      } else {
-        // Dz/dh, dz/dv and the last odd coord are packed with undef. Also,
-        // in 1D, derivatives dx/dh and dx/dv are packed with undef.
-        if (((i + 1) >= (AddrIdx + NumMIVAddrs)) ||
-            ((NumGradients / 2) % 2 == 1 &&
-            (i == DimIdx + (NumGradients / 2) - 1 ||
-             i == DimIdx + NumGradients - 1))) {
-          AddrLo = Op.getOperand(i);
-          if (AddrLo.getValueType() != MVT::i16)
-            AddrLo = DAG.getBitcast(MVT::i16, Op.getOperand(i));
-          AddrLo = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i32, AddrLo);
-        } else {
-          AddrLo = DAG.getBuildVector(VectorVT, DL,
-                                      {Op.getOperand(i), Op.getOperand(i + 1)});
-          i++;
-        }
-        AddrLo = DAG.getBitcast(MVT::f32, AddrLo);
-      }
-      VAddrs.push_back(AddrLo);
+    if (BaseOpcode->Gradients && !IsA16) {
+      if (!ST->hasG16()) {
+        LLVM_DEBUG(dbgs() << "Failed to lower image intrinsic: Target does not "
+                             "support 16 bit derivatives\n");
+        return Op;
+      // Activate g16
+      const AMDGPU::MIMGG16MappingInfo *G16MappingInfo =
+          AMDGPU::getMIMGG16MappingInfo(Intr->BaseOpcode);
+      IntrOpcode = G16MappingInfo->G16; // set new opcode to variant with _g16
+
+    // Don't compress addresses for G16
+    const int PackEndIdx = IsA16 ? CoordsEnd : CoordIdx;
+    packImageA16AddressToDwords(DAG, Op, PackVectorVT, VAddrs, DimIdx,
+                                PackEndIdx, NumGradients);
+
+    if (!IsA16) {
+      // Add uncompressed address
+      for (unsigned I = CoordIdx; I < CoordsEnd; I++)
+        VAddrs.push_back(Op.getOperand(I));
     }
   } else {
-    for (unsigned i = 0; i < NumMIVAddrs; ++i)
-      VAddrs.push_back(Op.getOperand(AddrIdx + i));
+    for (unsigned I = DimIdx; I < CoordsEnd; I++)
+      VAddrs.push_back(Op.getOperand(I));
   }
 
   // If the register allocator cannot place the address registers contiguously
   // without introducing moves, then using the non-sequential address encoding
   // is always preferable, since it saves VALU instructions and is usually a
   // wash in terms of code size or even better.
   //
   // However, we currently have no way of hinting to the register allocator that
   // MIMG addresses should be placed contiguously when it is possible to do so,
   // so force non-NSA for the common 2-address case as a heuristic.
   //
   // SIShrinkInstructions will convert NSA encodings to non-NSA after register
   // allocation when possible.
   bool UseNSA =

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

 #include "llvm/Target/TargetMachine.h"
 #include <cassert>
 #include <cstdint>
 #include <iterator>
 #include <utility>
 
 using namespace llvm;
 
+#define DEBUG_TYPE "si-instr-info"
+
 #define GET_INSTRINFO_CTOR_DTOR
 #include "AMDGPUGenInstrInfo.inc"
 
 namespace llvm {
 namespace AMDGPU {
 #define GET_D16ImageDimIntrinsics_IMPL
 #define GET_ImageDimIntrinsicTable_IMPL
 #define GET_RsrcIntrinsics_IMPL
if (DimOp) {
       if (ST.hasR128A16()) {
         const MachineOperand *R128A16 = getNamedOperand(MI, AMDGPU::OpName::r128);
         IsA16 = R128A16->getImm() != 0;
       } else if (ST.hasGFX10A16()) {
         const MachineOperand *A16 = getNamedOperand(MI, AMDGPU::OpName::a16);
         IsA16 = A16->getImm() != 0;
       }
 
-      bool PackDerivatives = IsA16; // Either A16 or G16
+      bool PackDerivatives = IsA16 || BaseOpcode->G16;
       bool IsNSA = SRsrcIdx - VAddr0Idx > 1;
 
       unsigned AddrWords = BaseOpcode->NumExtraArgs;
       unsigned AddrComponents = (BaseOpcode->Coordinates ? Dim->NumCoords : 0) +
                                 (BaseOpcode->LodOrClampOrMip ? 1 : 0);
       if (IsA16)
         AddrWords += (AddrComponents + 1) / 2;
       else
if (DimOp) {
           AddrWords = 8;
         else if (AddrWords == 4)
           AddrWords = 4;
         else if (AddrWords == 3)
           AddrWords = 3;
       }
 
       if (VAddrWords != AddrWords) {
+        LLVM_DEBUG(dbgs() << "bad vaddr size, expected " << AddrWords
+                          << " but got " << VAddrWords << "\n");
         ErrInfo = "bad vaddr size";
         return false;
       }
     }
   }
 
   const MachineOperand *DppCt = getNamedOperand(MI, AMDGPU::OpName::dpp_ctrl);
   if (DppCt) {

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h

struct MIMGBaseOpcodeInfo {
   bool Store;
   bool Atomic;
   bool AtomicX2;
   bool Sampler;
   bool Gather4;
 
   uint8_t NumExtraArgs;
   bool Gradients;
+  bool G16;
   bool Coordinates;
   bool LodOrClampOrMip;
   bool HasD16;
 };
 
 LLVM_READONLY
 const MIMGBaseOpcodeInfo *getMIMGBaseOpcodeInfo(unsigned BaseOpcode);
 
struct MIMGLZMappingInfo {
   MIMGBaseOpcode LZ;
 };
 
 struct MIMGMIPMappingInfo {
   MIMGBaseOpcode MIP;
   MIMGBaseOpcode NONMIP;
 };
 
+struct MIMGG16MappingInfo {
+  MIMGBaseOpcode G;
+  MIMGBaseOpcode G16;
+};
+
 LLVM_READONLY
 const MIMGLZMappingInfo *getMIMGLZMappingInfo(unsigned L);
 
 LLVM_READONLY
-const MIMGMIPMappingInfo *getMIMGMIPMappingInfo(unsigned L);
+const MIMGMIPMappingInfo *getMIMGMIPMappingInfo(unsigned MIP);
+
+LLVM_READONLY
+const MIMGG16MappingInfo *getMIMGG16MappingInfo(unsigned G);
 
 LLVM_READONLY
 int getMIMGOpcode(unsigned BaseOpcode, unsigned MIMGEncoding,
                   unsigned VDataDwords, unsigned VAddrDwords);
 
 LLVM_READONLY
 int getMaskedMIMGOp(unsigned Opc, unsigned NewChannels);
 
default:
     return false;
   }
 }
 
 bool hasXNACK(const MCSubtargetInfo &STI);
 bool hasSRAMECC(const MCSubtargetInfo &STI);
 bool hasMIMG_R128(const MCSubtargetInfo &STI);
 bool hasGFX10A16(const MCSubtargetInfo &STI);
+bool hasG16(const MCSubtargetInfo &STI);
 bool hasPackedD16(const MCSubtargetInfo &STI);
 
 bool isSI(const MCSubtargetInfo &STI);
 bool isCI(const MCSubtargetInfo &STI);
 bool isVI(const MCSubtargetInfo &STI);
 bool isGFX9(const MCSubtargetInfo &STI);
 bool isGFX10(const MCSubtargetInfo &STI);
 

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp

 
 namespace AMDGPU {
 
 #define GET_MIMGBaseOpcodesTable_IMPL
 #define GET_MIMGDimInfoTable_IMPL
 #define GET_MIMGInfoTable_IMPL
 #define GET_MIMGLZMappingTable_IMPL
 #define GET_MIMGMIPMappingTable_IMPL
+#define GET_MIMGG16MappingTable_IMPL
 #include "AMDGPUGenSearchableTables.inc"
 
 int getMIMGOpcode(unsigned BaseOpcode, unsigned MIMGEncoding,
                   unsigned VDataDwords, unsigned VAddrDwords) {
   const MIMGInfo *Info = getMIMGOpcodeHelper(BaseOpcode, MIMGEncoding,
                                              VDataDwords, VAddrDwords);
   return Info ? Info->Opcode : -1;
 }

 bool hasMIMG_R128(const MCSubtargetInfo &STI) {
   return STI.getFeatureBits()[AMDGPU::FeatureMIMG_R128] && !STI.getFeatureBits()[AMDGPU::FeatureR128A16];
 }
 
 bool hasGFX10A16(const MCSubtargetInfo &STI) {
   return STI.getFeatureBits()[AMDGPU::FeatureGFX10A16];
 }
 
+bool hasG16(const MCSubtargetInfo &STI) {
+  return STI.getFeatureBits()[AMDGPU::FeatureG16];
+}
+
 bool hasPackedD16(const MCSubtargetInfo &STI) {
   return !STI.getFeatureBits()[AMDGPU::FeatureUnpackedD16VMem];
 }
 
 bool isSI(const MCSubtargetInfo &STI) {
   return STI.getFeatureBits()[AMDGPU::FeatureSouthernIslands];
 }
 

llvm/test/CodeGen/AMDGPU/GlobalISel/image_ls_mipmap_zero.a16.ll

define amdgpu_ps <4 x float> @load_mip_1d(<8 x i32> inreg %rsrc, i16 %s) {
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[DEF]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: load_mip_1d
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[DEF]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.load.mip.1d.v4f32.i16(i32 15, i16 %s, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
define amdgpu_ps <4 x float> @load_mip_2d(<8 x i32> inreg %rsrc, i16 %s, i16 %t) {
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: load_mip_2d
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.load.mip.2d.v4f32.i16(i32 15, i16 %s, i16 %t, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
define amdgpu_ps <4 x float> @load_mip_3d(<8 x i32> inreg %rsrc, i16 %s, i16 %t, i16 %u) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: load_mip_3d
   ; GFX10: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_3d(<8 x i32> inreg %rsrc, i16 %s, i16 %t, i16 %u) {
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.load.mip.3d.v4f32.i16(i32 15, i16 %s, i16 %t, i16 %u, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
define amdgpu_ps <4 x float> @load_mip_1darray(<8 x i32> inreg %rsrc, i16 %s, i16 %t) {
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: load_mip_1darray
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.load.mip.1darray.v4f32.i16(i32 15, i16 %s, i16 %t, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
define amdgpu_ps <4 x float> @load_mip_2darray(<8 x i32> inreg %rsrc, i16 %s, i16 %t, i16 %u) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: load_mip_2darray
   ; GFX10: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_2darray(<8 x i32> inreg %rsrc, i16 %s, i16 %t, i16 %u) {
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.load.mip.2darray.v4f32.i16(i32 15, i16 %s, i16 %t, i16 %u, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
define amdgpu_ps <4 x float> @load_mip_cube(<8 x i32> inreg %rsrc, i16 %s, i16 %t, i16 %u) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10-LABEL: name: load_mip_cube
   ; GFX10: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_cube(<8 x i32> inreg %rsrc, i16 %s, i16 %t, i16 %u) {
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY8]](s32)
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
-  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %v = call <4 x float> @llvm.amdgcn.image.load.mip.cube.v4f32.i16(i32 15, i16 %s, i16 %t, i16 %u, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
define amdgpu_ps void @store_mip_1d(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s) {
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr4
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10-LABEL: name: store_mip_1d
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr4
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[DEF]](s32)
-  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10:   S_ENDPGM 0
 main_body:
   call void @llvm.amdgcn.image.store.mip.1d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @store_mip_2d(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t) {
   ; GFX9-LABEL: name: store_mip_2d
define amdgpu_ps void @store_mip_2d(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t) {
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr4
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr5
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10-LABEL: name: store_mip_2d
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr4
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr5
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
-  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10:   S_ENDPGM 0
 main_body:
   call void @llvm.amdgcn.image.store.mip.2d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @store_mip_3d(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX9-LABEL: name: store_mip_3d
define amdgpu_ps void @store_mip_3d(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
   ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10-LABEL: name: store_mip_3d
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_3d(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
   ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
-  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10:   S_ENDPGM 0
 main_body:
   call void @llvm.amdgcn.image.store.mip.3d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %u, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @store_mip_1darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t) {
   ; GFX9-LABEL: name: store_mip_1darray
define amdgpu_ps void @store_mip_1darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t) {
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr4
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr5
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10-LABEL: name: store_mip_1darray
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr4
   ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr5
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
-  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10:   S_ENDPGM 0
 main_body:
   call void @llvm.amdgcn.image.store.mip.1darray.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @store_mip_2darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX9-LABEL: name: store_mip_2darray
define amdgpu_ps void @store_mip_2darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
   ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10-LABEL: name: store_mip_2darray
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_2darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
   ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
-  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10:   S_ENDPGM 0
 main_body:
   call void @llvm.amdgcn.image.store.mip.2darray.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %u, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @store_mip_cube(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX9-LABEL: name: store_mip_cube
define amdgpu_ps void @store_mip_cube(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
   ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10-LABEL: name: store_mip_cube
   ; GFX10: bb.1.main_body:
   ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
   ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_cube(<8 x i32> inreg %rsrc, <4 x float> %vdata, i16 %s, i16 %t, i16 %u) {
   ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
   ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
   ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
-  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, 0, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10:   S_ENDPGM 0
 main_body:
   call void @llvm.amdgcn.image.store.mip.cube.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %u, i16 0, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 declare <4 x float> @llvm.amdgcn.image.load.mip.1d.v4f32.i16(i32 immarg, i16, i16, <8 x i32>, i32 immarg, i32 immarg) #0
 declare <4 x float> @llvm.amdgcn.image.load.mip.2d.v4f32.i16(i32 immarg, i16, i16, i16, <8 x i32>, i32 immarg, i32 immarg) #0

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.atomic.dim.a16.ll

define amdgpu_ps float @atomic_swap_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.swap.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.swap.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_swap_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.swap.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.swap.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.swap.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_add_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_add_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_sub_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_sub_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.sub.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.sub.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_sub_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.sub.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.sub.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.sub.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_smin_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_smin_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_smin_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.smin.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_umin_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_umin_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umin.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.umin.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_smax_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_smax_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_smax_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.smax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.smax.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_umax_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_umax_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_umax_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.umax.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.umax.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_and_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_and_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.and.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.and.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_and_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.and.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.and.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.and.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_or_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_or_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.or.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.or.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_or_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.or.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.or.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.or.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_xor_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_xor_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.xor.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.xor.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_xor_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.xor.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.xor.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.xor.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_inc_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_inc_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.inc.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.inc.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_inc_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.inc.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.inc.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.inc.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_dec_1d(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_dec_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.dec.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.dec.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_dec_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.dec.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.dec.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.dec.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_cmpswap_1d(<8 x i32> inreg %rsrc, i32 %cmp, i32 %swap, i16 %s) {
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY10]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_cmpswap_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY10]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.1d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.cmpswap.1d.i32.i16(i32 %cmp, i32 %swap, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_add_2d(<8 x i32> inreg %rsrc, i32 %data, i16 %s, i16 %t) {
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_2d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.2d.i32.i16(i32 %data, i16 %s, i16 %t, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_add_3d(<8 x i32> inreg %rsrc, i32 %data, i16 %s, i16 %t, i16 %r) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.3d), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.3d), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_3d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.3d), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.3d), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.3d.i32.i16(i32 %data, i16 %s, i16 %t, i16 %r, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_add_cube(<8 x i32> inreg %rsrc, i32 %data, i16 %s, i16 %t, i16 %face) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.cube), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.cube), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_cube
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.cube), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.cube), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.cube.i32.i16(i32 %data, i16 %s, i16 %t, i16 %face, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_add_1darray(<8 x i32> inreg %rsrc, i32 %data, i16 %s, i16 %slice) {
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_1darray
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY11]](s32), [[COPY12]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.1darray.i32.i16(i32 %data, i16 %s, i16 %slice, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_add_2darray(<8 x i32> inreg %rsrc, i32 %data, i16 %s, i16 %t, i16 %slice) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darray), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darray), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_2darray
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darray), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darray), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.2darray.i32.i16(i32 %data, i16 %s, i16 %t, i16 %slice, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_add_2dmsaa(<8 x i32> inreg %rsrc, i32 %data, i16 %s, i16 %t, i16 %fragid) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2dmsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2dmsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_2dmsaa
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2dmsaa), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2dmsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.2dmsaa.i32.i16(i32 %data, i16 %s, i16 %t, i16 %fragid, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_add_2darraymsaa(<8 x i32> inreg %rsrc, i32 %data, i16 %s, i16 %t, i16 %slice, i16 %fragid) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darraymsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darraymsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_2darraymsaa
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
define amdgpu_ps float @atomic_add_2darraymsaa(<8 x i32> inreg %rsrc, i32 %data, i16 %s, i16 %t, i16 %slice, i16 %fragid) {
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr4
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY9]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darraymsaa), [[COPY8]](s32), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.2darraymsaa), [[COPY8]](s32), [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.2darraymsaa.i32.i16(i32 %data, i16 %s, i16 %t, i16 %slice, i16 %fragid, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
 define amdgpu_ps float @atomic_add_1d_slc(<8 x i32> inreg %rsrc, i32 %data, i16 %s) {
   ; GFX9-LABEL: name: atomic_add_1d_slc
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_add_1d_slc
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY9]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.add.1d), [[COPY8]](s32), [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.add.1d.i32.i16(i32 %data, i16 %s, <8 x i32> %rsrc, i32 0, i32 2)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_cmpswap_2d(<8 x i32> inreg %rsrc, i32 %cmp, i32 %swap, i16 %s, i16 %t) {
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_cmpswap_2d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.cmpswap.2d.i32.i16(i32 %cmp, i32 %swap, i16 %s, i16 %t, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_cmpswap_3d(<8 x i32> inreg %rsrc, i32 %cmp, i32 %swap, i16 %s, i16 %t, i16 %r) {
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.3d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.3d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_cmpswap_3d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
define amdgpu_ps float @atomic_cmpswap_3d(<8 x i32> inreg %rsrc, i32 %cmp, i32 %swap, i16 %s, i16 %t, i16 %r) {
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.3d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.3d), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.cmpswap.3d.i32.i16(i32 %cmp, i32 %swap, i16 %s, i16 %t, i16 %r, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 
define amdgpu_ps float @atomic_cmpswap_2darraymsaa(<8 x i32> inreg %rsrc, i32 %cmp, i32 %swap, i16 %s, i16 %t, i16 %slice, i16 %fragid) {
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2darraymsaa), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2darraymsaa), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: atomic_cmpswap_2darraymsaa
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
define amdgpu_ps float @atomic_cmpswap_2darraymsaa(<8 x i32> inreg %rsrc, i32 %cmp, i32 %swap, i16 %s, i16 %t, i16 %slice, i16 %fragid) {
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY10]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY11]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
   ; GFX10NSA:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2darraymsaa), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.atomic.cmpswap.2darraymsaa), [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (volatile dereferenceable load store 4 on custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %v = call i32 @llvm.amdgcn.image.atomic.cmpswap.2darraymsaa.i32.i16(i32 %cmp, i32 %swap, i16 %s, i16 %t, i16 %slice, i16 %fragid, <8 x i32> %rsrc, i32 0, i32 0)
   %out = bitcast i32 %v to float
   ret float %out
 }
 

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.dim.a16.ll

 ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX9 %s
-; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -mattr=+r128-a16 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX10NSA %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX10NSA %s
 
 define amdgpu_ps <4 x float> @load_1d(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9-LABEL: name: load_1d
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @load_2d(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX9:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_2d
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_2d(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @load_3d(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_3d
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_3d(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_cube(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_cube
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_cube(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.cube), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_1darray(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX9:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_1darray
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_1darray(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @load_2darray(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_2darray
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_2darray(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_2dmsaa(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2dmsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2dmsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_2dmsaa
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_2dmsaa(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2dmsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2dmsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_2darraymsaa(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_2darraymsaa
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_2darraymsaa(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_mip_1d(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX9:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_mip_1d
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_1d(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @load_mip_2d(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_mip_2d
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_2d(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_mip_3d(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_mip_3d
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_3d(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_mip_cube(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_mip_cube
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_cube(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_mip_1darray(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_mip_1darray
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_1darray(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.1darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps <4 x float> @load_mip_2darray(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_mip_2darray
   ; GFX10NSA: bb.1.main_body:
define amdgpu_ps <4 x float> @load_mip_2darray(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.mip.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
define amdgpu_ps void @store_1d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[COPY12:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   call void @llvm.amdgcn.image.store.1d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @store_2d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
define amdgpu_ps void @store_2d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX9:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_2d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_2d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   %t = extractelement <2 x i16> %coords, i32 1
   call void @llvm.amdgcn.image.store.2d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
define amdgpu_ps void @store_3d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_3d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_3d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %r = extractelement <2 x i16> %coords_hi, i32 0
   call void @llvm.amdgcn.image.store.3d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %r, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
define amdgpu_ps void @store_cube(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_cube
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_cube(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %slice = extractelement <2 x i16> %coords_hi, i32 0
   call void @llvm.amdgcn.image.store.cube.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %slice, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
define amdgpu_ps void @store_1darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX9:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_1darray
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_1darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   %slice = extractelement <2 x i16> %coords, i32 1
   call void @llvm.amdgcn.image.store.1darray.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %slice, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
define amdgpu_ps void @store_2darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_2darray
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_2darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %slice = extractelement <2 x i16> %coords_hi, i32 0
   call void @llvm.amdgcn.image.store.2darray.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %slice, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
define amdgpu_ps void @store_2dmsaa(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2dmsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2dmsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_2dmsaa
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_2dmsaa(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2dmsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2dmsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %fragid = extractelement <2 x i16> %coords_hi, i32 0
   call void @llvm.amdgcn.image.store.2dmsaa.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %fragid, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
define amdgpu_ps void @store_2darraymsaa(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darraymsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darraymsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_2darraymsaa
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_2darraymsaa(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darraymsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2darraymsaa), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %slice = extractelement <2 x i16> %coords_hi, i32 0
   %fragid = extractelement <2 x i16> %coords_hi, i32 1
   call void @llvm.amdgcn.image.store.2darraymsaa.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %slice, i16 %fragid, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
define amdgpu_ps void @store_mip_1d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX9:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_mip_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_1d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY13]](s32), [[COPY14]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   %mip = extractelement <2 x i16> %coords, i32 1
   call void @llvm.amdgcn.image.store.mip.1d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %mip, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
define amdgpu_ps void @store_mip_2d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_mip_2d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_2d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %mip = extractelement <2 x i16> %coords_hi, i32 0
   call void @llvm.amdgcn.image.store.mip.2d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %mip, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
define amdgpu_ps void @store_mip_3d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_mip_3d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_3d(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.3d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %r = extractelement <2 x i16> %coords_hi, i32 0
   %mip = extractelement <2 x i16> %coords_hi, i32 1
   call void @llvm.amdgcn.image.store.mip.3d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %r, i16 %mip, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
define amdgpu_ps void @store_mip_cube(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_mip_cube
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_cube(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.cube), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %slice = extractelement <2 x i16> %coords_hi, i32 0
   %mip = extractelement <2 x i16> %coords_hi, i32 1
   call void @llvm.amdgcn.image.store.mip.cube.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %slice, i16 %mip, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
define amdgpu_ps void @store_mip_1darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_mip_1darray
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_1darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.1darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %slice = extractelement <2 x i16> %coords_lo, i32 1
   %mip = extractelement <2 x i16> %coords_hi, i32 0
   call void @llvm.amdgcn.image.store.mip.1darray.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %slice, i16 %mip, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
define amdgpu_ps void @store_mip_2darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_mip_2darray
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
define amdgpu_ps void @store_mip_2darray(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY13]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
   ; GFX10NSA:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.mip.2darray), [[BUILD_VECTOR1]](<4 x s32>), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords_lo, i32 0
   %t = extractelement <2 x i16> %coords_lo, i32 1
   %slice = extractelement <2 x i16> %coords_hi, i32 0
   %mip = extractelement <2 x i16> %coords_hi, i32 1
   call void @llvm.amdgcn.image.store.mip.2darray.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, i16 %t, i16 %slice, i16 %mip, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
define amdgpu_ps <4 x float> @getresinfo_1d(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: getresinfo_1d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %mip = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @getresinfo_2d(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: getresinfo_2d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %mip = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @getresinfo_3d(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.3d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.3d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: getresinfo_3d
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.3d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.3d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %mip = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @getresinfo_cube(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.cube), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.cube), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: getresinfo_cube
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.cube), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.cube), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %mip = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @getresinfo_1darray(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: getresinfo_1darray
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.1darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %mip = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @getresinfo_2darray(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: getresinfo_2darray
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darray), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %mip = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @getresinfo_2dmsaa(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2dmsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2dmsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: getresinfo_2dmsaa
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2dmsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2dmsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %mip = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @getresinfo_2darraymsaa(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darraymsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darraymsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: getresinfo_2darraymsaa
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darraymsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.getresinfo.2darraymsaa), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %mip = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps float @load_1d_V1(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 8, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 8, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: load_1d_V1
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 8, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 8, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   %v = call float @llvm.amdgcn.image.load.1d.f32.i16(i32 8, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   ret float %v
 }
 
 define amdgpu_ps <2 x float> @load_1d_V2(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9-LABEL: name: load_1d_V2
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 9, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 9, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   ; GFX10NSA-LABEL: name: load_1d_V2
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 9, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 9, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   %v = call <2 x float> @llvm.amdgcn.image.load.1d.v2f32.i16(i32 9, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   ret <2 x float> %v
define amdgpu_ps void @store_1d_V1(<8 x i32> inreg %rsrc, float %vdata, <2 x i16> %coords) {
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[COPY8]](s32), 2, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[COPY8]](s32), 2, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 4 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_1d_V1
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[COPY8]](s32), 2, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[COPY8]](s32), 2, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 4 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   call void @llvm.amdgcn.image.store.1d.f32.i16(float %vdata, i32 2, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @store_1d_V2(<8 x i32> inreg %rsrc, <2 x float> %vdata, <2 x i16> %coords) {
define amdgpu_ps void @store_1d_V2(<8 x i32> inreg %rsrc, <2 x float> %vdata, <2 x i16> %coords) {
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY10]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<2 x s32>), 12, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<2 x s32>), 12, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 8 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_1d_V2
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY10]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<2 x s32>), 12, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<2 x s32>), 12, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 3 :: (dereferenceable store 8 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   call void @llvm.amdgcn.image.store.1d.v2f32.i16(<2 x float> %vdata, i32 12, i16 %s, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps <4 x float> @load_1d_glc(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9-LABEL: name: load_1d_glc
   ; GFX9: bb.1.main_body:
   ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_1d_glc
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @load_1d_slc(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_1d_slc
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps <4 x float> @load_1d_glc_slc(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_1d_glc_slc
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
define amdgpu_ps void @store_1d_glc(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[COPY12:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_1d_glc
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 1, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   call void @llvm.amdgcn.image.store.1d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, <8 x i32> %rsrc, i32 0, i32 1)
   ret void
 }
 
 define amdgpu_ps void @store_1d_slc(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
define amdgpu_ps void @store_1d_slc(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[COPY12:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_1d_slc
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 2, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   call void @llvm.amdgcn.image.store.1d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, <8 x i32> %rsrc, i32 0, i32 2)
   ret void
 }
 
 define amdgpu_ps void @store_1d_glc_slc(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
define amdgpu_ps void @store_1d_glc_slc(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX9:   [[COPY12:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX9:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX9:   S_ENDPGM 0
   ; GFX10NSA-LABEL: name: store_1d_glc_slc
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr4
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY12]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
+  ; GFX10NSA:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.1d), [[BUILD_VECTOR1]](<4 x s32>), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 0, 3, 3 :: (dereferenceable store 16 into custom "TargetCustom8")
   ; GFX10NSA:   S_ENDPGM 0
 main_body:
   %s = extractelement <2 x i16> %coords, i32 0
   call void @llvm.amdgcn.image.store.1d.v4f32.i16(<4 x float> %vdata, i32 15, i16 %s, <8 x i32> %rsrc, i32 0, i32 3)
   ret void
 }
 
 define amdgpu_ps <4 x float> @getresinfo_dmask0(<8 x i32> inreg %rsrc, <4 x float> %vdata, <2 x i16> %coords) {
define amdgpu_ps <4 x float> @load_1d_tfe(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX9:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX9:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 1, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX9:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_1d_tfe
   ; GFX10NSA: bb.1.main_body:
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), 1, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX10NSA:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
define amdgpu_ps <4 x float> @load_2d_tfe(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX9:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GFX9:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX9:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX9:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX9:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_2d_tfe
define amdgpu_ps <4 x float> @load_2d_tfe(<8 x i32> inreg %rsrc, <2 x i16> %coords) {
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY8]](<2 x s16>)
   ; GFX10NSA:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX10NSA:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
define amdgpu_ps <4 x float> @load_3d_tfe(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF1]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX9:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_3d_tfe
define amdgpu_ps <4 x float> @load_3d_tfe(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; GFX10NSA:   [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[DEF1]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX10NSA:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:
define amdgpu_ps <4 x float> @load_2darraymsaa_tfe(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX9:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
   ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
-  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX9:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX9:   $vgpr0 = COPY [[UV]](s32)
   ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_2darraymsaa_tfe
define amdgpu_ps <4 x float> @load_2darraymsaa_tfe(<8 x i32> inreg %rsrc, <2 x i16> %coords_lo, <2 x i16> %coords_hi) {
   ; GFX10NSA:   [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[COPY9]](<2 x s16>)
   ; GFX10NSA:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY10]](s32), [[COPY11]](s32)
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; GFX10NSA:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY12]](s32), [[COPY13]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX10NSA:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
 main_body:

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.d16.ll

define amdgpu_ps half @image_load_f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s16) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s16) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[AMDGPU_INTRIN_IMAGE_LOAD]](s16)
   ; UNPACKED:   $vgpr0 = COPY [[ANYEXT]](s32)
   ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; PACKED-LABEL: name: image_load_f16
   ; PACKED: bb.1 (%ir-block.0):
   ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; PACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; PACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; PACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; PACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s16) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s16) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[AMDGPU_INTRIN_IMAGE_LOAD]](s16)
   ; PACKED:   $vgpr0 = COPY [[ANYEXT]](s32)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %tex = call half @llvm.amdgcn.image.load.2d.f16.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret half %tex
 }
 
 define amdgpu_ps <2 x half> @image_load_v2f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED-LABEL: name: image_load_v2f16
   ; UNPACKED: bb.1 (%ir-block.0):
   ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; UNPACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; UNPACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; UNPACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; UNPACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <2 x half> @image_load_v2f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; PACKED:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %tex = call <2 x half> @llvm.amdgcn.image.load.2d.v2f16.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <2 x half> %tex
 }
 
 define amdgpu_ps <3 x half> @image_load_v3f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED-LABEL: name: image_load_v3f16
   ; UNPACKED: bb.1 (%ir-block.0):
   ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; UNPACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; UNPACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; UNPACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; UNPACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <3 x half> @image_load_v3f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
   ; PACKED:   [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s16>)
   ; PACKED:   [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[UV]](<2 x s16>), [[UV1]](<2 x s16>), [[DEF]](<2 x s16>)
   ; PACKED:   [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV2]](<3 x s16>), 0
   ; PACKED:   [[EXTRACT:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
   ; PACKED:   [[EXTRACT1:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 32
define amdgpu_ps <4 x half> @image_load_v4f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <4 x half> @image_load_v4f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s16>)
   ; PACKED:   $vgpr0 = COPY [[UV]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[UV1]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %tex = call <4 x half> @llvm.amdgcn.image.load.2d.v4f16.i32(i32 15, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <4 x half> %tex
 }
 
define amdgpu_ps half @image_load_tfe_f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   $vgpr0 = COPY [[COPY10]](s32)
   ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; PACKED-LABEL: name: image_load_tfe_f16
   ; PACKED: bb.1 (%ir-block.0):
   ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; PACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; PACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; PACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; PACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; PACKED:   $vgpr0 = COPY [[COPY10]](s32)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %res = call { half, i32 } @llvm.amdgcn.image.load.2d.sl_f16i32s.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { half, i32 } %res, 0
   %tfe = extractvalue { half, i32 } %res, 1
define amdgpu_ps <2 x half> @image_load_tfe_v2f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <2 x half> @image_load_tfe_v2f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   $vgpr0 = COPY [[BITCAST]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %res = call { <2 x half>, i32 } @llvm.amdgcn.image.load.2d.sl_v2f16i32s.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { <2 x half>, i32 } %res, 0
   %tfe = extractvalue { <2 x half>, i32 } %res, 1
define amdgpu_ps <3 x half> @image_load_tfe_v3f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <3 x half> @image_load_tfe_v3f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV1]](s32)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[DEF1]](<2 x s16>)
   ; PACKED:   [[UV3:%[0-9]+]]:_(<3 x s16>), [[UV4:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
   ; PACKED:   G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
define amdgpu_ps <4 x half> @image_load_tfe_v4f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; UNPACKED:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
define amdgpu_ps <4 x half> @image_load_tfe_v4f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV1]](s32)
   ; PACKED:   G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   $vgpr0 = COPY [[BITCAST]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[BITCAST1]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %res = call { <4 x half>, i32 } @llvm.amdgcn.image.load.2d.sl_v4f16i32s.i32(i32 15, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
define amdgpu_ps <2 x half> @image_load_v2f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
   ; UNPACKED:   [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
   ; UNPACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
define amdgpu_ps <2 x half> @image_load_v2f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %tex = call <2 x half> @llvm.amdgcn.image.load.2d.v2f16.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <2 x half> %tex
 }
 
 define amdgpu_ps <2 x half> @image_load_v2f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED-LABEL: name: image_load_v2f16_dmask_0000
define amdgpu_ps <3 x half> @image_load_v3f16_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <3 x half> @image_load_v3f16_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; PACKED:   [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>), [[DEF]](<2 x s16>), [[DEF]](<2 x s16>)
   ; PACKED:   [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
   ; PACKED:   [[EXTRACT:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
   ; PACKED:   [[EXTRACT1:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 32
   ; PACKED:   $vgpr0 = COPY [[EXTRACT]](<2 x s16>)
define amdgpu_ps <3 x half> @image_load_v3f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
   ; UNPACKED:   [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
   ; UNPACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
define amdgpu_ps <3 x half> @image_load_v3f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>), [[DEF]](<2 x s16>), [[DEF]](<2 x s16>)
   ; PACKED:   [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
   ; PACKED:   [[EXTRACT:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 0
   ; PACKED:   [[EXTRACT1:%[0-9]+]]:_(<2 x s16>) = G_EXTRACT [[INSERT]](<4 x s16>), 32
   ; PACKED:   $vgpr0 = COPY [[EXTRACT]](<2 x s16>)
define amdgpu_ps <4 x half> @image_load_v4f16_dmask_1110(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <4 x half> @image_load_v4f16_dmask_1110(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
   ; PACKED:   [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s16>)
   ; PACKED:   $vgpr0 = COPY [[UV]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[UV1]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %tex = call <4 x half> @llvm.amdgcn.image.load.2d.v4f16.i32(i32 7, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <4 x half> %tex
 }
 
 define amdgpu_ps <4 x half> @image_load_v4f16_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED-LABEL: name: image_load_v4f16_dmask_1100
   ; UNPACKED: bb.1 (%ir-block.0):
   ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; UNPACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; UNPACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; UNPACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; UNPACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <4 x half> @image_load_v4f16_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; PACKED:   [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[DEF]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %tex = call <4 x half> @llvm.amdgcn.image.load.2d.v4f16.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <4 x half> %tex
 }
 
 define amdgpu_ps <4 x half> @image_load_v4f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED-LABEL: name: image_load_v4f16_dmask_1000
   ; UNPACKED: bb.1 (%ir-block.0):
   ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; UNPACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; UNPACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; UNPACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; UNPACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; UNPACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
   ; UNPACKED:   [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
   ; UNPACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
define amdgpu_ps <4 x half> @image_load_v4f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s16>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[DEF]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %tex = call <4 x half> @llvm.amdgcn.image.load.2d.v4f16.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <4 x half> %tex
 }
 
define amdgpu_ps half @image_load_tfe_f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   $vgpr0 = COPY [[COPY10]](s32)
   ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; PACKED-LABEL: name: image_load_tfe_f16_dmask_0000
   ; PACKED: bb.1 (%ir-block.0):
   ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; PACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; PACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; PACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; PACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; PACKED:   $vgpr0 = COPY [[COPY10]](s32)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %res = call { half, i32 } @llvm.amdgcn.image.load.2d.sl_f16i32s.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { half, i32 } %res, 0
   %tfe = extractvalue { half, i32 } %res, 1
define amdgpu_ps <2 x half> @image_load_tfe_v2f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
   ; UNPACKED:   [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
define amdgpu_ps <2 x half> @image_load_tfe_v2f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   $vgpr0 = COPY [[BITCAST]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %res = call { <2 x half>, i32 } @llvm.amdgcn.image.load.2d.sl_v2f16i32s.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { <2 x half>, i32 } %res, 0
   %tfe = extractvalue { <2 x half>, i32 } %res, 1
define amdgpu_ps <2 x half> @image_load_tfe_v2f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
   ; UNPACKED:   [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
define amdgpu_ps <2 x half> @image_load_tfe_v2f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   $vgpr0 = COPY [[BITCAST]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %res = call { <2 x half>, i32 } @llvm.amdgcn.image.load.2d.sl_v2f16i32s.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { <2 x half>, i32 } %res, 0
   %tfe = extractvalue { <2 x half>, i32 } %res, 1
define amdgpu_ps <3 x half> @image_load_tfe_v3f16_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
define amdgpu_ps <3 x half> @image_load_tfe_v3f16_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[DEF1]](<2 x s16>), [[DEF1]](<2 x s16>)
   ; PACKED:   [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[UV2]](<3 x s16>), 0
define amdgpu_ps <3 x half> @image_load_tfe_v3f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
   ; UNPACKED:   [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
define amdgpu_ps <3 x half> @image_load_tfe_v3f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[DEF1]](<2 x s16>), [[DEF1]](<2 x s16>)
   ; PACKED:   [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[UV2]](<3 x s16>), 0
define amdgpu_ps <3 x half> @image_load_tfe_v3f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
   ; UNPACKED:   [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
define amdgpu_ps <3 x half> @image_load_tfe_v3f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[DEF1]](<2 x s16>), [[DEF1]](<2 x s16>)
   ; PACKED:   [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[UV2]](<3 x s16>), 0
define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_1110(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; UNPACKED:   G_STORE [[UV3]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_1110(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 6 from custom "TargetCustom8", align 8)
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV1]](s32)
   ; PACKED:   G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   $vgpr0 = COPY [[BITCAST]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[BITCAST1]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %res = call { <4 x half>, i32 } @llvm.amdgcn.image.load.2d.sl_v4f16i32s.i32(i32 7, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; UNPACKED:   G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
   ; UNPACKED:   [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   $vgpr0 = COPY [[BITCAST]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[DEF1]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %res = call { <4 x half>, i32 } @llvm.amdgcn.image.load.2d.sl_v4f16i32s.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   $vgpr0 = COPY [[BITCAST]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[DEF1]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %res = call { <4 x half>, i32 } @llvm.amdgcn.image.load.2d.sl_v4f16i32s.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; UNPACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; UNPACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; UNPACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; UNPACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
   ; UNPACKED:   [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C]]
   ; UNPACKED:   [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
   ; UNPACKED:   [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[SHL:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C2]](s32)
define amdgpu_ps <4 x half> @image_load_tfe_v4f16_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
+  ; PACKED:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 2 from custom "TargetCustom8")
   ; PACKED:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; PACKED:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[UV]](s32)
   ; PACKED:   [[DEF1:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; PACKED:   $vgpr0 = COPY [[BITCAST]](<2 x s16>)
   ; PACKED:   $vgpr1 = COPY [[DEF1]](<2 x s16>)
   ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %res = call { <4 x half>, i32 } @llvm.amdgcn.image.load.2d.sl_v4f16i32s.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2d.ll

define amdgpu_ps float @image_load_f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %tex = call float @llvm.amdgcn.image.load.2d.f32.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret float %tex
 }
 
 define amdgpu_ps <2 x float> @image_load_v2f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN-LABEL: name: image_load_v2f32
   ; GCN: bb.1 (%ir-block.0):
   ; GCN:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GCN:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GCN:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GCN:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GCN:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %tex = call <2 x float> @llvm.amdgcn.image.load.2d.v2f32.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <2 x float> %tex
 }
 
 define amdgpu_ps <3 x float> @image_load_v3f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN-LABEL: name: image_load_v3f32
   ; GCN: bb.1 (%ir-block.0):
   ; GCN:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1
   ; GCN:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GCN:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GCN:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GCN:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[UV2]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
   %tex = call <3 x float> @llvm.amdgcn.image.load.2d.v3f32.i32(i32 7, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <3 x float> %tex
 }
define amdgpu_ps <4 x float> @image_load_v4f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[UV2]](s32)
   ; GCN:   $vgpr3 = COPY [[UV3]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   %tex = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 15, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <4 x float> %tex
define amdgpu_ps float @image_load_tfe_f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %res = call { float, i32 } @llvm.amdgcn.image.load.2d.sl_f32i32s.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { float, i32 } %res, 0
   %tfe = extractvalue { float, i32 } %res, 1
   store i32 %tfe, i32 addrspace(1)* undef
define amdgpu_ps <2 x float> @image_load_tfe_v2f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; GCN:   G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %res = call { <2 x float>, i32 } @llvm.amdgcn.image.load.2d.sl_v2f32i32s.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { <2 x float>, i32 } %res, 0
   %tfe = extractvalue { <2 x float>, i32 } %res, 1
define amdgpu_ps <3 x float> @image_load_tfe_v3f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GCN:   G_STORE [[UV3]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[UV2]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
   %res = call { <3 x float>, i32 } @llvm.amdgcn.image.load.2d.sl_v3f32i32s.i32(i32 7, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { <3 x float>, i32 } %res, 0
define amdgpu_ps <4 x float> @image_load_tfe_v4f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GCN:   G_STORE [[UV4]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[UV2]](s32)
   ; GCN:   $vgpr3 = COPY [[UV3]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   %res = call { <4 x float>, i32 } @llvm.amdgcn.image.load.2d.sl_v4f32i32s.i32(i32 15, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
define amdgpu_ps <2 x float> @image_load_v2f32_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %tex = call <2 x float> @llvm.amdgcn.image.load.2d.v2f32.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <2 x float> %tex
 }
 
define amdgpu_ps <3 x float> @image_load_v3f32_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
   %tex = call <3 x float> @llvm.amdgcn.image.load.2d.v3f32.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <3 x float> %tex
define amdgpu_ps <3 x float> @image_load_v3f32_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
   %tex = call <3 x float> @llvm.amdgcn.image.load.2d.v3f32.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <3 x float> %tex
 }
define amdgpu_ps <4 x float> @image_load_v4f32_dmask_1110(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; GCN:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[UV2]](s32)
   ; GCN:   $vgpr3 = COPY [[DEF]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   %tex = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 7, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
define amdgpu_ps <4 x float> @image_load_v4f32_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF]](s32)
   ; GCN:   $vgpr3 = COPY [[DEF]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   %tex = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
define amdgpu_ps <4 x float> @image_load_v4f32_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF]](s32)
   ; GCN:   $vgpr3 = COPY [[DEF]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   %tex = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret <4 x float> %tex
define amdgpu_ps float @image_load_tfe_f32_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %res = call { float, i32 } @llvm.amdgcn.image.load.2d.sl_f32i32s.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { float, i32 } %res, 0
   %tfe = extractvalue { float, i32 } %res, 1
   store i32 %tfe, i32 addrspace(1)* undef
define amdgpu_ps <2 x float> @image_load_tfe_v2f32_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %res = call { <2 x float>, i32 } @llvm.amdgcn.image.load.2d.sl_v2f32i32s.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { <2 x float>, i32 } %res, 0
define amdgpu_ps <2 x float> @image_load_tfe_v2f32_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
   %res = call { <2 x float>, i32 } @llvm.amdgcn.image.load.2d.sl_v2f32i32s.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { <2 x float>, i32 } %res, 0
define amdgpu_ps <3 x float> @image_load_tfe_v3f32_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
   %res = call { <3 x float>, i32 } @llvm.amdgcn.image.load.2d.sl_v3f32i32s.i32(i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
define amdgpu_ps <3 x float> @image_load_tfe_v3f32_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF1]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
   %res = call { <3 x float>, i32 } @llvm.amdgcn.image.load.2d.sl_v3f32i32s.i32(i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
define amdgpu_ps <3 x float> @image_load_tfe_v3f32_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF1]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
   %res = call { <3 x float>, i32 } @llvm.amdgcn.image.load.2d.sl_v3f32i32s.i32(i32 0, i32 %s, i32 %t, <8 x i32> %rsrc, i32 1, i32 0)
define amdgpu_ps <4 x float> @image_load_tfe_v4f32_dmask_1110(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 12 from custom "TargetCustom8", align 16)
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV3]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[UV2]](s32)
   ; GCN:   $vgpr3 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
define amdgpu_ps <4 x float> @image_load_tfe_v4f32_dmask_1100(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<3 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 8 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<3 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV2]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[UV1]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF1]](s32)
   ; GCN:   $vgpr3 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
define amdgpu_ps <4 x float> @image_load_tfe_v4f32_dmask_1000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF1]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF1]](s32)
   ; GCN:   $vgpr3 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
define amdgpu_ps <4 x float> @image_load_tfe_v4f32_dmask_0000(<8 x i32> inreg %rsrc, i32 %s, i32 %t) {
   ; GCN:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GCN:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GCN:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GCN:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GCN:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GCN:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GCN:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GCN:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GCN:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2d), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GCN:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GCN:   [[DEF1:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
   ; GCN:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GCN:   $vgpr0 = COPY [[UV]](s32)
   ; GCN:   $vgpr1 = COPY [[DEF1]](s32)
   ; GCN:   $vgpr2 = COPY [[DEF1]](s32)
   ; GCN:   $vgpr3 = COPY [[DEF1]](s32)
   ; GCN:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.2darraymsaa.ll

define amdgpu_ps <4 x float> @load_2darraymsaa(<8 x i32> inreg %rsrc, i32 %s, i32 %t, i32 %slice, i32 %fragid) {
   ; GFX6:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX6:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX6:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX6:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX6:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX6:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX6:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX6:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
-  ; GFX6:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR1]](<4 x s32>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX6:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR1]](<4 x s32>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX6:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX6:   $vgpr0 = COPY [[UV]](s32)
   ; GFX6:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX6:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX6:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX6:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_2darraymsaa
   ; GFX10NSA: bb.1 (%ir-block.0):
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   %v = call <4 x float> @llvm.amdgcn.image.load.2darraymsaa.v4f32.i32(i32 15, i32 %s, i32 %t, i32 %slice, i32 %fragid, <8 x i32> %rsrc, i32 0, i32 0)
   ret <4 x float> %v
define amdgpu_ps <4 x float> @load_2darraymsaa_tfe(<8 x i32> inreg %rsrc, i32 addrspace(1)* inreg %out, i32 %s, i32 %t, i32 %slice, i32 %fragid) {
   ; GFX6:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX6:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX6:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX6:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX6:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX6:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX6:   [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
   ; GFX6:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32), [[COPY13]](s32)
-  ; GFX6:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR1]](<4 x s32>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX6:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[BUILD_VECTOR1]](<4 x s32>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX6:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX6:   G_STORE [[UV4]](s32), [[MV]](p1) :: (store 4 into %ir.out, addrspace 1)
   ; GFX6:   $vgpr0 = COPY [[UV]](s32)
   ; GFX6:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX6:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX6:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX6:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   ; GFX10NSA-LABEL: name: load_2darraymsaa_tfe
define amdgpu_ps <4 x float> @load_2darraymsaa_tfe(<8 x i32> inreg %rsrc, i32 addrspace(1)* inreg %out, i32 %s, i32 %t, i32 %slice, i32 %fragid) {
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY11:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr3
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[MV:%[0-9]+]]:_(p1) = G_MERGE_VALUES [[COPY8]](s32), [[COPY9]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32), [[COPY13]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<5 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.2darraymsaa), 15, [[COPY10]](s32), [[COPY11]](s32), [[COPY12]](s32), [[COPY13]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<5 x s32>)
   ; GFX10NSA:   G_STORE [[UV4]](s32), [[MV]](p1) :: (store 4 into %ir.out, addrspace 1)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   $vgpr1 = COPY [[UV1]](s32)
   ; GFX10NSA:   $vgpr2 = COPY [[UV2]](s32)
   ; GFX10NSA:   $vgpr3 = COPY [[UV3]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
   %v = call { <4 x float>, i32 } @llvm.amdgcn.image.load.2darraymsaa.sl_v4f32i32s.i32(i32 15, i32 %s, i32 %t, i32 %slice, i32 %fragid, <8 x i32> %rsrc, i32 1, i32 0)

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.load.3d.ll

define amdgpu_ps float @image_load_3d_f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t, i32 %r) {
   ; GFX6:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX6:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX6:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX6:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX6:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX6:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX6:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX6:   [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX6:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[BUILD_VECTOR1]](<3 x s32>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX6:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[BUILD_VECTOR1]](<3 x s32>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GFX6:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX6:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: image_load_3d_f32
   ; GFX10NSA: bb.1 (%ir-block.0):
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GFX10NSA:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %tex = call float @llvm.amdgcn.image.load.3d.f32.i32(i32 1, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 0, i32 0)
   ret float %tex
 }
 
 define amdgpu_ps float @image_load_3d_tfe_f32(<8 x i32> inreg %rsrc, i32 %s, i32 %t, i32 %r) {
   ; GFX6-LABEL: name: image_load_3d_tfe_f32
   ; GFX6: bb.1 (%ir-block.0):
   ; GFX6:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; GFX6:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX6:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX6:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX6:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX6:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX6:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX6:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX6:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX6:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX6:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX6:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX6:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX6:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
   ; GFX6:   [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32)
-  ; GFX6:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[BUILD_VECTOR1]](<3 x s32>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX6:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[BUILD_VECTOR1]](<3 x s32>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GFX6:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GFX6:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX6:   $vgpr0 = COPY [[UV]](s32)
   ; GFX6:   SI_RETURN_TO_EPILOG implicit $vgpr0
   ; GFX10NSA-LABEL: name: image_load_3d_tfe_f32
   ; GFX10NSA: bb.1 (%ir-block.0):
   ; GFX10NSA:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; GFX10NSA:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; GFX10NSA:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; GFX10NSA:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; GFX10NSA:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; GFX10NSA:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; GFX10NSA:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; GFX10NSA:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; GFX10NSA:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; GFX10NSA:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; GFX10NSA:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; GFX10NSA:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; GFX10NSA:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; GFX10NSA:   [[DEF:%[0-9]+]]:_(p1) = G_IMPLICIT_DEF
-  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX10NSA:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.3d), 1, [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[BUILD_VECTOR]](<8 x s32>), 1, 0, 0 :: (dereferenceable load 4 from custom "TargetCustom8")
   ; GFX10NSA:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
   ; GFX10NSA:   G_STORE [[UV1]](s32), [[DEF]](p1) :: (store 4 into `i32 addrspace(1)* undef`, addrspace 1)
   ; GFX10NSA:   $vgpr0 = COPY [[UV]](s32)
   ; GFX10NSA:   SI_RETURN_TO_EPILOG implicit $vgpr0
   %val = call { float, i32 } @llvm.amdgcn.image.load.3d.sl_f32i32s.i32(i32 1, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 1, i32 0)
   %tex = extractvalue { float, i32 } %val, 0
   %tfe = extractvalue { float, i32 } %val, 1
   store i32 %tfe, i32 addrspace(1)* undef
   ret float %tex
 }
 
 declare float @llvm.amdgcn.image.load.3d.f32.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0
 declare { float, i32 } @llvm.amdgcn.image.load.3d.sl_f32i32s.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0
 
 attributes #0 = { nounwind readonly }

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.sample.a16.ll

+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX9 %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX10 %s
+
+define amdgpu_ps <4 x float> @sample_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s) {
+  ; GFX9-LABEL: name: sample_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY12]](s32)
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY12]](s32)
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f16(i32 15, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f16(i32 15, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %r) {
+  ; GFX9-LABEL: name: sample_3d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_3d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.3d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f16(i32 15, half %s, half %t, half %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cube(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %face) {
+  ; GFX9-LABEL: name: sample_cube
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_cube
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cube), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cube.v4f32.f16(i32 15, half %s, half %t, half %face, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_1darray(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %slice) {
+  ; GFX9-LABEL: name: sample_1darray
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_1darray
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1darray), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.1darray.v4f32.f16(i32 15, half %s, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_2darray(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %slice) {
+  ; GFX9-LABEL: name: sample_2darray
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_2darray
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.2darray), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.2darray.v4f32.f16(i32 15, half %s, half %t, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s) {
+  ; GFX9-LABEL: name: sample_c_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.1d.v4f32.f16(i32 15, float %zcompare, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_c_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %clamp) {
+  ; GFX9-LABEL: name: sample_cl_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cl.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cl.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cl.1d.v4f32.f16(i32 15, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %clamp) {
+  ; GFX9-LABEL: name: sample_cl_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cl.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cl.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cl.2d.v4f32.f16(i32 15, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %clamp) {
+  ; GFX9-LABEL: name: sample_c_cl_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cl.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cl.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cl.1d.v4f32.f16(i32 15, float %zcompare, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t, half %clamp) {
+  ; GFX9-LABEL: name: sample_c_cl_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cl.2d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s) {
+  ; GFX9-LABEL: name: sample_b_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_b_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f32.f16(i32 15, float %bias, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_b_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_b_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f32.f16(i32 15, float %bias, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s) {
+  ; GFX9-LABEL: name: sample_c_b_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.1d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_b_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.1d), 15, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_c_b_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.2d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_b_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.2d), 15, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %clamp) {
+  ; GFX9-LABEL: name: sample_b_cl_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_b_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f32.f16(i32 15, float %bias, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, half %s, half %t, half %clamp) {
+  ; GFX9-LABEL: name: sample_b_cl_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.2d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_b_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.b.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f32.f16(i32 15, float %bias, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %clamp) {
+  ; GFX9-LABEL: name: sample_c_b_cl_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.1d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_b_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.1d), 15, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, half %s, half %t, half %clamp) {
+  ; GFX9-LABEL: name: sample_c_b_cl_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[COPY18]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.2d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_b_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[COPY18]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.b.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f32.f16(i32 15, float %bias, float %zcompare, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, half %s) {
+  ; GFX9-LABEL: name: sample_d_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.1d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_d_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f16(i32 15, half %dsdh, half %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_d_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.2d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_d_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, half %s, half %t, half %r) {
+  ; GFX9-LABEL: name: sample_d_3d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY24]](s32), [[COPY25]](s32)
+  ; GFX9:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY26]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY27:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[COPY28:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC4:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY27]](s32), [[COPY28]](s32)
+  ; GFX9:   [[COPY29:%[0-9]+]]:_(s32) = COPY [[COPY20]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC5:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY29]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), [[BUILD_VECTOR_TRUNC4]](<2 x s16>), [[BUILD_VECTOR_TRUNC5]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.3d), 15, [[CONCAT_VECTORS]](<12 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_d_3d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY24]](s32), [[COPY25]](s32)
+  ; GFX10:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY26]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY27:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[COPY28:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC4:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY27]](s32), [[COPY28]](s32)
+  ; GFX10:   [[COPY29:%[0-9]+]]:_(s32) = COPY [[COPY20]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC5:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY29]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), [[BUILD_VECTOR_TRUNC4]](<2 x s16>), [[BUILD_VECTOR_TRUNC5]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, half %s, half %t, half %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, half %s) {
+  ; GFX9-LABEL: name: sample_c_d_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.1d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_d_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.1d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_c_d_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.2d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_d_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, half %s, half %clamp) {
+  ; GFX9-LABEL: name: sample_d_cl_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.cl.1d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_d_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.cl.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f16(i32 15, half %dsdh, half %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp) {
+  ; GFX9-LABEL: name: sample_d_cl_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX9:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY25]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.cl.2d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_d_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX10:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY25]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.cl.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, half %s, half %clamp) {
+  ; GFX9-LABEL: name: sample_c_d_cl_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.cl.1d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_d_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.cl.1d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp) {
+  ; GFX9-LABEL: name: sample_c_d_cl_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY24]](s32), [[COPY25]](s32)
+  ; GFX9:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY26]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<10 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.cl.2d), 15, [[CONCAT_VECTORS]](<10 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_d_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY24]](s32), [[COPY25]](s32)
+  ; GFX10:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY26]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, half %s) {
+  ; GFX9-LABEL: name: sample_cd_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.1d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_cd_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f16(i32 15, half %dsdh, half %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_cd_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.2d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_cd_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, half %s) {
+  ; GFX9-LABEL: name: sample_c_cd_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.1d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_cd_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.1d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dsdv, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_c_cd_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.2d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_cd_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, half %s, half %clamp) {
+  ; GFX9-LABEL: name: sample_cd_cl_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.cl.1d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_cd_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.cl.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f16(i32 15, half %dsdh, half %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp) {
+  ; GFX9-LABEL: name: sample_cd_cl_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX9:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY25]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.cl.2d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_cd_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX10:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY25]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.cl.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f16(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, half %s, half %clamp) {
+  ; GFX9-LABEL: name: sample_c_cd_cl_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.cl.1d), 15, [[CONCAT_VECTORS]](<8 x s16>), $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_cd_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.cl.1d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dsdv, half %s, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp) {
+  ; GFX9-LABEL: name: sample_c_cd_cl_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY24]](s32), [[COPY25]](s32)
+  ; GFX9:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY26]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<10 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.cl.2d), 15, [[CONCAT_VECTORS]](<10 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_cd_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY24]](s32), [[COPY25]](s32)
+  ; GFX10:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY26]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f32.f16(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_l_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %lod) {
+  ; GFX9-LABEL: name: sample_l_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.l.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_l_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.l.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.l.1d.v4f32.f16(i32 15, half %s, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t, half %lod) {
+  ; GFX9-LABEL: name: sample_l_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.l.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_l_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.l.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.l.2d.v4f32.f16(i32 15, half %s, half %t, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_l_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %lod) {
+  ; GFX9-LABEL: name: sample_c_l_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.l.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_l_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.l.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.l.1d.v4f32.f16(i32 15, float %zcompare, half %s, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t, half %lod) {
+  ; GFX9-LABEL: name: sample_c_l_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.l.2d), 15, [[CONCAT_VECTORS]](<6 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_l_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[COPY17]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.l.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.l.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, half %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_lz_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s) {
+  ; GFX9-LABEL: name: sample_lz_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY12]](s32)
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.lz.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_lz_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY12]](s32)
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.lz.1d), 15, [[TRUNC]](s16), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.lz.1d.v4f32.f16(i32 15, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_lz_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_lz_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.lz.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_lz_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[COPY15]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.lz.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.lz.2d.v4f32.f16(i32 15, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_lz_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s) {
+  ; GFX9-LABEL: name: sample_c_lz_1d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.lz.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_lz_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY14]](s32), [[DEF]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.lz.1d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.lz.1d.v4f32.f16(i32 15, float %zcompare, half %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_lz_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %s, half %t) {
+  ; GFX9-LABEL: name: sample_c_lz_2d
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.lz.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX9:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  ; GFX10-LABEL: name: sample_c_lz_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[COPY16]](s32)
+  ; GFX10:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.lz.2d), 15, [[CONCAT_VECTORS]](<4 x s16>), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.lz.2d.v4f32.f16(i32 15, float %zcompare, half %s, half %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %slice) {
+  ; GFX9-LABEL: name: sample_c_d_o_2darray_V1
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX9:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY25]](s32), [[COPY26]](s32)
+  ; GFX9:   [[COPY27:%[0-9]+]]:_(s32) = COPY [[COPY20]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY27]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.o.2darray), 4, [[CONCAT_VECTORS]](<12 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX9:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  ; GFX10-LABEL: name: sample_c_d_o_2darray_V1
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX10:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY25]](s32), [[COPY26]](s32)
+  ; GFX10:   [[COPY27:%[0-9]+]]:_(s32) = COPY [[COPY20]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY27]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.o.2darray), 4, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX10:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0
+main_body:
+  %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f32.f16.f16(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret float %v
+}
+
+define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %slice) {
+  ; GFX9-LABEL: name: sample_c_d_o_2darray_V2
+  ; GFX9: bb.1.main_body:
+  ; GFX9:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX9:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX9:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX9:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX9:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX9:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX9:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX9:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX9:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX9:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX9:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX9:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX9:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX9:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX9:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX9:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX9:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX9:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX9:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX9:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX9:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX9:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX9:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX9:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX9:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX9:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX9:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX9:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX9:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX9:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX9:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX9:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX9:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY25]](s32), [[COPY26]](s32)
+  ; GFX9:   [[COPY27:%[0-9]+]]:_(s32) = COPY [[COPY20]](s32)
+  ; GFX9:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX9:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY27]](s32), [[DEF]](s32)
+  ; GFX9:   [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>)
+  ; GFX9:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.o.2darray), 6, [[CONCAT_VECTORS]](<12 x s16>), $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GFX9:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
+  ; GFX9:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX9:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX9:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
+  ; GFX10-LABEL: name: sample_c_d_o_2darray_V2
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX10:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY18]](s32)
+  ; GFX10:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY19]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY25]](s32), [[COPY26]](s32)
+  ; GFX10:   [[COPY27:%[0-9]+]]:_(s32) = COPY [[COPY20]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY27]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.o.2darray), 6, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), $noreg, $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 3 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
+main_body:
+  %v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f32.f16(i32 6, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, half %s, half %t, half %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <2 x float> %v
+}
+
+declare <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f16(i32, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <8 x float> @llvm.amdgcn.image.sample.1d.v8f32.f16(i32, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cube.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.1darray.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.2darray.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.c.1d.v4f32.f16(i32, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.2d.v4f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cl.1d.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cl.2d.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cl.1d.v4f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f32.f16(i32, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f32.f16(i32, float, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f32.f16(i32, float, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f32.f16(i32, float, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f32.f16(i32, float, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f16(i32, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f16(i32, half, half, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f32.f16(i32, float, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f16(i32, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f16(i32, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f32.f16(i32, float, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f32.f16(i32, float, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f16(i32, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f32.f16(i32, float, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f16(i32, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f16(i32, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f32.f16(i32, float, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f32.f16(i32, float, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.l.1d.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.l.2d.v4f32.f16(i32, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.l.1d.v4f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.l.2d.v4f32.f16(i32, float, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.lz.1d.v4f32.f16(i32, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.lz.2d.v4f32.f16(i32, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.lz.1d.v4f32.f16(i32, float, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.lz.2d.v4f32.f16(i32, float, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare float @llvm.amdgcn.image.sample.c.d.o.2darray.f32.f16.f16(i32, i32, float, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f32.f16(i32, i32, float, half, half, half, half, half, half, half, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readonly }
+attributes #2 = { nounwind readnone }

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.sample.g16.ll

+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -stop-after=legalizer -o - %s | FileCheck -check-prefix=GFX10 %s
+
+define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+  ; GFX10-LABEL: name: sample_d_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY14]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+  ; GFX10-LABEL: name: sample_d_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY16]](s32), [[COPY17]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r) {
+  ; GFX10-LABEL: name: sample_d_3d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY25:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY24]](s32), [[COPY25]](s32)
+  ; GFX10:   [[COPY26:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY26]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.3d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[BUILD_VECTOR_TRUNC2]](<2 x s16>), [[BUILD_VECTOR_TRUNC3]](<2 x s16>), [[COPY18]](s32), [[COPY19]](s32), [[COPY20]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
+  ; GFX10-LABEL: name: sample_c_d_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.1d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY15]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+  ; GFX10-LABEL: name: sample_c_d_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY17]](s32), [[COPY18]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
+  ; GFX10-LABEL: name: sample_d_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.cl.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY14]](s32), [[COPY15]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+  ; GFX10-LABEL: name: sample_d_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.d.cl.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY16]](s32), [[COPY17]](s32), [[COPY18]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp) {
+  ; GFX10-LABEL: name: sample_c_d_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.cl.1d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY15]](s32), [[COPY16]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+  ; GFX10-LABEL: name: sample_c_d_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY17]](s32), [[COPY18]](s32), [[COPY19]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+  ; GFX10-LABEL: name: sample_cd_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY15]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY14]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+  ; GFX10-LABEL: name: sample_cd_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[COPY19]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY16]](s32), [[COPY17]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
+  ; GFX10-LABEL: name: sample_c_cd_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.1d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY15]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+  ; GFX10-LABEL: name: sample_c_cd_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY17]](s32), [[COPY18]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
+  ; GFX10-LABEL: name: sample_cd_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY16]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.cl.1d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY14]](s32), [[COPY15]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+  ; GFX10-LABEL: name: sample_cd_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY [[COPY12]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY19]](s32), [[COPY20]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.cd.cl.2d), 15, [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY16]](s32), [[COPY17]](s32), [[COPY18]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp) {
+  ; GFX10-LABEL: name: sample_c_cd_cl_1d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY17]](s32), [[DEF]](s32)
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY18]](s32), [[DEF]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.cl.1d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY15]](s32), [[COPY16]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+  ; GFX10-LABEL: name: sample_c_cd_cl_2d
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY [[COPY13]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY20]](s32), [[COPY21]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY22]](s32), [[COPY23]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.cd.cl.2d), 15, [[BITCAST]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY17]](s32), [[COPY18]](s32), [[COPY19]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   $vgpr2 = COPY [[UV2]](s32)
+  ; GFX10:   $vgpr3 = COPY [[UV3]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
+  ; GFX10-LABEL: name: sample_c_d_o_2darray_V1
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(s32) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.o.2darray), 4, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY18]](s32), [[COPY19]](s32), [[COPY20]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 4 from custom "TargetCustom8")
+  ; GFX10:   $vgpr0 = COPY [[AMDGPU_INTRIN_IMAGE_LOAD]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0
+main_body:
+  %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret float %v
+}
+
+define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
+  ; GFX10-LABEL: name: sample_c_d_o_2darray_V2
+  ; GFX10: bb.1.main_body:
+  ; GFX10:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GFX10:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
+  ; GFX10:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
+  ; GFX10:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
+  ; GFX10:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
+  ; GFX10:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
+  ; GFX10:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
+  ; GFX10:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
+  ; GFX10:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
+  ; GFX10:   [[COPY8:%[0-9]+]]:_(s32) = COPY $sgpr10
+  ; GFX10:   [[COPY9:%[0-9]+]]:_(s32) = COPY $sgpr11
+  ; GFX10:   [[COPY10:%[0-9]+]]:_(s32) = COPY $sgpr12
+  ; GFX10:   [[COPY11:%[0-9]+]]:_(s32) = COPY $sgpr13
+  ; GFX10:   [[COPY12:%[0-9]+]]:_(s32) = COPY $vgpr0
+  ; GFX10:   [[COPY13:%[0-9]+]]:_(s32) = COPY $vgpr1
+  ; GFX10:   [[COPY14:%[0-9]+]]:_(s32) = COPY $vgpr2
+  ; GFX10:   [[COPY15:%[0-9]+]]:_(s32) = COPY $vgpr3
+  ; GFX10:   [[COPY16:%[0-9]+]]:_(s32) = COPY $vgpr4
+  ; GFX10:   [[COPY17:%[0-9]+]]:_(s32) = COPY $vgpr5
+  ; GFX10:   [[COPY18:%[0-9]+]]:_(s32) = COPY $vgpr6
+  ; GFX10:   [[COPY19:%[0-9]+]]:_(s32) = COPY $vgpr7
+  ; GFX10:   [[COPY20:%[0-9]+]]:_(s32) = COPY $vgpr8
+  ; GFX10:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GFX10:   [[BUILD_VECTOR1:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GFX10:   [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY12]](s32)
+  ; GFX10:   [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[COPY13]](s32)
+  ; GFX10:   [[COPY21:%[0-9]+]]:_(s32) = COPY [[COPY14]](s32)
+  ; GFX10:   [[COPY22:%[0-9]+]]:_(s32) = COPY [[COPY15]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY21]](s32), [[COPY22]](s32)
+  ; GFX10:   [[COPY23:%[0-9]+]]:_(s32) = COPY [[COPY16]](s32)
+  ; GFX10:   [[COPY24:%[0-9]+]]:_(s32) = COPY [[COPY17]](s32)
+  ; GFX10:   [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[COPY23]](s32), [[COPY24]](s32)
+  ; GFX10:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:_(<2 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.c.d.o.2darray), 6, [[BITCAST]](<2 x s16>), [[BITCAST1]](<2 x s16>), [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[COPY18]](s32), [[COPY19]](s32), [[COPY20]](s32), $noreg, $noreg, [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 2 :: (dereferenceable load 8 from custom "TargetCustom8")
+  ; GFX10:   [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[AMDGPU_INTRIN_IMAGE_LOAD]](<2 x s32>)
+  ; GFX10:   $vgpr0 = COPY [[UV]](s32)
+  ; GFX10:   $vgpr1 = COPY [[UV1]](s32)
+  ; GFX10:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
+main_body:
+  %v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32 6, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <2 x float> %v
+}
+
+declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32, half, half, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32, float, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32, float, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32, float, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32, float, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32, i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32, i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readonly }
+attributes #2 = { nounwind readnone }

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-llvm.amdgcn.image.store.2d.d16.ll

define amdgpu_ps void @image_store_f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t, half %data) {
   ; UNPACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; UNPACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; UNPACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; UNPACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; UNPACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; UNPACKED:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY10]](s32)
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; UNPACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[TRUNC]](s16), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 2 into custom "TargetCustom8")
+  ; UNPACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[TRUNC]](s16), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable store 2 into custom "TargetCustom8")
   ; UNPACKED:   S_ENDPGM 0
   ; PACKED-LABEL: name: image_store_f16
   ; PACKED: bb.1 (%ir-block.0):
   ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; PACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; PACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; PACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; PACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[COPY10:%[0-9]+]]:_(s32) = COPY $vgpr2
   ; PACKED:   [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY10]](s32)
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[TRUNC]](s16), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 2 into custom "TargetCustom8")
+  ; PACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[TRUNC]](s16), 1, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable store 2 into custom "TargetCustom8")
   ; PACKED:   S_ENDPGM 0
   call void @llvm.amdgcn.image.store.2d.f16.i32(half %data, i32 1, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @image_store_v2f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t, <2 x half> %in) {
   ; UNPACKED-LABEL: name: image_store_v2f16
   ; UNPACKED: bb.1 (%ir-block.0):
define amdgpu_ps void @image_store_v2f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t, <2 x half> %in) {
   ; UNPACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; UNPACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
   ; UNPACKED:   [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY10]](<2 x s16>)
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
   ; UNPACKED:   [[COPY11:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; UNPACKED:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; UNPACKED:   [[BUILD_VECTOR2:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY11]](s32), [[COPY12]](s32)
-  ; UNPACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<2 x s32>), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8")
+  ; UNPACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<2 x s32>), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8")
   ; UNPACKED:   S_ENDPGM 0
   ; PACKED-LABEL: name: image_store_v2f16
   ; PACKED: bb.1 (%ir-block.0):
   ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2
   ; PACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; PACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; PACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; PACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[COPY10:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[COPY10]](<2 x s16>), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8")
+  ; PACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[COPY10]](<2 x s16>), 3, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable store 4 into custom "TargetCustom8")
   ; PACKED:   S_ENDPGM 0
   call void @llvm.amdgcn.image.store.2d.v2f16.i32(<2 x half> %in, i32 3, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @image_store_v3f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t, <3 x half> %in) {
   ; UNPACKED-LABEL: name: image_store_v3f16
   ; UNPACKED: bb.1 (%ir-block.0):
define amdgpu_ps void @image_store_v3f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t, <3 x half> %in) {
   ; UNPACKED:   [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
   ; UNPACKED:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
   ; UNPACKED:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
   ; UNPACKED:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; UNPACKED:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; UNPACKED:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; UNPACKED:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
   ; UNPACKED:   [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY12]](s32), [[COPY13]](s32), [[COPY14]](s32)
-  ; UNPACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<3 x s32>), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 6 into custom "TargetCustom8", align 8)
+  ; UNPACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<3 x s32>), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable store 6 into custom "TargetCustom8", align 8)
   ; UNPACKED:   S_ENDPGM 0
   ; PACKED-LABEL: name: image_store_v3f16
   ; PACKED: bb.1 (%ir-block.0):
   ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; PACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; PACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; PACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; PACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[COPY10:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
   ; PACKED:   [[COPY11:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[DEF:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[COPY10]](<2 x s16>), [[COPY11]](<2 x s16>), [[DEF]](<2 x s16>)
   ; PACKED:   [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<6 x s16>)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[UV]](<3 x s16>), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 6 into custom "TargetCustom8", align 8)
+  ; PACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[UV]](<3 x s16>), 7, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable store 6 into custom "TargetCustom8", align 8)
   ; PACKED:   S_ENDPGM 0
   call void @llvm.amdgcn.image.store.2d.v3f16.i32(<3 x half> %in, i32 7, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 define amdgpu_ps void @image_store_v4f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t, <4 x half> %in) {
   ; UNPACKED-LABEL: name: image_store_v4f16
   ; UNPACKED: bb.1 (%ir-block.0):
define amdgpu_ps void @image_store_v4f16(<8 x i32> inreg %rsrc, i32 %s, i32 %t, <4 x half> %in) {
   ; UNPACKED:   [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
   ; UNPACKED:   [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY11]](<2 x s16>)
   ; UNPACKED:   [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
   ; UNPACKED:   [[COPY12:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
   ; UNPACKED:   [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
   ; UNPACKED:   [[COPY14:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
   ; UNPACKED:   [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
   ; UNPACKED:   [[BUILD_VECTOR2:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[COPY12]](s32), [[COPY13]](s32), [[COPY14]](s32), [[COPY15]](s32)
-  ; UNPACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<4 x s32>), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8")
+  ; UNPACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[BUILD_VECTOR2]](<4 x s32>), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8")
   ; UNPACKED:   S_ENDPGM 0
   ; PACKED-LABEL: name: image_store_v4f16
   ; PACKED: bb.1 (%ir-block.0):
   ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3
   ; PACKED:   [[COPY:%[0-9]+]]:_(s32) = COPY $sgpr2
   ; PACKED:   [[COPY1:%[0-9]+]]:_(s32) = COPY $sgpr3
   ; PACKED:   [[COPY2:%[0-9]+]]:_(s32) = COPY $sgpr4
   ; PACKED:   [[COPY3:%[0-9]+]]:_(s32) = COPY $sgpr5
   ; PACKED:   [[COPY4:%[0-9]+]]:_(s32) = COPY $sgpr6
   ; PACKED:   [[COPY5:%[0-9]+]]:_(s32) = COPY $sgpr7
   ; PACKED:   [[COPY6:%[0-9]+]]:_(s32) = COPY $sgpr8
   ; PACKED:   [[COPY7:%[0-9]+]]:_(s32) = COPY $sgpr9
   ; PACKED:   [[COPY8:%[0-9]+]]:_(s32) = COPY $vgpr0
   ; PACKED:   [[COPY9:%[0-9]+]]:_(s32) = COPY $vgpr1
   ; PACKED:   [[COPY10:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
   ; PACKED:   [[COPY11:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr3
   ; PACKED:   [[BUILD_VECTOR:%[0-9]+]]:_(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
   ; PACKED:   [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[COPY10]](<2 x s16>), [[COPY11]](<2 x s16>)
   ; PACKED:   [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32)
-  ; PACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[CONCAT_VECTORS]](<4 x s16>), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8")
+  ; PACKED:   G_AMDGPU_INTRIN_IMAGE_STORE intrinsic(@llvm.amdgcn.image.store.2d), [[CONCAT_VECTORS]](<4 x s16>), 15, [[BUILD_VECTOR1]](<2 x s32>), $noreg, [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable store 8 into custom "TargetCustom8")
   ; PACKED:   S_ENDPGM 0
   call void @llvm.amdgcn.image.store.2d.v4f16.i32(<4 x half> %in, i32 15, i32 %s, i32 %t, <8 x i32> %rsrc, i32 0, i32 0)
   ret void
 }
 
 declare void @llvm.amdgcn.image.store.2d.f16.i32(half, i32 immarg, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0
 declare void @llvm.amdgcn.image.store.2d.v2f16.i32(<2 x half>, i32 immarg, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0
 declare void @llvm.amdgcn.image.store.2d.v3f16.i32(<3 x half>, i32 immarg, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0
 declare void @llvm.amdgcn.image.store.2d.v4f16.i32(<4 x half>, i32 immarg, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0
 
 attributes #0 = { nounwind writeonly }

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.image.sample.g16.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10 %s
+
+define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_d_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v3, s12
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v3, s12
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_d_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v6, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v3, v2, v6, v3
+; GFX10-NEXT:    v_and_or_b32 v10, v0, v6, v1
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v10, v3, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r) {
+; GFX10-LABEL: sample_d_3d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v11, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX10-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v11, v1
+; GFX10-NEXT:    v_and_or_b32 v1, v2, v11, s12
+; GFX10-NEXT:    v_and_or_b32 v2, v3, v11, v4
+; GFX10-NEXT:    v_and_or_b32 v3, v5, v11, s12
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v0, v1, v2, v3, v6, v7, v8], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_c_d_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v4, s12
+; GFX10-NEXT:    v_and_or_b32 v2, v2, v4, s12
+; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_c_d_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX10-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v7, v2
+; GFX10-NEXT:    v_and_or_b32 v3, v3, v7, v4
+; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_d_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v7, s12
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v7, s12
+; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_d_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v9, 16, v3
+; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v3, v2, v7, v9
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v7, v1
+; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], [v0, v3, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_c_d_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v7, s12
+; GFX10-NEXT:    v_and_or_b32 v2, v2, v7, s12
+; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_c_d_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v8, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX10-NEXT:    v_lshlrev_b32_e32 v10, 16, v4
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v8, v2
+; GFX10-NEXT:    v_and_or_b32 v2, v3, v8, v10
+; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], [v0, v1, v2, v5, v6, v7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_cd_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v3, s12
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v3, s12
+; GFX10-NEXT:    image_sample_cd_g16 v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_cd_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v6, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v3, v2, v6, v3
+; GFX10-NEXT:    v_and_or_b32 v10, v0, v6, v1
+; GFX10-NEXT:    image_sample_cd_g16 v[0:3], [v10, v3, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_c_cd_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v4, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v4, s12
+; GFX10-NEXT:    v_and_or_b32 v2, v2, v4, s12
+; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_c_cd_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX10-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v7, v2
+; GFX10-NEXT:    v_and_or_b32 v3, v3, v7, v4
+; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_cd_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v7, s12
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v7, s12
+; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_cd_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v9, 16, v3
+; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v3, v2, v7, v9
+; GFX10-NEXT:    v_and_or_b32 v0, v0, v7, v1
+; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], [v0, v3, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_c_cd_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    s_lshl_b32 s12, s0, 16
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v7, s12
+; GFX10-NEXT:    v_and_or_b32 v2, v2, v7, s12
+; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_c_cd_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v8, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX10-NEXT:    v_lshlrev_b32_e32 v10, 16, v4
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v1, v1, v8, v2
+; GFX10-NEXT:    v_and_or_b32 v2, v3, v8, v10
+; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], [v0, v1, v2, v5, v6, v7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
+; GFX10-LABEL: sample_c_d_o_2darray_V1:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX10-NEXT:    v_lshlrev_b32_e32 v11, 16, v5
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v2, v2, v9, v3
+; GFX10-NEXT:    v_and_or_b32 v3, v4, v9, v11
+; GFX10-NEXT:    image_sample_c_d_o_g16 v0, [v0, v1, v2, v3, v6, v7, v8], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret float %v
+}
+
+define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
+; GFX10-LABEL: sample_c_d_o_2darray_V2:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX10-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
+; GFX10-NEXT:    v_lshlrev_b32_e32 v11, 16, v5
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_or_b32 v2, v2, v9, v3
+; GFX10-NEXT:    v_and_or_b32 v3, v4, v9, v11
+; GFX10-NEXT:    image_sample_c_d_o_g16 v[0:1], [v0, v1, v2, v3, v6, v7, v8], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32 6, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <2 x float> %v
+}
+
+declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32, half, half, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32, float, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32, float, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32, float, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32, float, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32, i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32, i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readonly }
+attributes #2 = { nounwind readnone }

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.image.load.1d.ll

 ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck %s
-; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck -check-prefix=FAST %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck -check-prefix=GREEDY %s
 
 ; Natural mapping
 define amdgpu_ps void @load_1d_vgpr_vaddr__sgpr_srsrc(<8 x i32> inreg %rsrc, i32 %s) {
-  ; CHECK-LABEL: name: load_1d_vgpr_vaddr__sgpr_srsrc
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
-  ; CHECK:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
-  ; CHECK:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
-  ; CHECK:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
-  ; CHECK:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
-  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: load_1d_vgpr_vaddr__sgpr_srsrc
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
+  ; FAST:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; FAST:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; FAST:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; FAST:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; FAST:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; FAST:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; FAST:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; FAST:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; FAST:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   [[COPY9:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: load_1d_vgpr_vaddr__sgpr_srsrc
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0
+  ; GREEDY:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; GREEDY:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; GREEDY:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; GREEDY:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; GREEDY:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; GREEDY:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; GREEDY:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; GREEDY:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; GREEDY:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   [[COPY9:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 ; Copy needed for VGPR argument
 define amdgpu_ps void @load_1d_sgpr_vaddr__sgpr_srsrc(<8 x i32> inreg %rsrc, i32 inreg %s) {
-  ; CHECK-LABEL: name: load_1d_sgpr_vaddr__sgpr_srsrc
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10
-  ; CHECK:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
-  ; CHECK:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
-  ; CHECK:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
-  ; CHECK:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
-  ; CHECK:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY [[COPY8]](s32)
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   [[COPY10:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: load_1d_sgpr_vaddr__sgpr_srsrc
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10
+  ; FAST:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; FAST:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; FAST:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; FAST:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; FAST:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; FAST:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; FAST:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; FAST:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; FAST:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY [[COPY8]](s32)
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   [[COPY10:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: load_1d_sgpr_vaddr__sgpr_srsrc
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10
+  ; GREEDY:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; GREEDY:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; GREEDY:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; GREEDY:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; GREEDY:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; GREEDY:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; GREEDY:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; GREEDY:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; GREEDY:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY [[COPY8]](s32)
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   [[COPY10:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 ; Waterfall loop needed for rsrc
 define amdgpu_ps void @load_1d_vgpr_vaddr__vgpr_srsrc(<8 x i32> %rsrc, i32 %s) {
-  ; CHECK-LABEL: name: load_1d_vgpr_vaddr__vgpr_srsrc
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   successors: %bb.2(0x80000000)
-  ; CHECK:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
-  ; CHECK:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
-  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
-  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
-  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
-  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
-  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
-  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-  ; CHECK:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
-  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-  ; CHECK: bb.2:
-  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
-  ; CHECK:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %19, %bb.2
-  ; CHECK:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %12(<4 x s32>), %bb.2
-  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
-  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
-  ; CHECK: bb.3:
-  ; CHECK:   successors: %bb.4(0x80000000)
-  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-  ; CHECK: bb.4:
-  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: load_1d_vgpr_vaddr__vgpr_srsrc
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   successors: %bb.2(0x80000000)
+  ; FAST:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; FAST:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; FAST:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; FAST:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; FAST:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; FAST:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; FAST:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+  ; FAST:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+  ; FAST:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
+  ; FAST:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; FAST:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
+  ; FAST:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; FAST: bb.2:
+  ; FAST:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; FAST:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %19, %bb.2
+  ; FAST:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %12(<4 x s32>), %bb.2
+  ; FAST:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; FAST:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; FAST:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
+  ; FAST:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
+  ; FAST:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
+  ; FAST:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; FAST:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; FAST:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; FAST: bb.3:
+  ; FAST:   successors: %bb.4(0x80000000)
+  ; FAST:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; FAST: bb.4:
+  ; FAST:   [[COPY9:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: load_1d_vgpr_vaddr__vgpr_srsrc
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   successors: %bb.2(0x80000000)
+  ; GREEDY:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GREEDY:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; GREEDY:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; GREEDY:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; GREEDY:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; GREEDY:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; GREEDY:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+  ; GREEDY:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+  ; GREEDY:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
+  ; GREEDY:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; GREEDY:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
+  ; GREEDY:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; GREEDY: bb.2:
+  ; GREEDY:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; GREEDY:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %19, %bb.2
+  ; GREEDY:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %12(<4 x s32>), %bb.2
+  ; GREEDY:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
+  ; GREEDY:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY8]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GREEDY:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GREEDY:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; GREEDY: bb.3:
+  ; GREEDY:   successors: %bb.4(0x80000000)
+  ; GREEDY:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; GREEDY: bb.4:
+  ; GREEDY:   [[COPY9:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY9]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 ; Waterfall loop needed for rsrc, copy needed for vaddr
 define amdgpu_ps void @load_1d_sgpr_vaddr__vgpr_srsrc(<8 x i32> %rsrc, i32 inreg %s) {
-  ; CHECK-LABEL: name: load_1d_sgpr_vaddr__vgpr_srsrc
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   successors: %bb.2(0x80000000)
-  ; CHECK:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
-  ; CHECK:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
-  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
-  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
-  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
-  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
-  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
-  ; CHECK:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY [[COPY8]](s32)
-  ; CHECK:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-  ; CHECK:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
-  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-  ; CHECK: bb.2:
-  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
-  ; CHECK:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %20, %bb.2
-  ; CHECK:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %12(<4 x s32>), %bb.2
-  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
-  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
-  ; CHECK: bb.3:
-  ; CHECK:   successors: %bb.4(0x80000000)
-  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-  ; CHECK: bb.4:
-  ; CHECK:   [[COPY10:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: load_1d_sgpr_vaddr__vgpr_srsrc
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   successors: %bb.2(0x80000000)
+  ; FAST:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; FAST:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; FAST:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; FAST:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; FAST:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; FAST:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; FAST:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+  ; FAST:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+  ; FAST:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
+  ; FAST:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY [[COPY8]](s32)
+  ; FAST:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; FAST:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
+  ; FAST:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; FAST: bb.2:
+  ; FAST:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; FAST:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %20, %bb.2
+  ; FAST:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %12(<4 x s32>), %bb.2
+  ; FAST:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; FAST:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; FAST:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
+  ; FAST:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
+  ; FAST:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
+  ; FAST:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; FAST:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; FAST:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; FAST: bb.3:
+  ; FAST:   successors: %bb.4(0x80000000)
+  ; FAST:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; FAST: bb.4:
+  ; FAST:   [[COPY10:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: load_1d_sgpr_vaddr__vgpr_srsrc
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   successors: %bb.2(0x80000000)
+  ; GREEDY:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7
+  ; GREEDY:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; GREEDY:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; GREEDY:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; GREEDY:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; GREEDY:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; GREEDY:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+  ; GREEDY:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+  ; GREEDY:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
+  ; GREEDY:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY [[COPY8]](s32)
+  ; GREEDY:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; GREEDY:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
+  ; GREEDY:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; GREEDY: bb.2:
+  ; GREEDY:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; GREEDY:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %20, %bb.2
+  ; GREEDY:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %12(<4 x s32>), %bb.2
+  ; GREEDY:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
+  ; GREEDY:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.load.1d), 15, [[COPY9]](s32), [[BUILD_VECTOR1]](<8 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GREEDY:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GREEDY:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; GREEDY: bb.3:
+  ; GREEDY:   successors: %bb.4(0x80000000)
+  ; GREEDY:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; GREEDY: bb.4:
+  ; GREEDY:   [[COPY10:%[0-9]+]]:vgpr(p1) = COPY [[DEF]](p1)
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[COPY10]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 declare <4 x float> @llvm.amdgcn.image.load.1d.v4f32.i32(i32 immarg, i32, <8 x i32>, i32 immarg, i32 immarg) #0
 
 attributes #0 = { nounwind readonly }

llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.image.sample.1d.ll

 ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck %s
-; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-fast -o - %s | FileCheck -check-prefix=FAST %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -stop-after=regbankselect -regbankselect-greedy -o - %s | FileCheck -check-prefix=GREEDY %s
 
 ; Natural mapping
 define amdgpu_ps void @sample_1d_vgpr_vaddr__sgpr_rsrc__sgpr_samp(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
-  ; CHECK-LABEL: name: sample_1d_vgpr_vaddr__sgpr_rsrc__sgpr_samp
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0
-  ; CHECK:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
-  ; CHECK:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
-  ; CHECK:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
-  ; CHECK:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
-  ; CHECK:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
-  ; CHECK:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr11
-  ; CHECK:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr12
-  ; CHECK:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr13
-  ; CHECK:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: sample_1d_vgpr_vaddr__sgpr_rsrc__sgpr_samp
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0
+  ; FAST:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; FAST:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; FAST:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; FAST:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; FAST:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; FAST:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; FAST:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; FAST:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; FAST:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
+  ; FAST:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr11
+  ; FAST:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr12
+  ; FAST:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr13
+  ; FAST:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: sample_1d_vgpr_vaddr__sgpr_rsrc__sgpr_samp
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $vgpr0
+  ; GREEDY:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; GREEDY:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; GREEDY:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; GREEDY:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; GREEDY:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; GREEDY:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; GREEDY:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; GREEDY:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; GREEDY:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
+  ; GREEDY:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr11
+  ; GREEDY:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr12
+  ; GREEDY:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr13
+  ; GREEDY:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 ; Copy required for VGPR input
 define amdgpu_ps void @sample_1d_sgpr_vaddr__sgpr_rsrc__sgpr_samp(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float inreg %s) {
-  ; CHECK-LABEL: name: sample_1d_sgpr_vaddr__sgpr_rsrc__sgpr_samp
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $sgpr14
-  ; CHECK:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
-  ; CHECK:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
-  ; CHECK:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
-  ; CHECK:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
-  ; CHECK:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
-  ; CHECK:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr11
-  ; CHECK:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr12
-  ; CHECK:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr13
-  ; CHECK:   [[COPY12:%[0-9]+]]:sgpr(s32) = COPY $sgpr14
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[COPY13:%[0-9]+]]:vgpr(s32) = COPY [[COPY12]](s32)
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY13]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: sample_1d_sgpr_vaddr__sgpr_rsrc__sgpr_samp
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $sgpr14
+  ; FAST:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; FAST:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; FAST:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; FAST:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; FAST:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; FAST:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; FAST:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; FAST:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; FAST:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
+  ; FAST:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr11
+  ; FAST:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr12
+  ; FAST:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr13
+  ; FAST:   [[COPY12:%[0-9]+]]:sgpr(s32) = COPY $sgpr14
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[COPY13:%[0-9]+]]:vgpr(s32) = COPY [[COPY12]](s32)
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY13]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: sample_1d_sgpr_vaddr__sgpr_rsrc__sgpr_samp
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $sgpr14
+  ; GREEDY:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; GREEDY:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; GREEDY:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; GREEDY:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; GREEDY:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; GREEDY:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; GREEDY:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; GREEDY:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; GREEDY:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr10
+  ; GREEDY:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr11
+  ; GREEDY:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr12
+  ; GREEDY:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr13
+  ; GREEDY:   [[COPY12:%[0-9]+]]:sgpr(s32) = COPY $sgpr14
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[COPY13:%[0-9]+]]:vgpr(s32) = COPY [[COPY12]](s32)
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY13]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 ; Waterfall loop for rsrc
 define amdgpu_ps void @sample_1d_vgpr_vaddr__vgpr_rsrc__sgpr_samp(<8 x i32> %rsrc, <4 x i32> inreg %samp, float %s) {
-  ; CHECK-LABEL: name: sample_1d_vgpr_vaddr__vgpr_rsrc__sgpr_samp
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   successors: %bb.2(0x80000000)
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
-  ; CHECK:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
-  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
-  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
-  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
-  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
-  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
-  ; CHECK:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-  ; CHECK:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
-  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-  ; CHECK: bb.2:
-  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
-  ; CHECK:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
-  ; CHECK:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
-  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
-  ; CHECK:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
-  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
-  ; CHECK: bb.3:
-  ; CHECK:   successors: %bb.4(0x80000000)
-  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-  ; CHECK: bb.4:
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: sample_1d_vgpr_vaddr__vgpr_rsrc__sgpr_samp
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   successors: %bb.2(0x80000000)
+  ; FAST:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; FAST:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; FAST:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; FAST:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; FAST:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; FAST:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; FAST:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+  ; FAST:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+  ; FAST:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
+  ; FAST:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; FAST:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; FAST:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; FAST:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; FAST:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; FAST:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
+  ; FAST:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; FAST: bb.2:
+  ; FAST:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; FAST:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
+  ; FAST:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
+  ; FAST:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; FAST:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; FAST:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
+  ; FAST:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
+  ; FAST:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
+  ; FAST:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; FAST:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; FAST:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; FAST: bb.3:
+  ; FAST:   successors: %bb.4(0x80000000)
+  ; FAST:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; FAST: bb.4:
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: sample_1d_vgpr_vaddr__vgpr_rsrc__sgpr_samp
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   successors: %bb.2(0x80000000)
+  ; GREEDY:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8
+  ; GREEDY:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; GREEDY:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; GREEDY:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; GREEDY:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; GREEDY:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; GREEDY:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+  ; GREEDY:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+  ; GREEDY:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
+  ; GREEDY:   [[COPY8:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; GREEDY:   [[COPY9:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; GREEDY:   [[COPY10:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; GREEDY:   [[COPY11:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; GREEDY:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[BUILD_VECTOR1:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; GREEDY:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
+  ; GREEDY:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; GREEDY: bb.2:
+  ; GREEDY:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; GREEDY:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
+  ; GREEDY:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
+  ; GREEDY:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
+  ; GREEDY:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR1]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_2]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GREEDY:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GREEDY:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; GREEDY: bb.3:
+  ; GREEDY:   successors: %bb.4(0x80000000)
+  ; GREEDY:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; GREEDY: bb.4:
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 ; Waterfall loop for sampler
 define amdgpu_ps void @sample_1d_vgpr_vaddr__sgpr_rsrc__vgpr_samp(<8 x i32> inreg %rsrc, <4 x i32> %samp, float %s) {
-  ; CHECK-LABEL: name: sample_1d_vgpr_vaddr__sgpr_rsrc__vgpr_samp
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   successors: %bb.2(0x80000000)
-  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
-  ; CHECK:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
-  ; CHECK:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
-  ; CHECK:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
-  ; CHECK:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
-  ; CHECK:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
-  ; CHECK:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
-  ; CHECK:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
-  ; CHECK:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
-  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY10:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
-  ; CHECK:   [[COPY11:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
-  ; CHECK:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-  ; CHECK:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR1]](<4 x s32>)
-  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-  ; CHECK: bb.2:
-  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
-  ; CHECK:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
-  ; CHECK:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
-  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
-  ; CHECK:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR2]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
-  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
-  ; CHECK: bb.3:
-  ; CHECK:   successors: %bb.4(0x80000000)
-  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-  ; CHECK: bb.4:
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: sample_1d_vgpr_vaddr__sgpr_rsrc__vgpr_samp
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   successors: %bb.2(0x80000000)
+  ; FAST:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; FAST:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; FAST:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; FAST:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; FAST:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; FAST:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; FAST:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; FAST:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; FAST:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; FAST:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; FAST:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; FAST:   [[COPY10:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; FAST:   [[COPY11:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; FAST:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[BUILD_VECTOR1:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; FAST:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR1]](<4 x s32>)
+  ; FAST:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; FAST: bb.2:
+  ; FAST:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; FAST:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
+  ; FAST:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
+  ; FAST:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; FAST:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; FAST:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; FAST:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR2]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; FAST:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; FAST:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; FAST: bb.3:
+  ; FAST:   successors: %bb.4(0x80000000)
+  ; FAST:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; FAST: bb.4:
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: sample_1d_vgpr_vaddr__sgpr_rsrc__vgpr_samp
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   successors: %bb.2(0x80000000)
+  ; GREEDY:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; GREEDY:   [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr2
+  ; GREEDY:   [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr3
+  ; GREEDY:   [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr4
+  ; GREEDY:   [[COPY3:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+  ; GREEDY:   [[COPY4:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+  ; GREEDY:   [[COPY5:%[0-9]+]]:sgpr(s32) = COPY $sgpr7
+  ; GREEDY:   [[COPY6:%[0-9]+]]:sgpr(s32) = COPY $sgpr8
+  ; GREEDY:   [[COPY7:%[0-9]+]]:sgpr(s32) = COPY $sgpr9
+  ; GREEDY:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; GREEDY:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; GREEDY:   [[COPY10:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; GREEDY:   [[COPY11:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; GREEDY:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[BUILD_VECTOR1:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; GREEDY:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR1]](<4 x s32>)
+  ; GREEDY:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; GREEDY: bb.2:
+  ; GREEDY:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; GREEDY:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
+  ; GREEDY:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
+  ; GREEDY:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; GREEDY:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR]](<8 x s32>), [[BUILD_VECTOR2]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GREEDY:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GREEDY:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; GREEDY: bb.3:
+  ; GREEDY:   successors: %bb.4(0x80000000)
+  ; GREEDY:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; GREEDY: bb.4:
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 ; Waterfall loop for rsrc and sampler
 define amdgpu_ps void @sample_1d_vgpr_vaddr__vgpr_rsrc__vgpr_samp(<8 x i32> %rsrc, <4 x i32> %samp, float %s) {
-  ; CHECK-LABEL: name: sample_1d_vgpr_vaddr__vgpr_rsrc__vgpr_samp
-  ; CHECK: bb.1 (%ir-block.0):
-  ; CHECK:   successors: %bb.2(0x80000000)
-  ; CHECK:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12
-  ; CHECK:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
-  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
-  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
-  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
-  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
-  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
-  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
-  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY $vgpr9
-  ; CHECK:   [[COPY10:%[0-9]+]]:vgpr(s32) = COPY $vgpr10
-  ; CHECK:   [[COPY11:%[0-9]+]]:vgpr(s32) = COPY $vgpr11
-  ; CHECK:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr12
-  ; CHECK:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
-  ; CHECK:   [[BUILD_VECTOR1:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
-  ; CHECK:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
-  ; CHECK:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
-  ; CHECK:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
-  ; CHECK:   [[UV4:%[0-9]+]]:vreg_64(s64), [[UV5:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR1]](<4 x s32>)
-  ; CHECK:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
-  ; CHECK: bb.2:
-  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
-  ; CHECK:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
-  ; CHECK:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
-  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
-  ; CHECK:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
-  ; CHECK:   [[V_READFIRSTLANE_B32_8:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV4]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_9:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV4]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV4:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_8]](s32), [[V_READFIRSTLANE_B32_9]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_4:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV4]](s64), [[UV4]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_3:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_4]], [[S_AND_B64_2]], implicit-def $scc
-  ; CHECK:   [[V_READFIRSTLANE_B32_10:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV5]].sub0(s64), implicit $exec
-  ; CHECK:   [[V_READFIRSTLANE_B32_11:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV5]].sub1(s64), implicit $exec
-  ; CHECK:   [[MV5:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_10]](s32), [[V_READFIRSTLANE_B32_11]](s32)
-  ; CHECK:   [[V_CMP_EQ_U64_e64_5:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV5]](s64), [[UV5]](s64), implicit $exec
-  ; CHECK:   [[S_AND_B64_4:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_5]], [[S_AND_B64_3]], implicit-def $scc
-  ; CHECK:   [[BUILD_VECTOR3:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_8]](s32), [[V_READFIRSTLANE_B32_9]](s32), [[V_READFIRSTLANE_B32_10]](s32), [[V_READFIRSTLANE_B32_11]](s32)
-  ; CHECK:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR3]](<4 x s32>), 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
-  ; CHECK:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_4]], implicit-def $exec, implicit-def $scc, implicit $exec
-  ; CHECK:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
-  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
-  ; CHECK: bb.3:
-  ; CHECK:   successors: %bb.4(0x80000000)
-  ; CHECK:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
-  ; CHECK: bb.4:
-  ; CHECK:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
-  ; CHECK:   S_ENDPGM 0
+  ; FAST-LABEL: name: sample_1d_vgpr_vaddr__vgpr_rsrc__vgpr_samp
+  ; FAST: bb.1 (%ir-block.0):
+  ; FAST:   successors: %bb.2(0x80000000)
+  ; FAST:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12
+  ; FAST:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; FAST:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; FAST:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; FAST:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; FAST:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; FAST:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+  ; FAST:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+  ; FAST:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
+  ; FAST:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
+  ; FAST:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY $vgpr9
+  ; FAST:   [[COPY10:%[0-9]+]]:vgpr(s32) = COPY $vgpr10
+  ; FAST:   [[COPY11:%[0-9]+]]:vgpr(s32) = COPY $vgpr11
+  ; FAST:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr12
+  ; FAST:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; FAST:   [[BUILD_VECTOR1:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; FAST:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; FAST:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; FAST:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
+  ; FAST:   [[UV4:%[0-9]+]]:vreg_64(s64), [[UV5:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR1]](<4 x s32>)
+  ; FAST:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; FAST: bb.2:
+  ; FAST:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; FAST:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
+  ; FAST:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
+  ; FAST:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; FAST:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; FAST:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
+  ; FAST:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
+  ; FAST:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
+  ; FAST:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; FAST:   [[V_READFIRSTLANE_B32_8:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV4]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_9:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV4]].sub1(s64), implicit $exec
+  ; FAST:   [[MV4:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_8]](s32), [[V_READFIRSTLANE_B32_9]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_4:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV4]](s64), [[UV4]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_3:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_4]], [[S_AND_B64_2]], implicit-def $scc
+  ; FAST:   [[V_READFIRSTLANE_B32_10:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV5]].sub0(s64), implicit $exec
+  ; FAST:   [[V_READFIRSTLANE_B32_11:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV5]].sub1(s64), implicit $exec
+  ; FAST:   [[MV5:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_10]](s32), [[V_READFIRSTLANE_B32_11]](s32)
+  ; FAST:   [[V_CMP_EQ_U64_e64_5:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV5]](s64), [[UV5]](s64), implicit $exec
+  ; FAST:   [[S_AND_B64_4:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_5]], [[S_AND_B64_3]], implicit-def $scc
+  ; FAST:   [[BUILD_VECTOR3:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_8]](s32), [[V_READFIRSTLANE_B32_9]](s32), [[V_READFIRSTLANE_B32_10]](s32), [[V_READFIRSTLANE_B32_11]](s32)
+  ; FAST:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR3]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; FAST:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_4]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; FAST:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; FAST:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; FAST: bb.3:
+  ; FAST:   successors: %bb.4(0x80000000)
+  ; FAST:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; FAST: bb.4:
+  ; FAST:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; FAST:   S_ENDPGM 0
+  ; GREEDY-LABEL: name: sample_1d_vgpr_vaddr__vgpr_rsrc__vgpr_samp
+  ; GREEDY: bb.1 (%ir-block.0):
+  ; GREEDY:   successors: %bb.2(0x80000000)
+  ; GREEDY:   liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12
+  ; GREEDY:   [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+  ; GREEDY:   [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+  ; GREEDY:   [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+  ; GREEDY:   [[COPY3:%[0-9]+]]:vgpr(s32) = COPY $vgpr3
+  ; GREEDY:   [[COPY4:%[0-9]+]]:vgpr(s32) = COPY $vgpr4
+  ; GREEDY:   [[COPY5:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+  ; GREEDY:   [[COPY6:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+  ; GREEDY:   [[COPY7:%[0-9]+]]:vgpr(s32) = COPY $vgpr7
+  ; GREEDY:   [[COPY8:%[0-9]+]]:vgpr(s32) = COPY $vgpr8
+  ; GREEDY:   [[COPY9:%[0-9]+]]:vgpr(s32) = COPY $vgpr9
+  ; GREEDY:   [[COPY10:%[0-9]+]]:vgpr(s32) = COPY $vgpr10
+  ; GREEDY:   [[COPY11:%[0-9]+]]:vgpr(s32) = COPY $vgpr11
+  ; GREEDY:   [[COPY12:%[0-9]+]]:vgpr(s32) = COPY $vgpr12
+  ; GREEDY:   [[BUILD_VECTOR:%[0-9]+]]:vgpr(<8 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32), [[COPY5]](s32), [[COPY6]](s32), [[COPY7]](s32)
+  ; GREEDY:   [[BUILD_VECTOR1:%[0-9]+]]:vgpr(<4 x s32>) = G_BUILD_VECTOR [[COPY8]](s32), [[COPY9]](s32), [[COPY10]](s32), [[COPY11]](s32)
+  ; GREEDY:   [[DEF:%[0-9]+]]:sgpr(p1) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF1:%[0-9]+]]:vgpr(<4 x s32>) = G_IMPLICIT_DEF
+  ; GREEDY:   [[DEF2:%[0-9]+]]:sreg_64_xexec = IMPLICIT_DEF
+  ; GREEDY:   [[UV:%[0-9]+]]:vreg_64(s64), [[UV1:%[0-9]+]]:vreg_64(s64), [[UV2:%[0-9]+]]:vreg_64(s64), [[UV3:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<8 x s32>)
+  ; GREEDY:   [[UV4:%[0-9]+]]:vreg_64(s64), [[UV5:%[0-9]+]]:vreg_64(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR1]](<4 x s32>)
+  ; GREEDY:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; GREEDY: bb.2:
+  ; GREEDY:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; GREEDY:   [[PHI:%[0-9]+]]:sreg_64_xexec = PHI [[DEF2]], %bb.1, %24, %bb.2
+  ; GREEDY:   [[PHI1:%[0-9]+]]:vgpr(<4 x s32>) = G_PHI [[DEF1]](<4 x s32>), %bb.1, %17(<4 x s32>), %bb.2
+  ; GREEDY:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV]](s64), [[UV]](s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV1]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV1:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV1]](s64), [[UV1]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV2]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV2:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV2]](s64), [[UV2]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_2]], [[S_AND_B64_]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV3]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV3:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV3]](s64), [[UV3]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_2:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_3]], [[S_AND_B64_1]], implicit-def $scc
+  ; GREEDY:   [[BUILD_VECTOR2:%[0-9]+]]:sgpr(<8 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_]](s32), [[V_READFIRSTLANE_B32_1]](s32), [[V_READFIRSTLANE_B32_2]](s32), [[V_READFIRSTLANE_B32_3]](s32), [[V_READFIRSTLANE_B32_4]](s32), [[V_READFIRSTLANE_B32_5]](s32), [[V_READFIRSTLANE_B32_6]](s32), [[V_READFIRSTLANE_B32_7]](s32)
+  ; GREEDY:   [[V_READFIRSTLANE_B32_8:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV4]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_9:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV4]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV4:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_8]](s32), [[V_READFIRSTLANE_B32_9]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_4:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV4]](s64), [[UV4]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_3:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_4]], [[S_AND_B64_2]], implicit-def $scc
+  ; GREEDY:   [[V_READFIRSTLANE_B32_10:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV5]].sub0(s64), implicit $exec
+  ; GREEDY:   [[V_READFIRSTLANE_B32_11:%[0-9]+]]:sreg_32_xm0(s32) = V_READFIRSTLANE_B32 [[UV5]].sub1(s64), implicit $exec
+  ; GREEDY:   [[MV5:%[0-9]+]]:sreg_64_xexec(s64) = G_MERGE_VALUES [[V_READFIRSTLANE_B32_10]](s32), [[V_READFIRSTLANE_B32_11]](s32)
+  ; GREEDY:   [[V_CMP_EQ_U64_e64_5:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[MV5]](s64), [[UV5]](s64), implicit $exec
+  ; GREEDY:   [[S_AND_B64_4:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_5]], [[S_AND_B64_3]], implicit-def $scc
+  ; GREEDY:   [[BUILD_VECTOR3:%[0-9]+]]:sgpr(<4 x s32>) = G_BUILD_VECTOR [[V_READFIRSTLANE_B32_8]](s32), [[V_READFIRSTLANE_B32_9]](s32), [[V_READFIRSTLANE_B32_10]](s32), [[V_READFIRSTLANE_B32_11]](s32)
+  ; GREEDY:   [[AMDGPU_INTRIN_IMAGE_LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_AMDGPU_INTRIN_IMAGE_LOAD intrinsic(@llvm.amdgcn.image.sample.1d), 15, [[COPY12]](s32), [[BUILD_VECTOR2]](<8 x s32>), [[BUILD_VECTOR3]](<4 x s32>), 0, 0, 0, 0 :: (dereferenceable load 16 from custom "TargetCustom8")
+  ; GREEDY:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_4]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; GREEDY:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; GREEDY:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; GREEDY: bb.3:
+  ; GREEDY:   successors: %bb.4(0x80000000)
+  ; GREEDY:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; GREEDY: bb.4:
+  ; GREEDY:   G_STORE [[AMDGPU_INTRIN_IMAGE_LOAD]](<4 x s32>), [[DEF]](p1) :: (store 16 into `<4 x float> addrspace(1)* undef`, addrspace 1)
+  ; GREEDY:   S_ENDPGM 0
   %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 false, i32 0, i32 0)
   store <4 x float> %v, <4 x float> addrspace(1)* undef
   ret void
 }
 
 declare <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 immarg, float, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #0
 
 attributes #0 = { nounwind readonly }

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.encode.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -show-mc-encoding < %s | FileCheck -check-prefixes=GFX10 %s
+
+define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_d_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0x88,0xf0,0x00,0x00,0x40,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_d_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff ; encoding: [0xff,0x02,0x0e,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v2, v7, v2 ; encoding: [0x07,0x05,0x04,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v0, v7, v0 ; encoding: [0x07,0x01,0x00,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v1, 16, v0 ; encoding: [0x03,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v3, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0x88,0xf0,0x03,0x00,0x40,0x00,0x02,0x04,0x05,0x00]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r) {
+; GFX10-LABEL: sample_d_3d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v9, 0xffff ; encoding: [0xff,0x02,0x12,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v9, v3 ; encoding: [0x09,0x07,0x06,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v0, v9, v0 ; encoding: [0x09,0x01,0x00,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3 ; encoding: [0x03,0x00,0x6f,0xd7,0x04,0x21,0x0d,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v0, v2, v3, v5, v6, v7, v8], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D ; encoding: [0x15,0x0f,0x88,0xf0,0x00,0x00,0x40,0x00,0x02,0x03,0x05,0x06,0x07,0x08,0x00,0x00]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_c_d_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xa8,0xf0,0x00,0x00,0x40,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_c_d_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v10, 0xffff ; encoding: [0xff,0x02,0x14,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v10, v3 ; encoding: [0x0a,0x07,0x06,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v1, v10, v1 ; encoding: [0x0a,0x03,0x02,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3 ; encoding: [0x03,0x00,0x6f,0xd7,0x04,0x21,0x0d,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1 ; encoding: [0x01,0x00,0x6f,0xd7,0x02,0x21,0x05,0x04]
+; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0xa8,0xf0,0x00,0x00,0x40,0x00,0x01,0x03,0x05,0x06]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_d_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0x8c,0xf0,0x00,0x00,0x40,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_d_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff ; encoding: [0xff,0x02,0x0e,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v0, v7, v0 ; encoding: [0x07,0x01,0x00,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v2, v7, v2 ; encoding: [0x07,0x05,0x04,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v3, 16, v2 ; encoding: [0x03,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
+; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], [v0, v3, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0x8c,0xf0,0x00,0x00,0x40,0x00,0x03,0x04,0x05,0x06]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_c_d_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xac,0xf0,0x00,0x00,0x40,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_c_d_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v8, 0xffff ; encoding: [0xff,0x02,0x10,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v8, v3 ; encoding: [0x08,0x07,0x06,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v1, v8, v1 ; encoding: [0x08,0x03,0x02,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3 ; encoding: [0x03,0x00,0x6f,0xd7,0x04,0x21,0x0d,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1 ; encoding: [0x01,0x00,0x6f,0xd7,0x02,0x21,0x05,0x04]
+; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], [v0, v1, v3, v5, v6, v7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0d,0x0f,0xac,0xf0,0x00,0x00,0x40,0x00,0x01,0x03,0x05,0x06,0x07,0x00,0x00,0x00]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_cd_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_cd_g16 v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xa0,0xf1,0x00,0x00,0x40,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_cd_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff ; encoding: [0xff,0x02,0x0e,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v2, v7, v2 ; encoding: [0x07,0x05,0x04,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v0, v7, v0 ; encoding: [0x07,0x01,0x00,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v1, 16, v0 ; encoding: [0x03,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    image_sample_cd_g16 v[0:3], [v3, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0xa0,0xf1,0x03,0x00,0x40,0x00,0x02,0x04,0x05,0x00]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_c_cd_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xa8,0xf1,0x00,0x00,0x40,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_c_cd_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v10, 0xffff ; encoding: [0xff,0x02,0x14,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v10, v3 ; encoding: [0x0a,0x07,0x06,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v1, v10, v1 ; encoding: [0x0a,0x03,0x02,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3 ; encoding: [0x03,0x00,0x6f,0xd7,0x04,0x21,0x0d,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1 ; encoding: [0x01,0x00,0x6f,0xd7,0x02,0x21,0x05,0x04]
+; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0xa8,0xf1,0x00,0x00,0x40,0x00,0x01,0x03,0x05,0x06]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_cd_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xa4,0xf1,0x00,0x00,0x40,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_cd_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff ; encoding: [0xff,0x02,0x0e,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v0, v7, v0 ; encoding: [0x07,0x01,0x00,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v2, v7, v2 ; encoding: [0x07,0x05,0x04,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0 ; encoding: [0x00,0x00,0x6f,0xd7,0x01,0x21,0x01,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v3, 16, v2 ; encoding: [0x03,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
+; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], [v0, v3, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0b,0x0f,0xa4,0xf1,0x00,0x00,0x40,0x00,0x03,0x04,0x05,0x06]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_c_cd_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ; encoding: [0x01,0x0f,0xac,0xf1,0x00,0x00,0x40,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_c_cd_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v8, 0xffff ; encoding: [0xff,0x02,0x10,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v8, v3 ; encoding: [0x08,0x07,0x06,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v1, v8, v1 ; encoding: [0x08,0x03,0x02,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3 ; encoding: [0x03,0x00,0x6f,0xd7,0x04,0x21,0x0d,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1 ; encoding: [0x01,0x00,0x6f,0xd7,0x02,0x21,0x05,0x04]
+; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], [v0, v1, v3, v5, v6, v7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ; encoding: [0x0d,0x0f,0xac,0xf1,0x00,0x00,0x40,0x00,0x01,0x03,0x05,0x06,0x07,0x00,0x00,0x00]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
+; GFX10-LABEL: sample_c_d_o_2darray_V1:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v9, 0xffff ; encoding: [0xff,0x02,0x12,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v4, v9, v4 ; encoding: [0x09,0x09,0x08,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v2, v9, v2 ; encoding: [0x09,0x05,0x04,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4 ; encoding: [0x04,0x00,0x6f,0xd7,0x05,0x21,0x11,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
+; GFX10-NEXT:    image_sample_c_d_o_g16 v0, [v0, v1, v2, v4, v6, v7, v8], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY ; encoding: [0x2d,0x04,0xe8,0xf0,0x00,0x00,0x40,0x00,0x01,0x02,0x04,0x06,0x07,0x08,0x00,0x00]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret float %v
+}
+
+define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
+; GFX10-LABEL: sample_c_d_o_2darray_V2:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v9, 0xffff ; encoding: [0xff,0x02,0x12,0x7e,0xff,0xff,0x00,0x00]
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v4, v9, v4 ; encoding: [0x09,0x09,0x08,0x36]
+; GFX10-NEXT:    v_and_b32_e32 v2, v9, v2 ; encoding: [0x09,0x05,0x04,0x36]
+; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4 ; encoding: [0x04,0x00,0x6f,0xd7,0x05,0x21,0x11,0x04]
+; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2 ; encoding: [0x02,0x00,0x6f,0xd7,0x03,0x21,0x09,0x04]
+; GFX10-NEXT:    image_sample_c_d_o_g16 v[0:1], [v0, v1, v2, v4, v6, v7, v8], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY ; encoding: [0x2d,0x06,0xe8,0xf0,0x00,0x00,0x40,0x00,0x01,0x02,0x04,0x06,0x07,0x08,0x00,0x00]
+; GFX10-NEXT:    s_waitcnt vmcnt(0) ; encoding: [0x70,0x3f,0x8c,0xbf]
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32 6, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <2 x float> %v
+}
+
+declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32, half, half, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32, float, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32, float, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32, float, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32, float, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32, i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32, i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readonly }
+attributes #2 = { nounwind readnone }

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.image.sample.g16.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10 %s
+
+define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_d_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_d_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v2, v7, v2
+; GFX10-NEXT:    v_and_b32_e32 v0, v7, v0
+; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_lshl_or_b32 v3, v1, 16, v0
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v3, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r) {
+; GFX10-LABEL: sample_d_3d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v9, v3
+; GFX10-NEXT:    v_and_b32_e32 v0, v9, v0
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    image_sample_d_g16 v[0:3], [v0, v2, v3, v5, v6, v7, v8], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %drdh, half %dsdv, half %dtdv, half %drdv, float %s, float %t, float %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_c_d_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_c_d_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v10, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v10, v3
+; GFX10-NEXT:    v_and_b32_e32 v1, v10, v1
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    image_sample_c_d_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_d_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_d_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v0, v7, v0
+; GFX10-NEXT:    v_and_b32_e32 v2, v7, v2
+; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_lshl_or_b32 v3, v3, 16, v2
+; GFX10-NEXT:    image_sample_d_cl_g16 v[0:3], [v0, v3, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_c_d_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_c_d_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v8, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v8, v3
+; GFX10-NEXT:    v_and_b32_e32 v1, v8, v1
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    image_sample_c_d_cl_g16 v[0:3], [v0, v1, v3, v5, v6, v7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_cd_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_cd_g16 v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_cd_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v2, v7, v2
+; GFX10-NEXT:    v_and_b32_e32 v0, v7, v0
+; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    v_lshl_or_b32 v3, v1, 16, v0
+; GFX10-NEXT:    image_sample_cd_g16 v[0:3], [v3, v2, v4, v5], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s) {
+; GFX10-LABEL: sample_c_cd_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t) {
+; GFX10-LABEL: sample_c_cd_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v10, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v10, v3
+; GFX10-NEXT:    v_and_b32_e32 v1, v10, v1
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    image_sample_c_cd_g16 v[0:3], [v0, v1, v3, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_cd_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_cd_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v7, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v0, v7, v0
+; GFX10-NEXT:    v_and_b32_e32 v2, v7, v2
+; GFX10-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX10-NEXT:    v_lshl_or_b32 v3, v3, 16, v2
+; GFX10-NEXT:    image_sample_cd_cl_g16 v[0:3], [v0, v3, v4, v5, v6], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32 15, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp) {
+; GFX10-LABEL: sample_c_cd_cl_1d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps <4 x float> @sample_c_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp) {
+; GFX10-LABEL: sample_c_cd_cl_2d:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v8, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v3, v8, v3
+; GFX10-NEXT:    v_and_b32_e32 v1, v8, v1
+; GFX10-NEXT:    v_lshl_or_b32 v3, v4, 16, v3
+; GFX10-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX10-NEXT:    image_sample_c_cd_cl_g16 v[0:3], [v0, v1, v3, v5, v6, v7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32 15, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
+; GFX10-LABEL: sample_c_d_o_2darray_V1:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v4, v9, v4
+; GFX10-NEXT:    v_and_b32_e32 v2, v9, v2
+; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
+; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    image_sample_c_d_o_g16 v0, [v0, v1, v2, v4, v6, v7, v8], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32 4, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret float %v
+}
+
+define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice) {
+; GFX10-LABEL: sample_c_d_o_2darray_V2:
+; GFX10:       ; %bb.0: ; %main_body
+; GFX10-NEXT:    v_mov_b32_e32 v9, 0xffff
+; GFX10-NEXT:    ; implicit-def: $vcc_hi
+; GFX10-NEXT:    v_and_b32_e32 v4, v9, v4
+; GFX10-NEXT:    v_and_b32_e32 v2, v9, v2
+; GFX10-NEXT:    v_lshl_or_b32 v4, v5, 16, v4
+; GFX10-NEXT:    v_lshl_or_b32 v2, v3, 16, v2
+; GFX10-NEXT:    image_sample_c_d_o_g16 v[0:1], [v0, v1, v2, v4, v6, v7, v8], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY
+; GFX10-NEXT:    s_waitcnt vmcnt(0)
+; GFX10-NEXT:    ; return to shader part epilog
+main_body:
+  %v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32 6, i32 %offset, float %zcompare, half %dsdh, half %dtdh, half %dsdv, half %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <2 x float> %v
+}
+
+declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.3d.v4f32.f16.f32(i32, half, half, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f16.f32(i32, float, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f16.f32(i32, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f16.f32(i32, float, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f16.f32(i32, float, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f16.f32(i32, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f16.f32(i32, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f16.f32(i32, float, half, half, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f16.f32(i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+declare float @llvm.amdgcn.image.sample.c.d.o.2darray.f16.f32.f32(i32, i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+declare <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f16.f32(i32, i32, float, half, half, half, half, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readonly }
+attributes #2 = { nounwind readnone }

llvm/test/CodeGen/AMDGPU/unsupported-image-g16.ll

+; RUN: not --crash llc -march=amdgcn -mcpu=fiji -verify-machineinstrs -o /dev/null %s 2>&1 | FileCheck -check-prefix=ERR %s
+; RUN: not --crash llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs -o /dev/null %s 2>&1 | FileCheck -check-prefix=ERR %s
+
+; Make sure this doesn't assert on targets without the g16 feature, and instead
+; generates a selection error.
+
+; ERR: LLVM ERROR: Cannot select: intrinsic %llvm.amdgcn.image.sample.d.1d
+
+define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, half %dsdh, half %dsdv, float %s) {
+main_body:
+  %v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32 15, half %dsdh, half %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
+  ret <4 x float> %v
+}
+
+declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f16.f32(i32, half, half, float, <8 x i32>, <4 x i32>, i1, i32, i32) #0
+
+attributes #0 = { nounwind readonly }