AMDGPU/GlobalISel: Select llvm.amdgcn.raw.tbuffer.load

[kerbowa]

LGTM

[arsenm]

09ed0e44d9afdcf68d93f79a0e58baaef1df43f3

llvm/lib/Target/AMDGPU/AMDGPUGISel.td

 def : GINodeEquiv<G_AMDGPU_ATOMIC_CMPXCHG, AMDGPUatomic_cmp_swap>;
 def : GINodeEquiv<G_AMDGPU_BUFFER_LOAD, SIbuffer_load>;
 def : GINodeEquiv<G_AMDGPU_BUFFER_LOAD_USHORT, SIbuffer_load_ushort>;
 def : GINodeEquiv<G_AMDGPU_BUFFER_LOAD_UBYTE, SIbuffer_load_ubyte>;
 def : GINodeEquiv<G_AMDGPU_BUFFER_LOAD_SSHORT, SIbuffer_load_short>;
 def : GINodeEquiv<G_AMDGPU_BUFFER_LOAD_SBYTE, SIbuffer_load_byte>;
 def : GINodeEquiv<G_AMDGPU_BUFFER_LOAD_FORMAT, SIbuffer_load_format>;
 def : GINodeEquiv<G_AMDGPU_BUFFER_LOAD_FORMAT_D16, SIbuffer_load_format_d16>;
+def : GINodeEquiv<G_AMDGPU_TBUFFER_LOAD_FORMAT, SItbuffer_load>;
+def : GINodeEquiv<G_AMDGPU_TBUFFER_LOAD_FORMAT_D16, SItbuffer_load_d16>;
 
 class GISelSop2Pat <
   SDPatternOperator node,
   Instruction inst,
   ValueType dst_vt,
   ValueType src0_vt = dst_vt, ValueType src1_vt = src0_vt>   : GCNPat <
 
   (dst_vt (node (src0_vt SReg_32:$src0), (src1_vt SReg_32:$src1))),

 }
 
 def gi_as_i32timm : GICustomOperandRenderer<"renderTruncImm32">,
   GISDNodeXFormEquiv<as_i32timm>;
 
 def gi_as_i16timm : GICustomOperandRenderer<"renderTruncTImm">,
   GISDNodeXFormEquiv<as_i16timm>;
 
+def gi_as_i8timm : GICustomOperandRenderer<"renderTruncTImm8">,
+  GISDNodeXFormEquiv<as_i8timm>;
+
 def gi_NegateImm : GICustomOperandRenderer<"renderNegateImm">,
   GISDNodeXFormEquiv<NegateImm>;
 
 def gi_bitcast_fpimm_to_i32 : GICustomOperandRenderer<"renderBitcastImm">,
   GISDNodeXFormEquiv<bitcast_fpimm_to_i32>;
 
 def gi_IMMPopCount : GICustomOperandRenderer<"renderPopcntImm">,
   GISDNodeXFormEquiv<IMMPopCount>;

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h

private:
   selectDS1Addr1Offset(MachineOperand &Root) const;
 
   void renderTruncImm32(MachineInstrBuilder &MIB, const MachineInstr &MI,
                         int OpIdx = -1) const;
 
   void renderTruncTImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
                        int OpIdx) const;
 
+  void renderTruncTImm8(MachineInstrBuilder &MIB, const MachineInstr &MI,
+                        int OpIdx) const {
+    renderTruncTImm(MIB, MI, OpIdx);
+  }
+
+  void renderTruncTImm16(MachineInstrBuilder &MIB, const MachineInstr &MI,
+                        int OpIdx) const {
+    renderTruncTImm(MIB, MI, OpIdx);
+  }
+
   void renderNegateImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
                        int OpIdx) const;
 
   void renderBitcastImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
                         int OpIdx) const;
 
   void renderPopcntImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
                        int OpIdx) const;

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.h

public:
   std::tuple<Register, unsigned, unsigned>
   splitBufferOffsets(MachineIRBuilder &B, Register OrigOffset) const;
 
   Register handleD16VData(MachineIRBuilder &B, MachineRegisterInfo &MRI,
                           Register Reg) const;
   bool legalizeRawBufferStore(MachineInstr &MI, MachineRegisterInfo &MRI,
                               MachineIRBuilder &B, bool IsFormat) const;
   bool legalizeBufferLoad(MachineInstr &MI, MachineRegisterInfo &MRI,
-                          MachineIRBuilder &B, bool IsFormat) const;
+                          MachineIRBuilder &B, bool IsFormat,
+                          bool IsTyped) const;
 
   bool legalizeIntrinsic(MachineInstr &MI, MachineRegisterInfo &MRI,
                          MachineIRBuilder &B) const override;
 
 };
 } // End llvm namespace.
 #endif

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

bool AMDGPULegalizerInfo::legalizeRawBufferStore(MachineInstr &MI,
   }
 
   return Ty == S32;
 }
 
 bool AMDGPULegalizerInfo::legalizeBufferLoad(MachineInstr &MI,
                                              MachineRegisterInfo &MRI,
                                              MachineIRBuilder &B,
-                                             bool IsFormat) const {
+                                             bool IsFormat,
+                                             bool IsTyped) const {
   B.setInstr(MI);
 
   // FIXME: Verifier should enforce 1 MMO for these intrinsics.
   MachineMemOperand *MMO = *MI.memoperands_begin();
   const int MemSize = MMO->getSize();
   const LLT S32 = LLT::scalar(32);
 
   Register Dst = MI.getOperand(0).getReg();
   Register RSrc = MI.getOperand(2).getReg();
 
+  // The typed intrinsics add an immediate after the registers.
+  const unsigned NumVIndexOps = IsTyped ? 8 : 7;
+
   // The struct intrinsic variants add one additional operand over raw.
-  const bool HasVIndex = MI.getNumOperands() == 7;
+  const bool HasVIndex = MI.getNumOperands() == NumVIndexOps;
   Register VIndex;
   int OpOffset = 0;
   if (HasVIndex) {
     VIndex = MI.getOperand(3).getReg();
     OpOffset = 1;
   }
 
   Register VOffset = MI.getOperand(3 + OpOffset).getReg();
   Register SOffset = MI.getOperand(4 + OpOffset).getReg();
+
+  unsigned Format = 0;
+  if (IsTyped) {
+    Format = MI.getOperand(5 + OpOffset).getImm();
+    ++OpOffset;
+  }
+
   unsigned AuxiliaryData = MI.getOperand(5 + OpOffset).getImm();
   unsigned ImmOffset;
   unsigned TotalOffset;
 
   LLT Ty = MRI.getType(Dst);
   LLT EltTy = Ty.getScalarType();
   const bool IsD16 = IsFormat && (EltTy.getSizeInBits() == 16);
   const bool Unpacked = ST.hasUnpackedD16VMem();
 
   std::tie(VOffset, ImmOffset, TotalOffset) = splitBufferOffsets(B, VOffset);
   if (TotalOffset != 0)
     MMO = B.getMF().getMachineMemOperand(MMO, TotalOffset, MemSize);
 
   unsigned Opc;
-  if (IsFormat) {
+
+  if (IsTyped) {
+    Opc = IsD16 ? AMDGPU::G_AMDGPU_TBUFFER_LOAD_FORMAT_D16 :
+                  AMDGPU::G_AMDGPU_TBUFFER_LOAD_FORMAT;
+  } else if (IsFormat) {
     Opc = IsD16 ? AMDGPU::G_AMDGPU_BUFFER_LOAD_FORMAT_D16 :
                   AMDGPU::G_AMDGPU_BUFFER_LOAD_FORMAT;
   } else {
     switch (MemSize) {
     case 1:
       Opc = AMDGPU::G_AMDGPU_BUFFER_LOAD_UBYTE;
       break;
     case 2:
bool AMDGPULegalizerInfo::legalizeBufferLoad(MachineInstr &MI,
   else if (Unpacked && IsD16 && Ty.isVector())
     LoadDstReg = B.getMRI()->createGenericVirtualRegister(UnpackedTy);
   else
     LoadDstReg = Dst;
 
   if (!VIndex)
     VIndex = B.buildConstant(S32, 0).getReg(0);
 
-  B.buildInstr(Opc)
+  auto MIB = B.buildInstr(Opc)
     .addDef(LoadDstReg)         // vdata
     .addUse(RSrc)               // rsrc
     .addUse(VIndex)             // vindex
     .addUse(VOffset)            // voffset
     .addUse(SOffset)            // soffset
-    .addImm(ImmOffset)          // offset(imm)
-    .addImm(AuxiliaryData)      // cachepolicy, swizzled buffer(imm)
+    .addImm(ImmOffset);         // offset(imm)
+
+  if (IsTyped)
+    MIB.addImm(Format);
+
+  MIB.addImm(AuxiliaryData)      // cachepolicy, swizzled buffer(imm)
      .addImm(HasVIndex ? -1 : 0) // idxen(imm)
      .addMemOperand(MMO);
 
   if (LoadDstReg != Dst) {
     B.setInsertPt(B.getMBB(), ++B.getInsertPt());
 
     // Widen result for extending loads was widened.
     if (IsExtLoad)
       B.buildTrunc(Dst, LoadDstReg);
     else {
case Intrinsic::amdgcn_wavefrontsize: {
     return true;
   }
   case Intrinsic::amdgcn_raw_buffer_store:
     return legalizeRawBufferStore(MI, MRI, B, false);
   case Intrinsic::amdgcn_raw_buffer_store_format:
     return legalizeRawBufferStore(MI, MRI, B, true);
   case Intrinsic::amdgcn_raw_buffer_load:
   case Intrinsic::amdgcn_struct_buffer_load:
-    return legalizeBufferLoad(MI, MRI, B, false);
+    return legalizeBufferLoad(MI, MRI, B, false, false);
   case Intrinsic::amdgcn_raw_buffer_load_format:
   case Intrinsic::amdgcn_struct_buffer_load_format:
-    return legalizeBufferLoad(MI, MRI, B, true);
+    return legalizeBufferLoad(MI, MRI, B, true, false);
+  case Intrinsic::amdgcn_raw_tbuffer_load:
+    return legalizeBufferLoad(MI, MRI, B, true, true);
   default:
     return true;
   }
 
   return true;
 }

llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp

case AMDGPU::G_INSERT_VECTOR_ELT: {
     return;
   }
   case AMDGPU::G_AMDGPU_BUFFER_LOAD:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_USHORT:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_SSHORT:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_UBYTE:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_SBYTE:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_FORMAT:
-  case AMDGPU::G_AMDGPU_BUFFER_LOAD_FORMAT_D16: {
+  case AMDGPU::G_AMDGPU_BUFFER_LOAD_FORMAT_D16:
+  case AMDGPU::G_AMDGPU_TBUFFER_LOAD_FORMAT:
+  case AMDGPU::G_AMDGPU_TBUFFER_LOAD_FORMAT_D16: {
     applyDefaultMapping(OpdMapper);
     executeInWaterfallLoop(MI, MRI, {1, 4});
     return;
   }
   case AMDGPU::G_INTRINSIC: {
     switch (MI.getIntrinsicID()) {
     case Intrinsic::amdgcn_s_buffer_load: {
       // FIXME: Move to G_INTRINSIC_W_SIDE_EFFECTS
case AMDGPU::G_UNMERGE_VALUES: {
     break;
   }
   case AMDGPU::G_AMDGPU_BUFFER_LOAD:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_UBYTE:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_SBYTE:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_USHORT:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_SSHORT:
   case AMDGPU::G_AMDGPU_BUFFER_LOAD_FORMAT:
-  case AMDGPU::G_AMDGPU_BUFFER_LOAD_FORMAT_D16: {
+  case AMDGPU::G_AMDGPU_BUFFER_LOAD_FORMAT_D16:
+  case AMDGPU::G_AMDGPU_TBUFFER_LOAD_FORMAT:
+  case AMDGPU::G_AMDGPU_TBUFFER_LOAD_FORMAT_D16: {
     OpdsMapping[0] = getVGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);
 
     // rsrc
     OpdsMapping[1] = getSGPROpMapping(MI.getOperand(1).getReg(), MRI, *TRI);
 
     // vindex
     OpdsMapping[2] = getVGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);
 
     // vindex
     OpdsMapping[3] = getVGPROpMapping(MI.getOperand(3).getReg(), MRI, *TRI);
 
     // soffset
     OpdsMapping[4] = getSGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI);
+
+    // Any remaining operands are immediates and were correctly null
+    // initialized.
     break;
   }
   case AMDGPU::G_INTRINSIC: {
     switch (MI.getIntrinsicID()) {
     default:
       return getInvalidInstructionMapping();
     case Intrinsic::amdgcn_div_fmas:
     case Intrinsic::amdgcn_div_fixup:

llvm/lib/Target/AMDGPU/BUFInstructions.td

 // tbuffer_load/store_format patterns
 //===----------------------------------------------------------------------===//
 
 multiclass MTBUF_LoadIntrinsicPat<SDPatternOperator name, ValueType vt,
                                   string opcode> {
   def : GCNPat<
     (vt (name v4i32:$rsrc, 0, 0, i32:$soffset, timm:$offset,
               timm:$format, timm:$auxiliary, 0)),
-    (!cast<MTBUF_Pseudo>(opcode # _OFFSET) $rsrc, $soffset, (as_i16imm $offset),
-      (as_i8imm $format),
+    (!cast<MTBUF_Pseudo>(opcode # _OFFSET) SReg_128:$rsrc, SCSrc_b32:$soffset, (as_i16timm $offset),
+      (as_i8timm $format),
       (extract_glc $auxiliary), (extract_slc $auxiliary), 0, (extract_dlc $auxiliary),
       (extract_swz $auxiliary))
   >;
 
   def : GCNPat<
     (vt (name v4i32:$rsrc, i32:$vindex, 0, i32:$soffset, timm:$offset,
               timm:$format, timm:$auxiliary, timm)),
-    (!cast<MTBUF_Pseudo>(opcode # _IDXEN) $vindex, $rsrc, $soffset, (as_i16imm $offset),
-      (as_i8imm $format),
+    (!cast<MTBUF_Pseudo>(opcode # _IDXEN) VGPR_32:$vindex, SReg_128:$rsrc, SCSrc_b32:$soffset, (as_i16timm $offset),
+      (as_i8timm $format),
       (extract_glc $auxiliary), (extract_slc $auxiliary), 0, (extract_dlc $auxiliary),
       (extract_swz $auxiliary))
   >;
 
   def : GCNPat<
     (vt (name v4i32:$rsrc, 0, i32:$voffset, i32:$soffset, timm:$offset,
               timm:$format, timm:$auxiliary, 0)),
-    (!cast<MTBUF_Pseudo>(opcode # _OFFEN) $voffset, $rsrc, $soffset, (as_i16imm $offset),
-      (as_i8imm $format),
+    (!cast<MTBUF_Pseudo>(opcode # _OFFEN) VGPR_32:$voffset, SReg_128:$rsrc, SCSrc_b32:$soffset, (as_i16timm $offset),
+      (as_i8timm $format),
       (extract_glc $auxiliary), (extract_slc $auxiliary), 0, (extract_dlc $auxiliary),
       (extract_swz $auxiliary))
   >;
 
   def : GCNPat<
     (vt (name v4i32:$rsrc, i32:$vindex, i32:$voffset, i32:$soffset, timm:$offset,
               timm:$format, timm:$auxiliary, timm)),
     (!cast<MTBUF_Pseudo>(opcode # _BOTHEN)
-      (REG_SEQUENCE VReg_64, $vindex, sub0, $voffset, sub1),
-      $rsrc, $soffset, (as_i16imm $offset),
-      (as_i8imm $format),
+      (REG_SEQUENCE VReg_64, VGPR_32:$vindex, sub0, VGPR_32:$voffset, sub1),
+      SReg_128:$rsrc, SCSrc_b32:$soffset, (as_i16timm $offset),
+      (as_i8timm $format),
       (extract_glc $auxiliary), (extract_slc $auxiliary), 0, (extract_dlc $auxiliary),
       (extract_swz $auxiliary))
   >;
 }
 
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, i32,   "TBUFFER_LOAD_FORMAT_X">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v2i32, "TBUFFER_LOAD_FORMAT_XY">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v3i32, "TBUFFER_LOAD_FORMAT_XYZ">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v4i32, "TBUFFER_LOAD_FORMAT_XYZW">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, f32,   "TBUFFER_LOAD_FORMAT_X">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v2f32, "TBUFFER_LOAD_FORMAT_XY">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v3f32, "TBUFFER_LOAD_FORMAT_XYZ">;
 defm : MTBUF_LoadIntrinsicPat<SItbuffer_load, v4f32, "TBUFFER_LOAD_FORMAT_XYZW">;
 
 let SubtargetPredicate = HasUnpackedD16VMem in {
   defm : MTBUF_LoadIntrinsicPat<SItbuffer_load_d16, f16,   "TBUFFER_LOAD_FORMAT_D16_X_gfx80">;
+  defm : MTBUF_LoadIntrinsicPat<SItbuffer_load_d16, i32,   "TBUFFER_LOAD_FORMAT_D16_X_gfx80">;
   defm : MTBUF_LoadIntrinsicPat<SItbuffer_load_d16, v2i32, "TBUFFER_LOAD_FORMAT_D16_XY_gfx80">;
   defm : MTBUF_LoadIntrinsicPat<SItbuffer_load_d16, v4i32, "TBUFFER_LOAD_FORMAT_D16_XYZW_gfx80">;
 } // End HasUnpackedD16VMem.
 
 let SubtargetPredicate = HasPackedD16VMem in {
   defm : MTBUF_LoadIntrinsicPat<SItbuffer_load_d16, f16,   "TBUFFER_LOAD_FORMAT_D16_X">;
+  defm : MTBUF_LoadIntrinsicPat<SItbuffer_load_d16, i32,   "TBUFFER_LOAD_FORMAT_D16_X">;
   defm : MTBUF_LoadIntrinsicPat<SItbuffer_load_d16, v2f16, "TBUFFER_LOAD_FORMAT_D16_XY">;
   defm : MTBUF_LoadIntrinsicPat<SItbuffer_load_d16, v4f16, "TBUFFER_LOAD_FORMAT_D16_XYZW">;
 } // End HasPackedD16VMem.
 
 multiclass MTBUF_StoreIntrinsicPat<SDPatternOperator name, ValueType vt,
                                    string opcode> {
   def : GCNPat<
     (name vt:$vdata, v4i32:$rsrc, 0, 0, i32:$soffset, timm:$offset,

llvm/lib/Target/AMDGPU/SIInstrInfo.td

 def as_i1imm : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getZExtValue(), SDLoc(N), MVT::i1);
 }]>;
 
 def as_i8imm : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getZExtValue(), SDLoc(N), MVT::i8);
 }]>;
 
+def as_i8timm : SDNodeXForm<timm, [{
+  return CurDAG->getTargetConstant(N->getSExtValue(), SDLoc(N), MVT::i16);
+}]>;
+
 def as_i16imm : SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getSExtValue(), SDLoc(N), MVT::i16);
 }]>;
 
 def as_i16timm : SDNodeXForm<timm, [{
   return CurDAG->getTargetConstant(N->getSExtValue(), SDLoc(N), MVT::i16);
 }]>;
 

llvm/lib/Target/AMDGPU/SIInstructions.td

class BufferLoadGenericInstruction : AMDGPUGenericInstruction {
   let OutOperandList = (outs type0:$dst);
   let InOperandList = (ins type1:$rsrc, type2:$vindex, type2:$voffset,
                            type2:$soffset, untyped_imm_0:$offset,
                            untyped_imm_0:$cachepolicy, untyped_imm_0:$idxen);
   let hasSideEffects = 0;
   let mayLoad = 1;
 }
 
+class TBufferLoadGenericInstruction : AMDGPUGenericInstruction {
+  let OutOperandList = (outs type0:$dst);
+  let InOperandList = (ins type1:$rsrc, type2:$vindex, type2:$voffset,
+                           type2:$soffset, untyped_imm_0:$offset, untyped_imm_0:$format,
+                           untyped_imm_0:$cachepolicy, untyped_imm_0:$idxen);
+  let hasSideEffects = 0;
+  let mayLoad = 1;
+}
+
 def G_AMDGPU_BUFFER_LOAD_UBYTE : BufferLoadGenericInstruction;
 def G_AMDGPU_BUFFER_LOAD_SBYTE : BufferLoadGenericInstruction;
 def G_AMDGPU_BUFFER_LOAD_USHORT : BufferLoadGenericInstruction;
 def G_AMDGPU_BUFFER_LOAD_SSHORT : BufferLoadGenericInstruction;
 def G_AMDGPU_BUFFER_LOAD : BufferLoadGenericInstruction;
 def G_AMDGPU_BUFFER_LOAD_FORMAT : BufferLoadGenericInstruction;
 def G_AMDGPU_BUFFER_LOAD_FORMAT_D16 : BufferLoadGenericInstruction;
+def G_AMDGPU_TBUFFER_LOAD_FORMAT : TBufferLoadGenericInstruction;
+def G_AMDGPU_TBUFFER_LOAD_FORMAT_D16 : TBufferLoadGenericInstruction;
+
 
 // Atomic cmpxchg. $cmpval ad $newval are packed in a single vector
 // operand Expects a MachineMemOperand in addition to explicit
 // operands.
 def G_AMDGPU_ATOMIC_CMPXCHG : AMDGPUGenericInstruction {
   let OutOperandList = (outs type0:$oldval);
   let InOperandList = (ins ptype1:$addr, type0:$cmpval_nnenwval);
   let hasSideEffects = 0;
   let mayLoad = 1;
   let mayStore = 1;
 }

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.raw.tbuffer.load.f16.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=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=UNPACKED %s
+; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx810 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=PACKED %s
+
+define amdgpu_ps half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; UNPACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; UNPACKED: bb.1 (%ir-block.0):
+  ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; UNPACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; UNPACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; UNPACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; UNPACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; UNPACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; UNPACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; UNPACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; UNPACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; UNPACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN]]
+  ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  ; PACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; PACKED: bb.1 (%ir-block.0):
+  ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; PACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; PACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; PACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; PACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; PACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; PACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; PACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; PACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; PACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret half %val
+}
+
+define amdgpu_ps <2 x half> @raw_tbuffer_load_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; UNPACKED-LABEL: name: raw_tbuffer_load_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; UNPACKED: bb.1 (%ir-block.0):
+  ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; UNPACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; UNPACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; UNPACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; UNPACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; UNPACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; UNPACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; UNPACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; UNPACKED:   [[TBUFFER_LOAD_FORMAT_D16_XY_gfx80_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_D16_XY_gfx80_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
+  ; UNPACKED:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XY_gfx80_OFFEN]].sub0
+  ; UNPACKED:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XY_gfx80_OFFEN]].sub1
+  ; UNPACKED:   [[V_BFE_U32_:%[0-9]+]]:vgpr_32 = V_BFE_U32 [[COPY6]], 0, 16, implicit $exec
+  ; UNPACKED:   [[V_BFE_U32_1:%[0-9]+]]:vgpr_32 = V_BFE_U32 [[COPY7]], 0, 16, implicit $exec
+  ; UNPACKED:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 16, implicit $exec
+  ; UNPACKED:   [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[V_MOV_B32_e32_]], [[V_BFE_U32_1]], implicit $exec
+  ; UNPACKED:   [[V_OR_B32_e32_:%[0-9]+]]:vgpr_32 = V_OR_B32_e32 [[V_BFE_U32_]], [[V_LSHLREV_B32_e64_]], implicit $exec
+  ; UNPACKED:   $vgpr0 = COPY [[V_OR_B32_e32_]]
+  ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  ; PACKED-LABEL: name: raw_tbuffer_load_v2f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; PACKED: bb.1 (%ir-block.0):
+  ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; PACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; PACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; PACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; PACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; PACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; PACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; PACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; PACKED:   [[TBUFFER_LOAD_FORMAT_D16_XY_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_XY_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
+  ; PACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_XY_OFFEN]]
+  ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call <2 x half> @llvm.amdgcn.raw.tbuffer.load.v2f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <2 x half> %val
+}
+
+; FIXME: Crashes
+; define amdgpu_ps <3 x half> @raw_tbuffer_load_v3f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+;   %val = call <3 x half> @llvm.amdgcn.raw.tbuffer.load.v3f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+;   ret <3 x half> %val
+; }
+
+define amdgpu_ps <4 x half> @raw_tbuffer_load_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; UNPACKED-LABEL: name: raw_tbuffer_load_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; UNPACKED: bb.1 (%ir-block.0):
+  ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; UNPACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; UNPACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; UNPACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; UNPACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; UNPACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; UNPACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; UNPACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; UNPACKED:   [[TBUFFER_LOAD_FORMAT_D16_XYZW_gfx80_OFFEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_D16_XYZW_gfx80_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8 from custom "TargetCustom7", align 1, addrspace 4)
+  ; UNPACKED:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XYZW_gfx80_OFFEN]].sub0
+  ; UNPACKED:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XYZW_gfx80_OFFEN]].sub1
+  ; UNPACKED:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XYZW_gfx80_OFFEN]].sub2
+  ; UNPACKED:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XYZW_gfx80_OFFEN]].sub3
+  ; UNPACKED:   [[V_BFE_U32_:%[0-9]+]]:vgpr_32 = V_BFE_U32 [[COPY6]], 0, 16, implicit $exec
+  ; UNPACKED:   [[V_BFE_U32_1:%[0-9]+]]:vgpr_32 = V_BFE_U32 [[COPY7]], 0, 16, implicit $exec
+  ; UNPACKED:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 16, implicit $exec
+  ; UNPACKED:   [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[V_MOV_B32_e32_]], [[V_BFE_U32_1]], implicit $exec
+  ; UNPACKED:   [[V_OR_B32_e32_:%[0-9]+]]:vgpr_32 = V_OR_B32_e32 [[V_BFE_U32_]], [[V_LSHLREV_B32_e64_]], implicit $exec
+  ; UNPACKED:   [[V_BFE_U32_2:%[0-9]+]]:vgpr_32 = V_BFE_U32 [[COPY8]], 0, 16, implicit $exec
+  ; UNPACKED:   [[V_BFE_U32_3:%[0-9]+]]:vgpr_32 = V_BFE_U32 [[COPY9]], 0, 16, implicit $exec
+  ; UNPACKED:   [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 16, implicit $exec
+  ; UNPACKED:   [[V_LSHLREV_B32_e64_1:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[V_MOV_B32_e32_1]], [[V_BFE_U32_3]], implicit $exec
+  ; UNPACKED:   [[V_OR_B32_e32_1:%[0-9]+]]:vgpr_32 = V_OR_B32_e32 [[V_BFE_U32_2]], [[V_LSHLREV_B32_e64_1]], implicit $exec
+  ; UNPACKED:   $vgpr0 = COPY [[V_OR_B32_e32_]]
+  ; UNPACKED:   $vgpr1 = COPY [[V_OR_B32_e32_1]]
+  ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
+  ; PACKED-LABEL: name: raw_tbuffer_load_v4f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; PACKED: bb.1 (%ir-block.0):
+  ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; PACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; PACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; PACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; PACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; PACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; PACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; PACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; PACKED:   [[TBUFFER_LOAD_FORMAT_D16_XYZW_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_D16_XYZW_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8 from custom "TargetCustom7", align 1, addrspace 4)
+  ; PACKED:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XYZW_OFFEN]].sub0
+  ; PACKED:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_D16_XYZW_OFFEN]].sub1
+  ; PACKED:   $vgpr0 = COPY [[COPY6]]
+  ; PACKED:   $vgpr1 = COPY [[COPY7]]
+  ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
+  %val = call <4 x half> @llvm.amdgcn.raw.tbuffer.load.v4f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <4 x half> %val
+}
+
+define amdgpu_ps half @raw_tbuffer_load_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 inreg %voffset, i32 %soffset) {
+  ; UNPACKED-LABEL: name: raw_tbuffer_load_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; UNPACKED: bb.1 (%ir-block.0):
+  ; UNPACKED:   successors: %bb.2(0x80000000)
+  ; UNPACKED:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; UNPACKED:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; UNPACKED:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; UNPACKED:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; UNPACKED:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; UNPACKED:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; UNPACKED:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; UNPACKED:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; UNPACKED:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
+  ; UNPACKED:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; UNPACKED:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; UNPACKED:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; UNPACKED: bb.2:
+  ; UNPACKED:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; UNPACKED:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub0, implicit $exec
+  ; UNPACKED:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub1, implicit $exec
+  ; UNPACKED:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; UNPACKED:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY7]], implicit $exec
+  ; UNPACKED:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub0, implicit $exec
+  ; UNPACKED:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub1, implicit $exec
+  ; UNPACKED:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; UNPACKED:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY8]], implicit $exec
+  ; UNPACKED:   [[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
+  ; UNPACKED:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; UNPACKED:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY5]], implicit $exec
+  ; UNPACKED:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY5]], implicit $exec
+  ; UNPACKED:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
+  ; UNPACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN [[COPY6]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; UNPACKED:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; UNPACKED:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; UNPACKED:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; UNPACKED: bb.3:
+  ; UNPACKED:   successors: %bb.4(0x80000000)
+  ; UNPACKED:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; UNPACKED: bb.4:
+  ; UNPACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN]]
+  ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  ; PACKED-LABEL: name: raw_tbuffer_load_f16__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; PACKED: bb.1 (%ir-block.0):
+  ; PACKED:   successors: %bb.2(0x80000000)
+  ; PACKED:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; PACKED:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; PACKED:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; PACKED:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; PACKED:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; PACKED:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; PACKED:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; PACKED:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; PACKED:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
+  ; PACKED:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; PACKED:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; PACKED:   [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec
+  ; PACKED: bb.2:
+  ; PACKED:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; PACKED:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub0, implicit $exec
+  ; PACKED:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub1, implicit $exec
+  ; PACKED:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; PACKED:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY7]], implicit $exec
+  ; PACKED:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub0, implicit $exec
+  ; PACKED:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub1, implicit $exec
+  ; PACKED:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; PACKED:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY8]], implicit $exec
+  ; PACKED:   [[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
+  ; PACKED:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; PACKED:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY5]], implicit $exec
+  ; PACKED:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY5]], implicit $exec
+  ; PACKED:   [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc
+  ; PACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY6]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; PACKED:   [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; PACKED:   $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc
+  ; PACKED:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; PACKED: bb.3:
+  ; PACKED:   successors: %bb.4(0x80000000)
+  ; PACKED:   $exec = S_MOV_B64_term [[S_MOV_B64_term]]
+  ; PACKED: bb.4:
+  ; PACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret half %val
+}
+
+define amdgpu_ps half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; UNPACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
+  ; UNPACKED: bb.1 (%ir-block.0):
+  ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; UNPACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; UNPACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; UNPACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; UNPACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; UNPACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; UNPACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; UNPACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; UNPACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 1, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; UNPACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN]]
+  ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  ; PACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
+  ; PACKED: bb.1 (%ir-block.0):
+  ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; PACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; PACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; PACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; PACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; PACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; PACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; PACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; PACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 1, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; PACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 1)
+  ret half %val
+}
+
+define amdgpu_ps half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; UNPACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
+  ; UNPACKED: bb.1 (%ir-block.0):
+  ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; UNPACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; UNPACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; UNPACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; UNPACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; UNPACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; UNPACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; UNPACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; UNPACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 1, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; UNPACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN]]
+  ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  ; PACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
+  ; PACKED: bb.1 (%ir-block.0):
+  ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; PACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; PACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; PACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; PACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; PACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; PACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; PACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; PACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 1, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; PACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 2)
+  ret half %val
+}
+
+define amdgpu_ps half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; UNPACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
+  ; UNPACKED: bb.1 (%ir-block.0):
+  ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; UNPACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; UNPACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; UNPACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; UNPACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; UNPACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; UNPACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; UNPACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; UNPACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 1, 1, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; UNPACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN]]
+  ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  ; PACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
+  ; PACKED: bb.1 (%ir-block.0):
+  ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; PACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; PACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; PACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; PACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; PACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; PACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; PACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; PACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 1, 1, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; PACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 3)
+  ret half %val
+}
+
+define amdgpu_ps half @raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; UNPACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
+  ; UNPACKED: bb.1 (%ir-block.0):
+  ; UNPACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; UNPACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; UNPACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; UNPACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; UNPACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; UNPACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; UNPACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; UNPACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; UNPACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 1, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; UNPACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_gfx80_OFFEN]]
+  ; UNPACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  ; PACKED-LABEL: name: raw_tbuffer_load_f16__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
+  ; PACKED: bb.1 (%ir-block.0):
+  ; PACKED:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; PACKED:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; PACKED:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; PACKED:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; PACKED:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; PACKED:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; PACKED:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; PACKED:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; PACKED:   [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_D16_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 1, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4)
+  ; PACKED:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_D16_X_OFFEN]]
+  ; PACKED:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 4)
+  ret half %val
+}
+
+declare half @llvm.amdgcn.raw.tbuffer.load.f16(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <2 x half> @llvm.amdgcn.raw.tbuffer.load.v2f16(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <3 x half> @llvm.amdgcn.raw.tbuffer.load.v3f16(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <4 x half> @llvm.amdgcn.raw.tbuffer.load.v4f16(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+
+attributes #0 = { nounwind readonly }

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.raw.tbuffer.load.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=instruction-select -verify-machineinstrs -o - %s | FileCheck %s
+
+define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret float %val
+}
+
+define amdgpu_ps <2 x float> @raw_tbuffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_v2f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
+  ; CHECK:   $vgpr0 = COPY [[COPY6]]
+  ; CHECK:   $vgpr1 = COPY [[COPY7]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1
+  %val = call <2 x float> @llvm.amdgcn.raw.tbuffer.load.v2f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <2 x float> %val
+}
+
+define amdgpu_ps <3 x float> @raw_tbuffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_v3f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 12 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub0
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub1
+  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub2
+  ; CHECK:   $vgpr0 = COPY [[COPY6]]
+  ; CHECK:   $vgpr1 = COPY [[COPY7]]
+  ; CHECK:   $vgpr2 = COPY [[COPY8]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
+  %val = call <3 x float> @llvm.amdgcn.raw.tbuffer.load.v3f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <3 x float> %val
+}
+
+define amdgpu_ps <4 x float> @raw_tbuffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_v4f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN:%[0-9]+]]:vreg_128 = TBUFFER_LOAD_FORMAT_XYZW_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 16 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub0
+  ; CHECK:   [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub1
+  ; CHECK:   [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub2
+  ; CHECK:   [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZW_OFFEN]].sub3
+  ; CHECK:   $vgpr0 = COPY [[COPY6]]
+  ; CHECK:   $vgpr1 = COPY [[COPY7]]
+  ; CHECK:   $vgpr2 = COPY [[COPY8]]
+  ; CHECK:   $vgpr3 = COPY [[COPY9]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
+  %val = call <4 x float> @llvm.amdgcn.raw.tbuffer.load.v4f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret <4 x float> %val
+}
+
+define amdgpu_ps float @raw_tbuffer_load_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 inreg %voffset, i32 %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_f32__vgpr_rsrc__sgpr_voffset__vgpr_soffset
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   liveins: $sgpr2, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+  ; CHECK:   [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+  ; CHECK:   [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
+  ; CHECK:   [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr4
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[COPY4]]
+  ; CHECK:   [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1
+  ; CHECK:   [[COPY8:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3
+  ; CHECK:   [[S_MOV_B32_term:%[0-9]+]]:sreg_32_xm0_xexec = S_MOV_B32_term $exec_lo
+  ; CHECK: bb.2:
+  ; CHECK:   successors: %bb.3(0x40000000), %bb.2(0x40000000)
+  ; CHECK:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY7]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY7]], implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub0, implicit $exec
+  ; CHECK:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY8]].sub1, implicit $exec
+  ; CHECK:   [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
+  ; CHECK:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY8]], implicit $exec
+  ; CHECK:   [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc
+  ; CHECK:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; CHECK:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY5]], implicit $exec
+  ; CHECK:   [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY5]], implicit $exec
+  ; CHECK:   [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 [[V_CMP_EQ_U32_e64_]], [[S_AND_B32_]], implicit-def $scc
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY6]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 78, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_1]], implicit-def $exec, implicit-def $scc, implicit $exec
+  ; CHECK:   $exec_lo = S_XOR_B32_term $exec_lo, [[S_AND_SAVEEXEC_B32_]], implicit-def $scc
+  ; CHECK:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
+  ; CHECK: bb.3:
+  ; CHECK:   successors: %bb.4(0x80000000)
+  ; CHECK:   $exec_lo = S_MOV_B32_term [[S_MOV_B32_term]]
+  ; CHECK: bb.4:
+  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 0)
+  ret float %val
+}
+
+define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_glc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 1, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 1)
+  ret float %val
+}
+
+define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 1, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 2)
+  ret float %val
+}
+
+define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc_glc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 1, 1, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 3)
+  ret float %val
+}
+
+define amdgpu_ps float @raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
+  ; CHECK-LABEL: name: raw_tbuffer_load_f32__sgpr_rsrc__vgpr_voffset__sgpr_soffset_dlc
+  ; CHECK: bb.1 (%ir-block.0):
+  ; CHECK:   liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0
+  ; CHECK:   $vcc_hi = IMPLICIT_DEF
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2
+  ; CHECK:   [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3
+  ; CHECK:   [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4
+  ; CHECK:   [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5
+  ; CHECK:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+  ; CHECK:   [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6
+  ; CHECK:   [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3
+  ; CHECK:   [[TBUFFER_LOAD_FORMAT_X_OFFEN:%[0-9]+]]:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN [[COPY4]], [[REG_SEQUENCE]], [[COPY5]], 0, 78, 0, 0, 0, 1, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4)
+  ; CHECK:   $vgpr0 = COPY [[TBUFFER_LOAD_FORMAT_X_OFFEN]]
+  ; CHECK:   SI_RETURN_TO_EPILOG implicit $vgpr0
+  %val = call float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 78, i32 4)
+  ret float %val
+}
+
+declare float @llvm.amdgcn.raw.tbuffer.load.f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <2 x float> @llvm.amdgcn.raw.tbuffer.load.v2f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <3 x float> @llvm.amdgcn.raw.tbuffer.load.v3f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+declare <4 x float> @llvm.amdgcn.raw.tbuffer.load.v4f32(<4 x i32>, i32, i32, i32 immarg, i32 immarg) #0
+
+attributes #0 = { nounwind readonly }